abstract
stringlengths
383
2.62k
TLDR
stringlengths
4
250
To understand and interpret neural networks, representation similarity metrics have been used to compare learned representations between and across networks. Recent experiments have compared these similarity metrics to find the best performing and the most robust metrics, noting that classic baselines perform surprisingly well. These experiments are mostly constrained to studying relatively low-dimensional representations because of the computational cost of prominent representation similarity metrics. We extend previous work to test representation similarity metrics on larger convolutional networks processing larger images. In order to make this work possible, we employ reformulated representation similarity metrics for use on very high-dimensional representations. Using these reformulated similarity metrics, we test how well each metric captures changes to representations induced by ablations in two popular convolutional networks. In order to ground the effects of changes to representations in function, we use linear decoding probes and network performance measures. These measures of function allow us to test how well similarity metrics capture changes in decodable information versus changes in network performance. Linear decoding methods index available information in the representation, while network performance measures index the information used by the network. We show that all the tested representation similarity metrics significantly predict changes in network function and decodability. Within these metrics, on average, Procrustes and CKA outperform regularized CCA-based methods. All metrics predict decodability changes significantly better than they do network function. Procrustes and CKA do not outperform regularized CCA-based metrics for all network and functionality measure combinations. We add to the growing literature on representational similarity metrics to facilitate the improvement of current metrics for network interpretability.
We evaluate representation similarity measures for sensitivity to decoding and network function using ablation on convolutional neural networks.
Predicting Vulnerable Road User (VRU) crossing intention is one of the major challenges in automated driving. Crossing intention prediction systems trained only on pedestrian features underperform in situations that are most obvious to humans, as the latter take additional context features into consideration. Moreover, such systems tend to be over-confident for out-of-distribution samples, therefore making them less reliable to be used by downstream tasks like sensor fusion and trajectory planning for automated vehicles. In this work, we demonstrate that the results of crossing intention prediction systems can be improved by incorporating traffic light status as an additional input. Further, we make the model robust and interpretable by estimating uncertainty. Experiments on the PIE dataset show that the F1-score improved from 0.77 to 0.82 and above for three different baseline systems when considering traffic-light context. By adding uncertainty, we show increased uncertainty values for out-of-distribution samples, therefore leading to interpretable and reliable predictions of crossing intention.
We improve pedestrian crossing intention model performance and robustness using traffic light status and predicting uncertainty estimation.
Offline Reinforcement Learning (RL) extracts effective policies from historical data without the need to interact with the environment. However, the learned policy often suffers large generalization errors in the online environment due to the distributional shift. While existing work mostly focuses on learning a generalizable policy, we propose to adapt the learned policy to fit the online environment with limited queries. The goals include querying reasonable actions with limited chances and efficiently modifying the policy. Our insight is to unify these two goals via a proper pseudometric. Intuitively, the metric can compare online and offline states to infer optimal query actions. Additionally, efficient policy updates require good knowledge of the similarity between query results and historical data. Therefore, we propose a unified framework, denoted Pseudometric Guided Offline-to-Online RL (PGO2). Specifically, in deep Q learning, PGO2 has a structural design between the Q-neural network and the Siamese network, which guarantees simultaneous Q-network updating and pseudometric learning, promoting Q-network fine-tuning. In the inference phase, PGO2 solves convex optimizations to identify optimal query actions. We also show that PGO2 training converges to the so-called bisimulation metric with strong theoretical guarantees. Finally, we demonstrate the superiority of PGO2 on diversified datasets.
We propose to utilize pseudometric to guide the online queries with optimality and efficient policy update.
Recognizing and telling similar objects apart is even hard for human beings. In this paper, we show that there is a phenomenon of class interference with all deep neural networks. Class interference represents the learning difficulty in data and it constitutes the largest percentage of generalization errors by deep networks. To understand class interference, we propose cross-class tests, class ego directions and interference models. We show how to use these definitions to study minima flatness and class interference of a trained model. We also show how to detect class interference during training through label dancing pattern and class dancing notes.
We show that there is a phenomenon of class interference with all deep neural networks.
Learning with confidence labels is an emerging weakly supervised learning paradigm, where training data are equipped with confidence labels instead of exact labels. Positive-confidence (Pconf) classification is a typical learning problem in this context, where we are given only positive data equipped with confidence. However, pointwise confidence may not be accessible in real-world scenarios. In this paper, we dive into a novel weakly supervised learning problem called confidence-difference (ConfDiff) classification. Instead of pointwise confidence, we are given only unlabeled data pairs equipped with confidence difference specifying the difference in the probabilities of being positive. An unbiased risk estimator is derived to tackle the problem, and we show that the estimation error bound achieves the optimal convergence rate. Extensive experiments on benchmark data sets validate the effectiveness of our proposed approaches in leveraging the supervision information of the confidence difference.
The difference of confidence labels on unlabeled data pairs, as a novel type of weak supervision, is sufficient to train binary classifiers with theoretical guarantees.
Large language models (LLM) have shown exceptional performance on a variety of natural language tasks. Yet, their capabilities for HTML understanding – i.e., parsing the raw HTML of a webpage, with applications to automation of web-based tasks, crawling, and browser-assisted retrieval – have not been fully explored. We contribute HTML understanding models (fine-tuned LLMs) and an in-depth analysis of their capabilities under three tasks: (i) Semantic Classification of HTML elements, (ii) Description Generation for HTML inputs, and (iii) Autonomous Web Navigation of HTML pages. While previous work has developed dedicated architectures and training procedures for HTML understanding, we show that LLMs pretrained on standard natural language corpora transfer remarkably well to HTML understanding tasks. For instance, fine-tuned LLMs are 12% more accurate at semantic classification compared to models trained exclusively on the task dataset. Moreover, when fine-tuned on data from the MiniWoB benchmark, LLMs successfully complete 50% more tasks using 192x less data compared to the previous best supervised model. To promote further research on LLMs for HTML understanding, we create and open-source a large-scale HTML dataset distilled and auto-labeled from CommonCrawl. We show evidence that T5-based models due to the bidirectional encoder-decoder architecture are the best choice and that for practitioners larger models are not necessarily better.
Large language models are very effective at understanding HTML including navigating web pages, classifying elements, and generating descriptions of elements.
Parameter sharing has greatly contributed to the success of multi-agent reinforcement learning in recent years. However, most existing parameter sharing mechanisms are static, and parameters are indiscriminately shared among individuals, ignoring the dynamic environments and different roles of multiple agents. In addition, although a single-level selective parameter sharing mechanism can promote the diversity of strategies, it is hard to establish complementary and cooperative relationships between agents. To address these issues, we propose a bi-level dynamic parameter sharing mechanism among individuals and teams for promoting effective collaborations (BDPS). Specifically, at the individual level, we define virtual dynamic roles based on the long-term cumulative advantages of agents and share parameters among agents in the same role. At the team level, we combine agents of different virtual roles and share parameters of agents in the same group. Through the joint efforts of these two levels, we achieve a dynamic balance between the individuality and commonality of agents, enabling agents to learn more complex and complementary collaborative relationships. We evaluate BDPS on a challenging set of StarCraft II micromanagement tasks. The experimental results show that our method outperforms the current state-of-the-art baselines, and we demonstrate the reliability of our proposed structure through ablation experiments.
We propose a bi-level dynamic parameter sharing mechanism between individuals and teams, which can not only promote agents to learn diversified strategies, but also promote agents to form more stable and complementary cooperative relationships.
We consider the problem of offline reinforcement learning where only a set of system transitions is made available for policy optimization. Following recent advances in the field, we consider a model-based reinforcement learning algorithm that infers the system dynamics from the available data and performs policy optimization on imaginary model rollouts. This approach is vulnerable to exploiting model errors which can lead to catastrophic failures on the real system. The standard solution is to rely on ensembles for uncertainty heuristics and to avoid exploiting the model where it is too uncertain. We challenge the popular belief that we must resort to ensembles by showing that better performance can be obtained with a single well-calibrated autoregressive model on the D4RL benchmark. We also analyze static metrics of model-learning and conclude on the important model properties for the final performance of the agent.
We show in model-based offline reinforcement learning a better performance can be obtained with a single well-calibrated autoregressive system model than with the usual ensembles.
While diffusion models have shown great success in image generation, their noise-inverting generative process does not explicitly consider the structure of images, such as their inherent multi-scale nature. Inspired by diffusion models and the empirical success of coarse-to-fine modelling, we propose a new diffusion-like model that generates images through stochastically reversing the heat equation, a PDE that locally erases fine-scale information when run over the 2D plane of the image. We interpret the solution of the forward heat equation with constant additive noise as a variational approximation in the diffusion latent variable model. Our new model shows emergent qualitative properties not seen in standard diffusion models, such as disentanglement of overall colour and shape in images. Spectral analysis on natural images highlights connections to diffusion models and reveals an implicit coarse-to-fine inductive bias in them.
We propose a generative model that iteratively reverses the heat equation, increasing the effective resolution of the image
Federated Learning has become a widely-used framework which allows learning a global model on decentralized local datasets under the condition of protecting local data privacy. However, federated learning faces severe optimization difficulty when training samples are not independently and identically distributed (non-i.i.d.). In this paper, we point out that the client sampling practice plays a decisive role in the aforementioned optimization difficulty. We find that the negative client sampling will cause the merged data distribution of currently sampled clients heavily inconsistent with that of all available clients, and further make the aggregated gradient unreliable. To address this issue, we propose a novel learning rate adaptation mechanism to adaptively adjust the server learning rate for the aggregated gradient in each round, according to the consistency between the merged data distribution of currently sampled clients and that of all available clients. Specifically, we make theoretical deductions to find a meaningful and robust indicator that is positively related to the optimal server learning rate and can effectively reflect the merged data distribution of sampled clients, and we utilize it for the server learning rate adaptation. Extensive experiments on multiple image and text classification tasks validate the great effectiveness of our method.
We find that existing federated optimization suffers from the unreliable aggregated gradients caused by the negative client sampling results, and propose a gradient similarity–aware learning rate adaptation mechanism to address this problem.
3D Point cloud is a critical data representation in many real-world applications, such as autonomous driving, robotics, and medical imaging. Although the success of deep learning further accelerates the adoption of 3D point clouds in the physical world, deep learning is notoriously vulnerable to adversarial attacks. Various defense solutions have been proposed to build robust models against adversarial attacks. In this work, we identify that the state-of-the-art empirical defense, adversarial training, has a major limitation in 3D point cloud models due to gradient obfuscation, resulting in significant degradation of robustness against strong attacks. To bridge the gap, we propose PointDP, a purification strategy that leverages diffusion models to defend against 3D adversarial attacks. Since PointDP does not rely on predefined adversarial examples for training, it can defend against diverse threats. We extensively evaluate PointDP on six representative 3D point cloud architectures and leverage sixteen strong and adaptive attacks to demonstrate its lower-bound robustness. Our evaluation shows that PointDP achieves significantly better (i.e., 12.6\%-40.3\%) adversarial robustness than state-of-the-art methods under strong attacks bounded by different $\ell_p$ norms.
We propose PointDP, a diffusion-driven purification strategy to defend against adversarial point cloud. PointDP consistently achieves the strongest robustness under various attacks.
We consider fair graph representation learning via data augmentations. While this direction has been explored previously, existing methods invariably rely on certain assumptions on the properties of fair graph data in order to design fixed strategies on data augmentations. Nevertheless, the exact properties of fair graph data may vary significantly in different scenarios. Hence, heuristically designed augmentations may not always generate fair graph data in different application scenarios. In this work, we propose a method, known as Graphair, to learn fair representations based on automated graph data augmentations. Such fairness-aware augmentations are themselves learned from data. Our Graphair is designed to automatically discover fairness-aware augmentations from input graphs in order to circumvent sensitive information while preserving other useful information. Experimental results demonstrate that our Graphair consistently outperforms many baselines on multiple node classification datasets in terms of fairness-accuracy trade-off performance. In addition, results indicate that Graphair can automatically learn to generate fair graph data without prior knowledge on fairness-relevant graph properties.
We propose an automated graph data augmentation method to learn fair graph representations.
Unsupervised skill learning aims to learn a rich repertoire of behaviors without external supervision, providing artificial agents with the ability to control and influence the environment. However, without appropriate knowledge and exploration, skills may provide control only over a restricted area of the environment, limiting their applicability. Furthermore, it is unclear how to leverage the learned skill behaviors for adapting to downstream tasks in a data-efficient manner. We present Choreographer, a model-based agent that exploits its world model to learn and adapt skills in imagination. Our method decouples the exploration and skill learning processes, being able to discover skills in the latent state space of the model. During adaptation, the agent uses a meta-controller to evaluate and adapt the learned skills efficiently by deploying them in parallel in imagination. Choreographer is able to learn skills both from offline data, and by collecting data simultaneously with an exploration policy. The skills can be used to effectively adapt to downstream tasks, as we show in the URL benchmark, where we outperform previous approaches from both pixels and states inputs. The skills also explore the environment thoroughly, finding sparse rewards more frequently, as shown in goal-reaching tasks from the DMC Suite and Meta-World. Project website: https://skillchoreographer.github.io/
Choreographer: a model-based agent that discovers and learns unsupervised skills in latent imagination, and it's able to efficiently coordinate and adapt the skills to solve downstream tasks.
When solving a problem, human beings have the adaptive ability in terms of the type of information they use, the procedure they take, and the amount of time they spend approaching and solving the problem. However, most standard neural networks have the same function type and fixed computation budget on different samples regardless of their nature and difficulty. Adaptivity is a powerful paradigm as it not only imbues practitioners with flexibility pertaining to the downstream usage of these models but can also serve as a powerful inductive bias for solving certain challenging classes of problems. In this work, we propose a new strategy, AdaTape, that enables dynamic computation in neural networks via adaptive tape tokens. AdaTape employs an elastic input sequence by equipping an existing architecture with a dynamic read and write tape. Specifically, we adaptively generate input sequences using tape tokens obtained from a tape bank that can either be trainable or generated from input data. We analyze the challenges and requirements to obtain dynamic sequence content and length, and propose the Adaptive Tape Reader (ATR) algorithm to achieve both objectives. Via extensive experiments on image recognition tasks, we show that AdaTape can achieve better performance while maintaining the computational cost.
We present a new perspective for embattling dynamic allocation of computation budget to different inputs via introducing elasticity to the input length.
Stochastic gradient descent is a workhorse for training deep neural networks due to its excellent generalization performance. Several studies demonstrated this success is attributed to the implicit bias of the method that prefers a flat minimum and developed new methods based on this perspective. Recently, Izmailov et al. (2018) empirically observed that an averaged stochastic gradient descent with a large step size can bring out the implicit bias more effectively and can converge more stably to a flat minimum than the vanilla stochastic gradient descent. In our work, we theoretically justify this observation by showing that the averaging scheme improves the bias-optimization tradeoff coming from the stochastic gradient noise: a large step size amplifies the bias but makes convergence unstable, and vice versa. Specifically, we show that the averaged stochastic gradient descent can get closer to a solution of a penalized objective on the sharpness than the vanilla stochastic gradient descent using the same step size under certain conditions. In experiments, we verify our theory and demonstrate this learning scheme significantly improves performance.
This paper shows that averaged SGD with a large step-size efficiently converges to flat regions.
Knowledge distillation is one of the primary methods of transferring knowledge from large to small models. However, it requires massive task-specific data, which may not be plausible in many real-world applications. Data augmentation methods such as representation interpolation, token replacement, or augmentation with models are applied to tackle this problem. However, these data augmentation methods either potentially cause shifts in decision boundaries (representation interpolation), are not expressive enough (token replacement), or introduce too much computational overhead (augmentation with models). To this end, we propose AugPro (Augmentation with Projection), an effective and efficient data augmentation method for distillation. Our method builds on top of representation interpolation augmentation methods to maintain the diversity of expressions and converts the augmented data to tokens to avoid shifting decision boundaries. It uses simple operations that come with little computational overhead. The results on multiple GLUE tasks show that our methods can improve distillation performance by a large margin at a low time cost.
We proposed an effective and efficient data augmentation paradigm for knowledge distillation
For small training set sizes $P$, the generalization error of wide neural networks is well-approximated by the error of an infinite width neural network (NN), either in the kernel or mean-field/feature-learning regime. However, after a critical sample size $P^*$, we empirically find the finite-width network generalization becomes worse than that of the infinite width network. In this work, we empirically study the transition from infinite-width behavior to this \textit{variance-limited} regime as a function of sample size $P$ and network width $N$. We find that finite-size effects can become relevant for very small dataset sizes on the order of $P^* \sim \sqrt{N}$ for polynomial regression with ReLU networks. We discuss the source of these effects using an argument based on the variance of the NN's final neural tangent kernel (NTK). This transition can be pushed to larger $P$ by enhancing feature learning or by ensemble averaging the networks. We find that the learning curve for regression with the final NTK is an accurate approximation of the NN learning curve. Using this, we provide a toy model which also exhibits $P^* \sim \sqrt{N}$ scaling and has $P$-dependent benefits from feature learning.
Empirical study of neural networks in the overparameterized regime shows how finite-width effects are brought on by initialization variance as sample size grows.
A soft tree is an actively studied variant of a decision tree that updates splitting rules using the gradient method. Although soft trees can take various architectures, their impact is not theoretically well known. In this paper, we formulate and analyze the Neural Tangent Kernel (NTK) induced by soft tree ensembles for arbitrary tree architectures. This kernel leads to the remarkable finding that only the number of leaves at each depth is relevant for the tree architecture in ensemble learning with an infinite number of trees. In other words, if the number of leaves at each depth is fixed, the training behavior in function space and the generalization performance are exactly the same across different tree architectures, even if they are not isomorphic. We also show that the NTK of asymmetric trees like decision lists does not degenerate when they get infinitely deep. This is in contrast to the perfect binary trees, whose NTK is known to degenerate and leads to worse generalization performance for deeper trees.
We formulate and analyze the Neural Tangent Kernel (NTK) induced by soft tree ensembles for arbitrary tree architectures
Finding the proper depth $d$ of a GNN that provides strong representation power has drawn significant attention, yet nonetheless largely remains an open problem for the graph learning community. Although noteworthy progress has been made, the depth or the number of layers of a corresponding GCN is realized by a series of graph convolution operations, which naturally makes $d$ a positive integer ($d \in \mathbb{N}+$). An interesting question is whether breaking the constraint of $\mathbb{N}+$ by making $d$ a real number ($d \in \mathbb{R}$) can bring new insights into graph learning mechanisms. In this work, by redefining GCN's depth $d$ as a trainable parameter continuously adjustable within $(-\infty,+\infty)$, we open a new door of controlling its expressiveness on graph signal processing to model graph homophily/heterophily (nodes with similar/dissimilar labels/attributes tend to inter-connect). A simple and powerful GCN model ReD-GCN, is proposed to retain the simplicity of GCN and meanwhile automatically search for the optimal $d$ without the prior knowledge regarding whether the input graph is homophilic or heterophilic. Negative-valued $d$ intrinsically enables high-pass frequency filtering functionality for graph heterophily. Variants extending the model flexibility/scalability are also developed. The theoretical feasibility of having a real-valued depth with explainable physical meanings is ensured via eigen-decomposition of the graph Laplacian and a properly designed transformation function from the perspective of functional calculus. Extensive experiments demonstrate the superiority of ReD-GCN on node classification tasks for a variety of graphs. Furthermore, by introducing the concept of eigengraph, a novel graph augmentation method is obtained: the optimal $d$ effectively generates a new topology through a properly weighted combination of eigengraphs, which dramatically boosts the performance even for a vanilla GCN.
Extend the depth of GCN from positive integer domain ($\mathbb{N}+$) to real number domain ($\mathbb{R}$). A novel problem of automatic GCN depth tuning for graph homophily/heterophily detection is formulated.
Graph Neural Networks (GNNs) have emerged as a powerful technique for learning on relational data. Owing to the relatively limited number of message passing steps they perform—and hence a smaller receptive field—there has been significant interest in improving their expressivity by incorporating structural aspects of the underlying graph. In this paper, we explore the use of affinity measures as features in graph neural networks, in particular measures arising from random walks, including effective resistance, hitting and commute times. We propose message passing networks based on these features and evaluate their performance on a variety of node and graph property prediction tasks.
We show how to use affinity measures arising from random walks (e.g., effective resistance) to design message passing networks that are shown to outperform various benchmarks with fewer message passing steps.
Systematic examination of learning tasks remains an important but understudied area of machine learning (ML) research. To date, most ML research has focused on measuring performance on new tasks or surpassing state of the art performance on existing tasks. These efforts are vital but do not explain why some tasks are more difficult than others. Understanding how task characteristics affect difficulty is critical to formalizing ML's strengths and limitations; a rigorous assessment of which types of tasks are well-suited to a specific algorithm and, conversely, which algorithms are well-suited to a specific task would mark an important step forward for the field. To assist researchers in this effort, we introduce a novel learning environment designed to study how task characteristics affect measured difficulty for the learner. This tool frames learning tasks as a ``board-clearing game,'' which we call the Game of Hidden Rules (GOHR). In each instance of the game, the researcher encodes a specific rule, unknown to the learner, that determines which moves are allowed at each state of the game. The learner must infer the rule through play. We detail the game's expressive rule syntax and show how it gives researchers granular control over learning tasks. We present sample rules, a sample ML algorithm, and methods to assess algorithm performance. Separately, we provide additional benchmark rules, a public leaderboard for performance on these rules, and documentation for installing and using the GOHR environment.
We present a new learning environment allowing researchers to rigorously study how the characteristics of learning tasks affect difficulty.
A distinguishing characteristic of federated learning is that the (local) client data could have statistical heterogeneity. This heterogeneity has motivated the design of personalized learning, where individual (personalized) models are trained, through collaboration. There have been various personalization methods proposed in literature, with seemingly very different forms and methods ranging from use of a single global model for local regularization and model interpolation, to use of multiple global models for personalized clustering, etc. In this work, we begin with a statistical framework that unifies several different algorithms as well as suggest new algorithms. We apply our framework to personalized estimation, and connect it to the classical empirical Bayes' methodology. We develop novel private personalized estimation under this framework. We then use our statistical framework to propose new personalized learning algorithms, including AdaPeD based on information-geometry regularization, which numerically outperforms several known algorithms. We develop privacy for personalized learning methods with guarantees for user-level privacy and composition. We numerically evaluate the performance as well as the privacy for both the estimation and learning problems, demonstrating the advantages of our proposed methods.
We utilize a statistical framework to enable our design of new personalized Federated Learning/Estimation algorithms with privacy guarantees.
Recurrent neural networks (RNNs) are well suited for solving sequence tasks in resource-constrained systems due to their expressivity and low computational requirements. However, there is still a need to bridge the gap between what RNNs are capable of in terms of efficiency and performance and real-world application requirements. The memory and computational requirements arising from propagating the activations of all the neurons at every time step to every connected neuron, together with the sequential dependence of activations, contribute to the inefficiency of training and using RNNs. We propose a solution inspired by biological neuron dynamics that makes the communication between RNN units sparse and discrete. This makes the backward pass with backpropagation through time (BPTT) computationally sparse and efficient as well. We base our model on the gated recurrent unit (GRU), extending it with units that emit discrete events for communication triggered by a threshold so that no information is communicated to other units in the absence of events. We show theoretically that the communication between units, and hence the computation required for both the forward and backward passes, scales with the number of events in the network. Our model achieves efficiency without compromising task performance, demonstrating competitive performance compared to state-of-the-art recurrent network models in real-world tasks, including language modeling. The dynamic activity sparsity mechanism also makes our model well suited for novel energy-efficient neuromorphic hardware. Code is available at https://github.com/KhaleelKhan/EvNN/.
We add a activity sparsity mechanism to the GRU using a thresholding function, which makes both the forward and backward passes computationally sparse. This model achieves competitive performance on various benchmarks including language modeling.
Many contrastive and meta-learning approaches learn representations by identifying common features in multiple views. However, the formalism for these approaches generally assumes features to be shared across views to be captured coherently. We consider the problem of learning a unified representation from partial observations, where useful features may be present in only some of the views. We approach this through a probabilistic formalism enabling views to map to representations with different levels of uncertainty in different components; these views can then be integrated with one another through marginalisation over that uncertainty. Our approach, Partial Observation Experts Modelling (POEM), then enables us to meta-learn consistent representations from partial observations. We evaluate our approach on an adaptation of a comprehensive few-shot learning benchmark, Meta-Dataset, and demonstrate the benefits of POEM over other meta-learning methods at representation learning from partial observations. We further demonstrate the utility of POEM by meta-learning to represent an environment from partial views observed by an agent exploring the environment.
An approach for meta-learning contrastive representations under partial observability.
Model compression is vital to the deployment of deep learning on edge devices. Low precision representations, achieved via quantization of weights and activations, can reduce inference time and memory requirements. However, quantifying and predicting the response of a model to the changes associated with this procedure remains challenging. This response is non-linear and heterogeneous throughout the network. Understanding which groups of parameters and activations are more sensitive to quantization than others is a critical stage in maximizing efficiency. For this purpose, we propose FIT. Motivated by an information geometric perspective, FIT combines the Fisher information with a model of quantization. We find that FIT can estimate the final performance of a network without retraining. FIT effectively fuses contributions from both parameter and activation quantization into a single metric. Additionally, FIT is fast to compute when compared to existing methods, demonstrating favourable convergence properties. These properties are validated experimentally across hundreds of quantization configurations, with a focus on layer-wise mixed-precision quantization.
We propose the Fisher Information Trace (FIT) metric, to quantify the effects of mixed-precision quantization. FIT facilitates zero-shot performance prediction of quantized models, and is fast to compute.
Despite the huge success of object detection, the training process still requires an immense amount of labeled data. Although various active learning solutions for object detection have been proposed, most existing works do not take advantage of epistemic uncertainty, which is an important metric for capturing the usefulness of the sample. Also, previous works pay little attention to the attributes of each bounding box (e.g., nearest object, box size) when computing the informativeness of an image. In this paper, we propose a new active learning strategy for object detection that overcomes the shortcomings of prior works. To make use of epistemic uncertainty, we adopt evidential deep learning (EDL) and propose a new module termed model evidence head (MEH), that makes EDL highly compatible with object detection. Based on the computed epistemic uncertainty of each bounding box, we propose hierarchical uncertainty aggregation (HUA) for obtaining the informativeness of an image. HUA realigns all bounding boxes into multiple levels based on the attributes and aggregates uncertainties in a bottom-up order, to effectively capture the context within the image. Experimental results show that our method outperforms existing state-of-the-art methods by a considerable margin.
We propose an active learning method for object detection using evidential deep learning and novel uncertainty aggregation method.
Lighter and faster image restoration (IR) models are crucial for the deployment on resource-limited devices. Binary neural network (BNN), one of the most promising model compression methods, can dramatically reduce the computations and parameters of full-precision convolutional neural networks (CNN). However, there are different properties between BNN and full-precision CNN, and we can hardly use the experience of designing CNN to develop BNN. In this study, we reconsider components in binary convolution, such as residual connection, BatchNorm, activation function, and structure, for IR tasks. We conduct systematic analyses to explain each component's role in binary convolution and discuss the pitfalls. Specifically, we find that residual connection can reduce the information loss caused by binarization; BatchNorm can solve the value range gap between residual connection and binary convolution; The position of the activation function dramatically affects the performance of BNN. Based on our findings and analyses, we design a simple yet efficient basic binary convolution unit (BBCU). Furthermore, we divide IR networks into four parts and specially design variants of BBCU for each part to explore the benefit of binarizing these parts. We conduct experiments on different IR tasks, and our BBCU significantly outperforms other BNNs and lightweight models, which shows that BBCU can serve as a basic unit for binarized IR networks. All codes and models will be released.
We reconsider the components in BNNs and design a strong, simple and efficient baisc binary convolution unit.
In this paper we present a novel method to estimate 3D human pose and shape from monocular videos. This task requires directly recovering pixel-alignment 3D human pose and body shape from monocular images or videos, which is challenging due to its inherent ambiguity. To improve precision, existing methods highly rely on the initialized mean pose and shape as prior estimates and parameter regression with an iterative error feedback manner. In addition, video-based approaches model the overall change over the image-level features to temporally enhance the single-frame feature, but fail to capture the rotational motion at the joint level, and cannot guarantee local temporal consistency. To address these issues, we propose a novel Transformer-based model with a design of independent tokens. First, we introduce three types of tokens independent of the image feature: \textit{joint rotation tokens, shape token, and camera token}. By progressively interacting with image features through Transformer layers, these tokens learn to encode the prior knowledge of human 3D joint rotations, body shape, and position information from large-scale data, and are updated to estimate SMPL parameters conditioned on a given image. Second, benefiting from the proposed token-based representation, we further use a temporal model to focus on capturing the rotational temporal information of each joint, which is empirically conducive to preventing large jitters in local parts. Despite being conceptually simple, the proposed method attains superior performances on the 3DPW and Human3.6M datasets. Using ResNet-50 and Transformer architectures, it obtains 42.0 mm error on the PA-MPJPE metric of the challenging 3DPW, outperforming state-of-the-art counterparts by a large margin. Code will be publicly available\footnote{\url{https://github.com/yangsenius/INT_HMR_Model}}.
We present a novel, effective and robust model with designed independent tokens to estimate 3D human pose and shape from monocular videos
Quantization of the weights and activations is one of the main methods to reduce the computational footprint of Deep Neural Networks (DNNs) training. Current methods enable 4-bit quantization of the forward phase. However, this constitutes only a third of the training process. Reducing the computational footprint of the entire training process requires the quantization of the neural gradients, i.e., the loss gradients with respect to the outputs of intermediate neural layers. Previous works separately showed that accurate 4-bit quantization of the neural gradients needs to (1) be unbiased and (2) have a log scale. However, no previous work aimed to combine both ideas, as we do in this work. Specifically, we examine the importance of having unbiased quantization in quantized neural network training, where to maintain it, and how to combine it with logarithmic quantization. Based on this, we suggest a $\textit{logarithmic unbiased quantization}$ (LUQ) method to quantize all both the forward and backward phase to 4-bit, achieving state-of-the-art results in 4-bit training without overhead. For example, in ResNet50 on ImageNet, we achieved a degradation of 1.1 %. We further improve this to degradation of only 0.32 % after three epochs of high precision fine-tunining, combined with a variance reduction method---where both these methods add overhead comparable to previously suggested methods. A reference implementation is supplied in the supplementary material.
A method to quantize all training matrix multiplication in 4 bit with standard formats
Data assimilation refers to a group of algorithms that combines numerical models with observations to obtain an optimal estimation of the system's states. In areas like earth science, numerical models are usually formulated by differential equations, also known as the prior dynamics. It is a great challenge for neural networks to properly exploit the dynamical characteristics for data assimilation, because first, it is difficult to represent complicated dynamical characteristics in neural networks, and second, the dynamics are likely to be biased. The state-of-the-art neural networks borrow from the traditional method to introduce dynamical characteristics by optimizing the 4D-Var objective function in which the dynamics are inherently quantified, but the iterative optimization process leads to high computational cost. In this paper, we develop a novel deep learning framework with neural operators for data assimilation. The key novelty of our proposed approach is that we design a so-called flow operator through self-supervised learning to explicitly learn dynamical characteristics for reconstructed states. Numerical experiments on the Lorenz-63 and Lorenz-96 systems, which are the standard benchmarks for data assimilation performance evaluation, show that the proposed method is at least three times faster than state-of-the-art neural networks, and reduces the dynamic loss by two orders of magnitude. It is also demonstrated that our method is well-adapted to biases in the prior dynamics.
A new deep learning framework is proposed for data assimilation issues.
In recent years, deep network pruning has attracted significant attention in order to enable the rapid deployment of AI into small devices with computation and memory constraints. Pruning is often achieved by dropping redundant weights, neurons, or layers of a deep network while attempting to retain a comparable test performance. Many deep pruning algorithms have been proposed with impressive empirical success. However, existing approaches lack a quantifiable measure to estimate the compressibility of a sub-network during each pruning iteration and thus may under-prune or over-prune the model. In this work, we propose PQ Index (PQI) to measure the potential compressibility of deep neural networks and use this to develop a Sparsity-informed Adaptive Pruning (SAP) algorithm. Our extensive experiments corroborate the hypothesis that for a generic pruning procedure, PQI decreases first when a large model is being effectively regularized and then increases when its compressibility reaches a limit that appears to correspond to the beginning of underfitting. Subsequently, PQI decreases again when the model collapse and significant deterioration in the performance of the model start to occur. Additionally, our experiments demonstrate that the proposed adaptive pruning algorithm with proper choice of hyper-parameters is superior to the iterative pruning algorithms such as the lottery ticket-based pruning methods, in terms of both compression efficiency and robustness.
This work develops PQ Index (PQI) as a new measure of sparsity and proposes a Sparsity-informed Adaptive Pruning (SAP) algorithm.
A standard hardware bottleneck when training deep neural networks is GPU memory. The bulk of memory is occupied by caching intermediate tensors for gradient computation in the backward pass. We propose a novel method to reduce this footprint - Dropping Intermediate Tensors (DropIT). DropIT drops min-k elements of the intermediate tensors and approximates gradients from the sparsified tensors in the backward pass. Theoretically, DropIT reduces noise on estimated gradients and therefore has a higher rate of convergence than vanilla-SGD. Experiments show that we can drop up to 90\% of the intermediate tensor elements in fully-connected and convolutional layers while achieving higher testing accuracy for Visual Transformers and Convolutional Neural Networks on various tasks (e.g., classification, object detection, instance segmentation). Our code and models are available at https://github.com/chenjoya/dropit.
DropIT can save memory & improve accuracy, providing a new perspective of dropping in activation compressed training than quantization.
Resource-constrained perception systems such as edge computing and vision-for-robotics require vision models to be both accurate and lightweight in computation and memory usage. Knowledge distillation is one effective strategy to improve the performance of lightweight classification models, but it is less well-explored for structured outputs such as object detection and instance segmentation, where the variable number of outputs and complex internal network modules complicate the distillation. In this paper, we propose a simple yet surprisingly effective sequential approach to knowledge distillation that progressively transfers the knowledge of a set of teachers to a given lightweight student. Our approach is inspired by curriculum learning: To distill knowledge from a highly accurate but complex teacher model, we construct a sequence of teachers to help the student gradually adapt. Our progressive distillation strategy can be easily combined with existing distillation mechanisms to consistently maximize student performance in various settings. To the best of our knowledge, we are the first to successfully distill knowledge from Transformer-based teacher detectors to convolution-based students, and unprecedentedly boost the performance of ResNet-50 based RetinaNet from 36.5% to 42.0% AP and Mask R-CNN from 38.2% to 42.5% AP on the MS COCO benchmark.
We propose a progressive approach to distill knowledge from multiple teacher detectors into a lightweight student.
Cloud based machine learning inference is an emerging paradigm where users share their data with a service provider. Due to increased concerns over data privacy, recent works have proposed using Adversarial Representation Learning (ARL) to learn a privacy-preserving encoding of sensitive user data before it is shared with an untrusted service provider. Traditionally, the privacy of these encodings is evaluated empirically as they lack formal guarantees. In this work, we develop a new framework that provides formal privacy guarantees for an arbitrarily trained neural network by linking its local Lipschitz constant with its local sensitivity. To utilize local sensitivity for guaranteeing privacy, we extend the Propose-Test-Release(PTR) framework to make it tractable for neural network based queries. We verify the efficacy of our framework experimentally on real-world datasets and elucidate the role of ARL in improving the privacy-utility tradeoff.
We present a framework for achieving formal privacy guarantees in adversarially trained ML models
Algorithmic fairness plays an increasingly critical role in machine learning research. Several group fairness notions and algorithms have been proposed. However, the fairness guarantee of existing fair classification methods mainly depend on specific data distributional assumptions, often requiring large sample sizes, and fairness could be violated when there is a modest number of samples, which is often the case in practice. In this paper, we propose FaiREE, a fair classification algorithm which can satisfy group fairness constraints with finite-sample and distribution-free theoretical guarantees. FaiREE can be adapted to satisfying various group fairness notions (e.g., Equality of Opportunity, Equalized Odds, Demographic Parity, etc.) and achieve the optimal accuracy. These theoretical guarantees are further supported by experiments on both synthetic and real data. FaiREE is shown to have favorable performance over state-of-the-art algorithms.
We propose a fair classification algorithm which can satisfy the group fairness constraints with finite-sample and distribution-free guarantees.
Operator learning, learning the mapping between infinite-dimensional function spaces, has been attracted as an alternative approach to traditional numerical methods to solve partial differential equations (PDEs). In practice, the functions of the physical systems are often observed by sparse or even irregularly distributed measurements, thus the functions are discretized and usually represented by finite structured arrays, which are given as data of input-output pairs. Through training with the arrays, the solution of the trained models should be independent of the discretization of the input function and can be queried at any point continuously. Therefore, the architectures for operator learning should be flexibly compatible with arbitrary sizes and locations of the measurements, otherwise, it can restrict the scalability when the observations have discrepancies between measurement formats. In this paper, we propose to treat the discretized functions as set-valued data and construct an attention-based model, called mesh-independent operator learner (MIOL), to provide proper treatments of input functions and query coordinates for the solution functions by detaching the dependencies on input and output meshes. Our models pre-trained with benchmark datasets of operator learning are evaluated by downstream tasks to demonstrate the generalization abilities to varying discretization formats of the system, which are natural characteristics of the continuous solution of the PDEs.
We propose an attention-based operator learning model for obtaining the continuous solution of PDEs, independent of the discretization formats.
Current fake audio detection algorithms achieve promising performances on most datasets. However, their performance may be significantly degraded when dealing with audio of a different dataset. The orthogonal weight modification to overcome catastrophic forgetting does not consider the similarity of some audio, including fake audio obtained by the same algorithm and genuine audio, on different datasets. To overcome this limitation, we propose a continual learning algorithm for fake audio detection to overcome catastrophic forgetting, called Regularized Adaptive Weight Modification (RAWM). Specifically, when fine-tuning a detection network, our approach adaptively computes the direction of weight modification according to the ratio of genuine utterances and fake utterances. The adaptive modification direction ensures the network can detect fake audio on the new dataset while preserving its knowledge of previous model, thus mitigating catastrophic forgetting. In addition, orthogonal weight modification of fake audios in the new dataset will skew the distribution of inferences on audio in the previous dataset with similar acoustic characteristics, so we introduce a regularization constraint to force the network to remember this distribution. We evaluate our approach across multiple datasets and obtain a significant performance improvement on cross-dataset experiments.
We propose a regularized adaptive weight modification algorithm to overcome catastrophic forgetting for fake audio detection.
Despite recent successes in natural language processing and computer vision, Transformer suffers from the scalability problem when dealing with graphs. Computing node-to-node attentions is infeasible on complicated graphs, e.g., knowledge graphs. One solution is to consider only the near neighbors, which, however, will lose the key merit of Transformer that attends to the elements at any distance. In this paper, we propose a new Transformer architecture, named dual-encoding Transformer (DET), which has a structural encoder to aggregate information from near neighbors and a semantic encoder to focus on useful semantically close neighbors. The two encoders can be incorporated to boost each other's performance. Our experiments demonstrate that DET achieves superior performance compared to the respective state-of-the-art attention-based methods in dealing with molecules, networks and knowledge graphs.
A Transformer model for diverse graph representation learning tasks
Training large neural networks is known to be time-consuming, where learning duration may stretch up to days or weeks. To address this problem, the approach of large-batch optimization was introduced, demonstrating that scaling mini-batch sizes with appropriate learning rate adjustments may speed up the training process by orders of magnitude. While long training time was not typically a major issue for model-free deep offline RL algorithms, recently introduced Q-ensemble methods achieving state-of-the-art performance made this issue more relevant, notably extending the training duration. In this work, we demonstrate how large-batch optimization, typically overlooked in deep offline RL community, can benefit this class of methods. We show that simply scaling the mini-batch size and naively adjusting the learning rate allows for (1) a reduced size of the Q-ensemble, (2) stronger penalization of out-of-distribution actions, and (3) improved convergence time, effectively shortening training durations by 2.5x times on average.
Large Batch Optimization for SAC-N allows to reduce size of the Q-ensemble and improves convergence time by 2.5x times on average
Despite the tremendous success, existing machine learning models still fall short of human-like systematic generalization—learning compositional rules from limited data and applying them to unseen combinations in various domains. We propose Neural-Symbolic Recursive Machine (NSR) to tackle this deficiency. The core representation of NSR is a Grounded Symbol System (GSS) with combina- torial syntax and semantics, which entirely emerges from training data. Akin to the neuroscience studies suggesting separate brain systems for perceptual, syntactic, and semantic processing, NSR implements analogous separate modules of neural perception, syntactic parsing, and semantic reasoning, which are jointly learned by a deduction-abduction algorithm. We prove that NSR is expressive enough to model various sequence-to-sequence tasks. Superior systematic generalization is achieved via the inductive biases of equivariance and recursiveness embedded in NSR. In experiments, NSR achieves state-of-the-art performance in three benchmarks from different domains: SCAN for semantic parsing, PCFG for string manipulation, and HINT for arithmetic reasoning. Specifically, NSR achieves 100% generalization accuracy on SCAN and PCFG and outperforms state-of-the-art models on HINT by about 23%. Our NSR demonstrates stronger generalization than pure neural networks due to its symbolic representation and inductive biases. NSR also demonstrates better transferability than existing neural-symbolic approaches due to less domain-specific knowledge required.
We present Neural-Symbolic Recursive Machine for systematic generalization which achieves state-of-the-art performance on SCAN, PCFG, and HINT.
Not forgetting old class knowledge is a key challenge for class-incremental learning (CIL) when the model continuously adapts to new coming classes. A common technique to address this is knowledge distillation (KD) which penalizes prediction inconsistencies between old and new models. Such prediction is made with almost new class data, as old class data is extremely scarce due to the strict memory limitation in CIL. In this paper, we take a deep dive into KD losses and find that “using new class data for KD” not only hinders the model adaption (for learning new classes) but also results in low efficiency for preserving old class knowledge. We address this by “using the placebos of old classes for KD”, where the placebos are chosen from a free image stream, such as Google Images, in an automatical and economical fashion. To this end, we train an online placebo selection policy to quickly evaluate the quality of streaming images (good or bad placebos) and use only good ones for one-time feed-forward computation of KD. We formulate the policy training process as an online Markov Decision Process (MDP), and introduce an online learning algorithm to solve this MDP problem without causing much computation costs. In experiments, we show that our method 1) is surprisingly effective even when there is no class overlap between placebos and original old class data, 2) does not require any additional supervision or memory budget, and 3) significantly outperforms a number of top-performing CIL methods, in particular when using lower memory budgets for old class exemplars, e.g., five exemplars per class. The code is available in the supplementary.
We design an online learning algorithm to quickly evaluate and select unlabeled data to improve the KD loss in class-incremental learning.
Advances in the expressivity of pretrained models have increased interest in the design of adaptation protocols which enable safe and effective transfer learning. Going beyond conventional linear probing (LP) and fine tuning (FT) strategies, protocols that can effectively control feature distortion, i.e., the failure to update features orthogonal to the in-distribution, have been found to achieve improved out-of-distribution generalization (OOD). In order to limit this distortion, the LP+FT protocol, which first learns a linear probe and then uses this initialization for subsequent FT, was proposed. However, in this paper, we find when adaptation protocols (LP, FT, LP+FT) are also evaluated on a variety of safety objectives (e.g., calibration, robustness, etc.), a complementary perspective to feature distortion is helpful to explain protocol behavior. To this end, we study the susceptibility of protocols to simplicity bias (SB), i.e. the well-known propensity of deep neural networks to rely upon simple features, as SB has recently been shown to underlie several problems in robust generalization. Using a synthetic dataset, we demonstrate the susceptibility of existing protocols to SB. Given the strong effectiveness of LP+FT, we then propose modified linear probes that help mitigate SB, and lead to better initializations for subsequent FT. We verify the effectiveness of the proposed LP+FT variants for decreasing SB in a controlled setting, and their ability to improve OOD generalization and safety on three adaptation datasets.
Mitigating feature distortion is not enough to ensure that transfer learning from large-scale, pretrained models leads to better safety and generalization on downstream tasks.
Offline reinforcement learning (RL) struggles in environments with rich and noisy inputs, where the agent only has access to a fixed dataset without environment interactions. Past works have proposed common workarounds based on the pre-training of state representations, followed by policy training. In this work, we introduce a simple, yet effective approach for learning state representations. Our method, Behavior Prior Representation (BPR), learns state representations with an easy-to-integrate objective based on behavior cloning of the dataset: we first learn a state representation by mimicking actions from the dataset, and then train a policy on top of the fixed representation, using any off-the-shelf Offline RL algorithm. Theoretically, we prove that BPR carries out performance guarantees when integrated into algorithms that have either policy improvement guarantees (conservative algorithms) or produce lower bounds of the policy values (pessimistic algorithms). Empirically, we show that BPR combined with existing state-of-the-art Offline RL algorithms leads to significant improvements across several offline control benchmarks. The code is available at \url{https://github.com/bit1029public/offline_bpr}
We propose a state representation learning method with a surprisingly simple, easy-to-integrate objective based on behavior cloning of the dataset
Traditional analyses of gradient descent show that when the largest eigenvalue of the Hessian, also known as the sharpness $S(\theta)$, is bounded by $2/\eta$, training is "stable" and the training loss decreases monotonically. Recent works, however, have observed that this assumption does not hold when training modern neural networks with full batch or large batch gradient descent. Most recently, Cohen at al. (2021) detailed two important phenomena. The first, dubbed \emph{progressive sharpening}, is that the sharpness steadily increases throughout training until it reaches the instability cutoff $2/\eta$. The second, dubbed \emph{edge of stability}, is that the sharpness hovers at $2/\eta$ for the remainder of training while the loss continues decreasing, albeit non-monotonically. We demonstrate that, far from being chaotic, the dynamics of gradient descent at the edge of stability can be captured by a cubic Taylor expansion: as the iterates diverge in direction of the top eigenvector of the Hessian due to instability, the cubic term in the local Taylor expansion of the loss function causes the curvature to decrease until stability is restored. This property, which we call \emph{self-stabilization}, is a general property of gradient descent and explains its behavior at the edge of stability. A key consequence of self-stabilization is that gradient descent at the edge of stability implicitly follows \emph{projected} gradient descent (PGD) under the constraint $S(\theta) \le 2/\eta$. Our analysis provides precise predictions for the loss, sharpness, and deviation from the PGD trajectory throughout training, which we verify both empirically in a number of standard settings and theoretically under mild conditions. Our analysis uncovers the mechanism for gradient descent's implicit bias towards stability.
We explain the mechanism behind the edge of stability phenomenon, where full batch gradient descent non-monotonically decreases the loss in the presence of instability.
The successes of artificial neural networks (ANNs) are largely attributed to mimicking the human brain structures. Recent advances in neuroscience revealed that neurons interact with each other through various kinds of connectivity patterns to process information, in which the common connectivity patterns are also called circuit motifs. However, many existing ANNs can only model one or two circuit motifs in their architectures, so that their performance may drastically vary among different types of machine learning tasks. In this paper, we propose a new type of neural network inspired by the architectures of neuronal circuits, namely Circuit Neural Network (CircuitNet). In CircuitNet, a group of densely connected neurons, namely circuit motif unit (CMU), form the basic unit of the network, which is capable of modeling universal circuit motifs by adjusting the weights within the CMUs. Compared with traditional feed-forward networks, CircuitNet has the ability to model more types of neuron connections such as feed-back and lateral motifs. Inspired by the locally dense and globally sparse structure of the human brain, several iterations of signal transmission among different CMUs are achieved by sparse connections through the input ports and output ports of different CMUs. Experiments have demonstrated that CircuitNet can outperform popular neural network architectures in function approximation, reinforcement learning, image classification, and time series forecasting tasks.
We proposed CircuitNet by modeling the universal circuit motifs and structures in human brains to function as a generic neural network and tested in several machine learning tasks.
Communication within or between complex systems is commonplace in the natural sciences and fields such as graph neural networks. The brain is a perfect example of such a complex system, where communication between brain regions is constantly being orchestrated. To analyze communication, the brain is often split up into anatomical regions that each perform certain computations. These regions must interact and communicate with each other to perform tasks and support higher-level cognition. On a macroscale, these regions communicate through signal propagation along the cortex and along white matter tracts over longer distances. When and what types of signals are communicated over time is an unsolved problem and is often studied using either functional or structural data. In this paper, we propose a non-linear generative approach to communication from functional data. We address three issues with common connectivity approaches by explicitly modeling the directionality of communication, finding communication at each timestep, and encouraging sparsity. To evaluate our model, we simulate temporal data that has sparse communication between nodes embedded in it and show that our model can uncover the expected communication dynamics. Subsequently, we apply our model to temporal neural data from multiple tasks and show that our approach models communication that is more specific to each task. The specificity of our method means it can have an impact on the understanding of psychiatric disorders, which are believed to be related to highly specific communication between brain regions compared to controls. In sum, we propose a general model for dynamic communication learning on graphs, and show its applicability to a subfield of the natural sciences, with potential widespread scientific impact.
We address three issues with common connectivity approaches by explicitly modeling the directionality of communication, finding communication at each timestep, and encouraging sparsity.
Classical wisdom suggests that estimators should avoid fitting noise to achieve good generalization. In contrast, modern overparameterized models can yield small test error despite interpolating noise — a phenomenon often called "benign overfitting" or "harmless interpolation". This paper argues that the degree to which interpolation is harmless hinges upon the strength of an estimator's inductive bias, i.e., how heavily the estimator favors solutions with a certain structure: while strong inductive biases prevent harmless interpolation, weak inductive biases can even require fitting noise to generalize well. Our main theoretical result establishes tight non-asymptotic bounds for high-dimensional kernel regression that reflect this phenomenon for convolutional kernels, where the filter size regulates the strength of the inductive bias. We further provide empirical evidence of the same behavior for deep neural networks with varying filter sizes and rotational invariance.
We show that the strength of a model’s inductive bias determines whether interpolation of noisy data is harmless or harmful.
We study the problem of crystal material property prediction. A crystal structure consists of a minimal unit cell that is repeated infinitely in 3D space. How to accurately represent such repetitive structures in machine learning models remains unresolved. Current methods construct graphs by establishing edges only between nearby nodes, thereby failing to faithfully capture infinite repeating patterns and distant interatomic interactions. In this work, we propose several innovations to overcome these limitations. First, we propose to model physics-principled interatomic potentials directly instead of only using distances as in existing methods. These potentials include the Coulomb potential, London dispersion potential, and Pauli repulsion potential. Second, we propose to model the complete set of potentials among all atoms, instead of only between nearby atoms as in prior methods. This is enabled by our approximations of infinite potential summations with provable error bounds. We further develop efficient algorithms to compute the approximations. Finally, we propose to incorporate our computations of complete interatomic potentials into message passing neural networks for representation learning. We perform experiments on the JARVIS and Materials Project benchmarks for evaluation. Results show that the use of complete interatomic potentials leads to consistent performance improvements with reasonable computational costs.
We propose to directly model complete interactions for crystals with potential summations
The transferability of adversarial perturbations between image models has been extensively studied. In this case, an attack is generated from a known surrogate \eg, the ImageNet trained model, and transferred to change the decision of an unknown (black-box) model trained on an image dataset. However, attacks generated from image models do not capture the dynamic nature of a moving object or a changing scene due to a lack of temporal cues within image models. This leads to reduced transferability of adversarial attacks from representation-enriched \emph{image} models such as Supervised Vision Transformers (ViTs), Self-supervised ViTs (\eg, DINO), and Vision-language models (\eg, CLIP) to black-box \emph{video} models. In this work, we induce dynamic cues within the image models without sacrificing their original performance on images. To this end, we optimize \emph{temporal prompts} through frozen image models to capture motion dynamics. Our temporal prompts are the result of a learnable transformation that allows optimizing for temporal gradients during an adversarial attack to fool the motion dynamics. Specifically, we introduce spatial (image) and temporal (video) cues within the same source model through task-specific prompts. Attacking such prompts maximizes the adversarial transferability from image-to-video and image-to-image models using the attacks designed for image models. As an example, an iterative attack launched from image model Deit-B with temporal prompts reduces generalization (top1 \% accuracy) of a video model by 35\% on Kinetics-400. Our approach also improves adversarial transferability to image models by 9\% on ImageNet w.r.t the current state-of-the-art approach. Our attack results indicate that the attacker does not need specialized architectures, \eg, divided space-time attention, 3D convolutions, or multi-view convolution networks for different data modalities. Image models are effective surrogates to optimize an adversarial attack to fool black-box models in a changing environment over time. Code is available at \url{https://bit.ly/3Xd9gRQ}
A new approach for optimizing temporal prompts through frozen image models to capture motion dynamics for better transferability
A backdoor attack aims to inject a backdoor into a deep model so that the model performs normally on benign samples while maliciously predicting the input as the attacker-defined target class when the backdoor is activated by a predefined trigger pattern. Most existing backdoor attacks use a pattern that rarely occurs in benign data as the trigger pattern. In this way, the impact of the attack on the label prediction of benign data can be mitigated. However, this practice also results in the attack being defended against with little performance degradation on benign data by preventing the trigger pattern from being activated. In this work, we present a new attack strategy to solve this dilemma. Unlike the conventional strategy, our strategy extracts the trigger pattern from benign training data, which frequently occurs in samples of the target class but rarely occurs in samples of the other classes. Compared with the prevailing strategy, our proposed strategy has two advantages. First, it can improve the efficiency of the attack because learning on benign samples of the target class can facilitate the fitting of the trigger pattern. Second, it increases the difficulty or cost of identifying the trigger pattern and preventing its activation, since many benign samples of the target class contain the trigger pattern. We empirically evaluate our strategy on four benchmark datasets. The experimental studies show that attacks performed with our strategy can achieve much better performance when poisoning only 0.1\% or more of the training data, and can achieve better performance against several benchmark defense algorithms.
A new strategy for developing the trigger pattern of backdoor attacks with great efficiency and stealthiness using benign training data.
Real-world machine learning problems often exhibit shifts between the source and target distributions, in which source data does not fully convey the desired behavior on target inputs. Different functions that achieve near-perfect source accuracy can make differing predictions on test inputs, and such ambiguity makes robustness to distribution shifts challenging. We propose DivDis, a simple two-stage framework for identifying and resolving ambiguity in data. DivDis first learns a diverse set of hypotheses that achieve low source loss but make differing predictions on target inputs. We then disambiguate by selecting one of the discovered functions using additional information, for example, a small number of target labels. Our experimental evaluation shows improved performance in subpopulation shift and domain generalization settings, demonstrating that DivDis can scalably adapt to distribution shifts in image and text classification benchmarks.
Given underspecified data, (1) find a diverse set of solutions and (2) choose the best one.
Weather and climate simulations produce petabytes of high-resolution data that are later analyzed by researchers in order to understand climate change or severe weather. We propose a new method of compressing this multidimensional weather and climate data: a coordinate-based neural network is trained to overfit the data, and the resulting parameters are taken as a compact representation of the original grid-based data. While compression ratios range from 300x to more than 3,000x, our method outperforms the state-of-the-art compressor SZ3 in terms of weighted RMSE, MAE. It can faithfully preserve important large scale atmosphere structures and does not introduce significant artifacts. When using the resulting neural network as a 790x compressed dataloader to train the WeatherBench forecasting model, its RMSE increases by less than 2%. The three orders of magnitude compression democratizes access to high-resolution climate data and enables numerous new research directions.
We compress weather and climate data into neural network weights.
Decision Trees are commonly used for many machine learning tasks due to their high interpretability. However, learning a decision tree from data is a difficult optimization problem, since it is non-convex and non-differentiable. Therefore, common approaches learn decision trees using a greedy growth algorithm that minimizes the impurity at each internal node. Unfortunately, this greedy procedure can lead to suboptimal trees. In this paper, we present a novel approach for learning univariate, axis-aligned decision trees with gradient descent. This is achieved by applying backpropagation with an adjusted gradient flow on a dense decision tree representation that optimizes all decision tree parameters jointly. We show that our gradient-based optimization outperforms existing baselines on several binary classification benchmarks and achieves competitive results for multi-class tasks. To the best of our knowledge, this is the first approach that attempts to learn univariate, axis-aligned decision trees with gradient descent.
A novel approach to learn univariate, axis-aligned decision trees with gradient descent using a dense tree representation and an adjusted backpropagation algorithm.
A common explanation for the failure of deep networks to generalize out-of-distribution is that they fail to recover the "correct" features. We challenge this notion with a simple experiment which suggests that ERM already learns sufficient features and that the current bottleneck is not feature learning, but robust regression. We therefore argue that devising simpler methods for learning predictors on existing features is a promising direction for future research. Towards this end, we introduce Domain-Adjusted Regression (DARE), a convex objective for learning a linear predictor that is provably robust under a new model of distribution shift. Rather than learning one function, DARE performs a domain-specific adjustment to unify the domains in a canonical latent space and learns to predict in this space. Under a natural model, we prove that the DARE solution is the minimax-optimal predictor for a constrained set of test distributions. Further, we provide the first finite-environment convergence guarantee to the minimax risk, improving over existing analyses which only yield minimax predictors after an environment threshold. Evaluated on finetuned features, we find that DARE compares favorably to prior methods, consistently achieving equal or better performance.
We show that features learned via ERM may be "good enough" for generalization, and that the main difficulty is robust classification. We give a new model of dist shift and an alg which is minimax-optimal and meets/exceeds SOTA on several benchmarks.
It is essential yet challenging for future home-assistant robots to understand and manipulate diverse 3D objects in daily human environments. Towards building scalable systems that can perform diverse manipulation tasks over various 3D shapes, recent works have advocated and demonstrated promising results learning visual actionable affordance, which labels every point over the input 3D geometry with an action likelihood of accomplishing the downstream task (e.g., pushing or picking-up). However, these works only studied single-gripper manipulation tasks, yet many real-world tasks require two hands to achieve collaboratively. In this work, we propose a novel learning framework, DualAfford, to learn collaborative affordance for dual-gripper manipulation tasks. The core design of the approach is to reduce the quadratic problem for two grippers into two disentangled yet interconnected subtasks for efficient learning. Using the large-scale PartNet-Mobility and ShapeNet datasets, we set up four benchmark tasks for dual-gripper manipulation. Experiments prove the effectiveness and superiority of our method over three baselines. We will release code and data upon acceptance.
We propose a novel learning framework to learn collaborative affordance for dual-gripper manipulation tasks.
Many modern machine learning algorithms are composed of simple private algorithms; thus, an increasingly important problem is to efficiently compute the overall privacy loss under composition. In this study, we introduce the Edgeworth Accountant, an analytical approach to composing differential privacy guarantees of private algorithms. The Edgeworth Accountant starts by losslessly tracking the privacy loss under composition using the $f$-differential privacy framework, which allows us to express the privacy guarantees using privacy-loss log-likelihood ratios (PLLRs). As the name suggests, this accountant next uses the Edgeworth expansion to the upper and lower bounds the probability distribution of the sum of the PLLRs. Moreover, by relying on a technique for approximating complex distributions using simple ones, we demonstrate that the Edgeworth Accountant can be applied to the composition of any noise-addition mechanism. Owing to certain appealing features of the Edgeworth expansion, the $(\epsilon, \delta)$-differential privacy bounds offered by this accountant are non-asymptotic, with essentially no extra computational cost, as opposed to the prior approaches, wherein the running times increase with the number of compositions. Finally, we demonstrate that our upper and lower $(\epsilon, \delta)$-differential privacy bounds are tight in federated analytics and certain regimes of training private deep learning models.
We developed an efficient analytical tool via the Edgeworth expansion with finite-sample bounds to to keep track of DP guarantees with a large number of compositions.
Mixup, which creates synthetic training instances by linearly interpolating random sample pairs, is a simple and yet effective regularization technique to boost the performance of deep models trained with SGD. In this work, we report a previously unobserved phenomenon in Mixup raining: on a number of standard datasets, the performance of Mixup-trained models starts to decay after training for a large number of epochs, giving rise to a U-shaped generalization curve. This behavior is further aggravated when the size of original dataset is reduced. To help understand such a behavior of Mixup, we show theoretically that Mixup training may introduce undesired data-dependent label noises to the synthesized data. Via analyzing a least-square regression problem with a random feature model, we explain why noisy labels may cause the U-shaped curve to occur: Mixup improves generalization through fitting the clean patterns at the early training stage, but as training progresses, Mixup becomes over-fitting to the noise in the synthetic data. Extensive experiments are performed on a variety of benchmark datasets, validating this explanation.
We empirically discovered a U-shaped generalization curve of Mixup training.
Current graph representation learning techniques use Graph Neural Networks (GNNs) to extract features from dataset embeddings. In this work, we examine the quality of these embeddings and assess how changing them can affect the ac- curacy of GNNs. We explore different embedding extraction techniques for both images and texts; and find that the choice of embedding biases the performance of different GNN architectures and thus the choice of embedding influences the selection of GNNs regardless of the underlying dataset. In addition, we only see an improvement in accuracy from some GNN models compared to the accuracy of models trained from scratch or fine-tuned on the underlying data without utilising the graph connections. As an alternative, we propose Graph-connected Network (GraNet) layers to better leverage existing unconnected models within a GNN. Existing language and vision models are thus improved by allowing neighbour- hood aggregation. This gives a chance for the model to use pre-trained weights, if possible, and we demonstrate that this approach improves the accuracy compared to traditional GNNs: on Flickr v2, GraNet beats GAT2 and GraphSAGE by 7.7% and 1.7% respectively.
We question current graph neural network embedding quality and present new GNN techniques to use large models (pre-trained or trained from scratch) to work directly on graph-connected data
Deep generative modeling is a rapidly-advancing field with a wealth of modeling choices developed in the past decades. Amongst them, Wasserstein gradient flows (WGF) are a powerful and theoretically rich class of methods. However, their applications to high-dimensional distributions remain relatively underexplored. In this paper, we present Deep Generative Wasserstein Gradient Flows (DGGF), which constructs a WGF between two distributions by minimizing the entropy-regularized $f$-divergence. We demonstrate how to train a deep density ratio estimator that is required for the WGF and apply it to the task of generative modeling. Experiments demonstrate that DGGF is able to synthesize high-fidelity images of resolutions up to $128\times128$, directly in data space. We demonstrate that DGGF has an interpretable diagnostic of sample quality by naturally estimating the KL divergence throughout the gradient flow. Finally, we show DGGF's modularity by composition with external density ratio estimators for conditional generation, as well as for unpaired image-to-image translation with no modifications to the framework.
We scale Wasserstein gradient flows to high dimensional image generation tasks.
With the prevailing of machine learning (ML), researchers have shown that ML models are also vulnerable to various privacy and security attacks. As one of the representative attacks, the property inference attack aims to infer the private/sensitive properties of the training data (e.g., race distribution) given the output of ML models. In this paper, we present a new side channel for property inference attacks, i.e., t-SNE plots, which are widely used to show feature distribution or demonstrate model performance. We show for the first time that the private/sensitive properties of the data that are used to generate the plot can be successfully predicted. Briefly, we leverage the publicly available model as the shadow model to generate t-SNE plots with different properties. We use those plots to train an attack model, which is a simple image classifier, to infer the specific property of a given t-SNE plot. Extensive evaluation on four datasets shows that our proposed attack can effectively infer the undisclosed property of the data presented in the t-SNE plots, even when the shadow model is different from the target model used to generate the t-SNE plots. We also reveal that the attacks are robust in various scenarios, such as constructing the attack with fewer t-SNE plots/different density settings and attacking t-SNE plots generated by fine-tuned target models. The simplicity of our attack method indicates that the potential risk of leaking sensitive properties in t-SNE plots is largely underestimated. As possible defenses, we observe that adding noise to the image embeddings or t-SNE coordinates effectively mitigates attacks but can be bypassed by adaptive attacks, which prompts the need for more effective defenses.
We present for the first time that t-SNE plots can be a new valid side channel for property inference attacks
Coverage conditions---which assert that the data logging distribution adequately covers the state space---play a fundamental role in determining the sample complexity of offline reinforcement learning. While such conditions might seem irrelevant to online reinforcement learning at first glance, we establish a new connection by showing---somewhat surprisingly---that the mere existence of a data distribution with good coverage can enable sample-efficient online RL. Concretely, we show that coverability---that is, existence of a data distribution that satisfies a ubiquitous coverage condition called concentrability---can be viewed as a structural property of the underlying MDP, and can be exploited by standard algorithms for sample-efficient exploration, even when the agent does not know said distribution. We complement this result by proving that several weaker notions of coverage, despite being sufficient for offline RL, are insufficient for online RL. We also show that existing complexity measures for online RL, including Bellman rank and Bellman-Eluder dimension, fail to optimally capture coverability, and propose a new complexity measure, the self-normalized coefficient, to provide a unification.
This paper shows surprising connections between online and offline learnability, in particular, how coverage in offline RL enables exploration in online RL.
Despite recent advancements in language models (LMs), their application to dialogue management (DM) problems and ability to carry on rich conversations remain a challenge. We use reinforcement learning (RL) to develop a dialogue agent that avoids being short-sighted (outputting generic utterances) and maximizes overall user satisfaction. Most existing RL approaches to DM train the agent at the word-level, and thus, have to deal with a combinatorially complex action space even for a medium-size vocabulary. As a result, they struggle to produce a successful and engaging dialogue even if they are warm-started with a pre-trained LM. To address this issue, we develop a RL-based DM using a novel mixture of expert language model (MoE-LM) that consists of (i) a LM capable of learning diverse semantics for conversation histories, (ii) a number of specialized LMs (or experts) capable of generating utterances corresponding to a particular attribute or personality, and (iii) a RL-based DM that performs dialogue planning with the utterances generated by the experts. Our MoE approach provides greater flexibility to generate sensible utterances with different intents and allows RL to focus on conversational-level DM. We compare it with SOTA baselines on open-domain dialogues and demonstrate its effectiveness both in terms of the diversity and sensibility of the generated utterances and the overall DM performance.
A mixture-of-expert based dialogue manager that is amenable to sequential decision making techniques
Modern high-scoring models of vision in the brain score competition do not stem from Vision Transformers. However, in this paper, we provide evidence against the unexpected trend of Vision Transformers (ViT) being not perceptually aligned with human visual representations by showing how a dual-stream Transformer, a CrossViT $~\textit{a la}$ Chen et. al. (2021), under a joint rotationally-invariant and adversarial optimization procedure yields 2nd place in the aggregate Brain-Score 2022 competition (Schrimpf et al., 2020b) averaged across all visual categories, and at the time of the competition held 1st place for the highest explainable variance of area V4. In addition, our current Transformer-based model also achieves greater explainable variance for areas V4, IT, and Behaviour than a biologically-inspired CNN (ResNet50) that integrates a frontal V1-like computation module (Dapello et al., 2020). To assess the contribution of the optimization scheme with respect to the CrossViT architecture, we perform several additional experiments on differently optimized CrossViT's regarding adversarial robustness, common corruption benchmarks, mid-ventral stimuli interpretation, and feature inversion. Against our initial expectations, our family of results provides tentative support for an $\textit{``All roads lead to Rome''}$ argument enforced via a joint optimization rule even for non biologically-motivated models of vision such as Vision Transformers.
We provide evidence that a specific Vision Transformer under a joint rotationally-invariant and adversarial optimization procedure can reach state of the art Brain-Score for Area V4
Implicit neural representations (INRs) have recently emerged as a powerful tool that provides an accurate and resolution-independent encoding of data. Their robustness as general approximators has been shown in a wide variety of data sources, with applications on image, sound, and 3D scene representation. However, little attention has been given to leveraging these architectures for the representation and analysis of time series data. In this paper, we propose a new INR architecture for time series (iSIREN) designed to perform an accurate reconstruction of univariate and multivariate data, while also providing an interpretable encoding of the signal. We compare our architecture against SIREN and INRs with different activations, in terms of training convergence, and the reconstruction accuracy of both the signal and its spectral distribution. To achieve generalization, we propose a hypernetwork architecture (HyperTime) that leverages iSIRENs to learn a latent representation of an entire time series dataset. In addition to the traditional reconstruction loss, we introduce an FFT-based loss that guides the training by enforcing a good match of the ground truth spectral distribution. We show how these architectures can be used for time series generation, and evaluate our method through fidelity metrics, presenting results that exceed the performance of state-of-the-art techniques. Finally, we propose an alternative hypernetwork architecture (iHyperTime) that incorporates interpretability into the latent representation, enabling the introduction of prior knowledge by imposing constraints into the generation process.
We propose a time series specific implicit neural representation architecture, and use it to generate synthetic data.
Optimization problems with expensive nonlinear cost functions and combinatorial constraints appear in many real-world applications, but remain challenging to solve efficiently. Existing combinatorial solvers like Mixed Integer Linear Programming can be fast in practice but cannot readily optimize nonlinear cost functions, while general nonlinear optimizers like gradient descent often do not handle complex combinatorial structures, may require many queries of the cost function, and are prone to local optima. To bridge this gap, we propose SurCo that learns linear Surrogate costs which can be used by existing Combinatorial solvers to output good solutions to the original nonlinear combinatorial optimization problem, combining the flexibility of gradient-based methods with the structure of linear combinatorial optimization. We learn these linear surrogates end-to-end with the nonlinear loss by differentiating through the linear surrogate solver. Three variants of SurCo are proposed: SurCo-zero operates on individual nonlinear problems, SurCo-prior trains a linear surrogate predictor on distributions of problems, and SurCo-hybrid uses a model trained offline to warm start online solving for SurCo-zero. We analyze our method theoretically and empirically, showing smooth convergence and improved performance. Experiments show that compared to state-of-the-art approaches and expert-designed heuristics, SurCo obtains lower cost solutions with comparable or faster solve time for two real-world industry-level applications: embedding table sharding and inverse photonic design.
SurCo learns linear surrogate problems for nonlinear combinatorial optimization by training high-quality linear surrogates using end-to-end gradient descent with better performance in two industrial domains
Inferring reward functions from human behavior is at the center of value alignment – aligning AI objectives with what we, humans, actually want. But doing so relies on models of how humans behave given their objectives. After decades of research in cognitive science, neuroscience, and behavioral economics, obtaining accurate human models remains an open research topic. This begs the question: how accurate do these models need to be in order for the reward inference to be accurate? On the one hand, if small errors in the model can lead to catastrophic error in inference, the entire framework of reward learning seems ill-fated, as we will never have perfect models of human behavior. On the other hand, if as our models improve, we can have a guarantee that reward accuracy also improves, this would show the benefit of more work on the modeling side. We study this question both theoretically and empirically. We do show that it is unfortunately possible to construct small adversarial biases in behavior that lead to arbitrarily large errors in the inferred reward. However, and arguably more importantly, we are also able to identify reasonable assumptions under which the reward inference error can be bounded linearly in the error in the human model. Finally, we verify our theoretical insights in discrete and continuous control tasks with simulated and human data.
We investigate the impact of assuming wrong human models on reward learning.
The rise of Vision Transformers (ViT) combined with better self-supervised learning pre-tasks has taken representation learning to the next level, beating supervised results on ImageNet. In particular, self-attention mechanism of ViT allows to easily visualize semantic information learned by the network. Following revealing of attention maps of DINO, many tried to leverage its representations for unsupervised segmentation. Despite very promising results for basic images with a single clear object in a simple background, representation of ViT are not able to segment images, with several classes and object instance, in an unsupervised fashion yet. In this paper, we propose SALT: Semi-supervised Segmentation with Self-supervised Attention Layers in Transformers, an interactive algorithm for multi-class/multi-instance segmentation. We follow previous works path and take it a step further by discriminating between different objects, using sparse human help to select said objects. We show that remarkable results are achieved with very sparse labels. Different pre-tasks are compared, and we show that self-supervised ones are more robust for panoptic segmentation, and overall achieve very similar performance. Evaluation is carried out on Pascal VOC 2007 and COCO-panoptic. Performance is evaluated for extreme conditions such as very noisy, and sparse interactions going to as little as one interaction per class.
Multi-instance interactive segmentation using Label Propagation and self-supervised representations from Vision Transformer.
Tabular data is among the oldest and most ubiquitous forms of data. However, the generation of synthetic samples with the original data’s characteristics remains a significant challenge for tabular data. While many generative models from the computer vision domain, such as variational autoencoders or generative adversarial networks, have been adapted for tabular data generation, less research has been directed towards recent transformer-based large language models (LLMs), which are also generative in nature. To this end, we propose GReaT (Generation of Realistic Tabular data), which exploits an auto-regressive generative LLM to sample synthetic and yet highly realistic tabular data. Furthermore, GReaT can model tabular data distributions by conditioning on any subset of features; the remaining features are sampled without additional overhead. We demonstrate the effectiveness of the proposed approach in a series of experiments that quantify the validity and quality of the produced data samples from multiple angles. We find that GReaT maintains state-of-the-art performance across numerous real-world and synthetic data sets with heterogeneous feature types coming in various sizes.
The GReaT approach utilizes the capabilities of large language models to synthesize realistic tabular data. A challenging set of experiments validates the proposed method’s efficiency.
Finding efficient optimization methods plays an important role for quantum optimization and quantum machine learning on near-term quantum computers. While backpropagation on classical computers is computationally efficient, obtaining gradients on quantum computers is not, because the computational complexity scales linearly with the number of parameters and measurements. In this paper, we connect Koopman operator theory, which has been successful in predicting nonlinear dynamics, with natural gradient methods in quantum optimization. We propose a data-driven approach using Koopman operator learning to accelerate quantum optimization and quantum machine learning. We develop two new families of methods: the sliding window dynamic mode decomposition (DMD) and the neural DMD for efficiently updating parameters on quantum computers. We show that our methods can predict gradient dynamics on quantum computers and accelerate the quantum variational eigensolver used in quantum optimization, as well as quantum machine learning. We further implement the learning algorithms on a real quantum computer and demonstrate their practical effectiveness.
Koopman opeartor learning for accelerating quantum optimization and quantum machine learning.
Graph attention networks estimate the relational importance of node neighbors to aggregate relevant information over local neighborhoods for a prediction task. However, the inferred attentions are vulnerable to spurious correlations and connectivity in the training data, hampering the generalizability of the model. We introduce CAR, a general-purpose regularization framework for graph attention networks. Embodying a causal inference approach, CAR aligns the attention mechanism with the causal effects of active interventions on graph connectivity in a scalable manner. CAR is compatible with a variety of graph attention architectures, and we show that it systematically improves generalizability on various node classification tasks. Our ablation studies indicate that CAR hones in on the aspects of graph structure most pertinent to the prediction (e.g., homophily), and does so more effectively than alternative approaches. Finally, we also show that CAR enhances interpretability of attention weights by accentuating node-neighbor relations that point to causal hypotheses. For social media network-sized graphs, a CAR-guided graph rewiring approach could allow us to combine the scalability of graph convolutional methods with the higher performance of graph attention.
We introduce a general regularization framework for graph attention networks that aligns attention weights with the causal effects of interventions on graph connectivity.
Stochastic dynamics are ubiquitous in many fields of science, from the evolution of quantum systems in physics to diffusion-based models in machine learning. Existing methods such as score matching can be used to simulate these physical processes by assuming that the dynamics is a diffusion, which is not always the case. In this work, we propose a method called "Action Matching" that enables us to learn a much broader family of stochastic dynamics. Our method requires access only to samples from different time-steps, makes no explicit assumptions about the underlying dynamics, and can be applied even when samples are uncorrelated (i.e., are not part of a trajectory). Action Matching directly learns an underlying mechanism to move samples in time without modeling the distributions at each time-step. In this work, we showcase how Action Matching can be used for several computer vision tasks such as generative modeling, super-resolution, colorization, and inpainting; and further discuss potential applications in other areas of science.
We propose Action Matching for modeling stochastic dynamics by learning an underlying mechanism to move samples.
Blind image super-resolution (Blind-SR) aims to recover a high-resolution (HR) image from its corresponding low-resolution (LR) input image with unknown degradations. Most of the existing works design an explicit degradation estimator for each degradation to guide SR. However, it is infeasible to provide concrete labels of multiple degradation combinations (\eg, blur, noise, jpeg compression) to supervise the degradation estimator training. In addition, these special designs for certain degradation, such as blur, impedes the models from being generalized to handle different degradations. To this end, it is necessary to design an implicit degradation estimator that can extract discriminative degradation representation for all degradations without relying on the supervision of degradation ground-truth. In this paper, we propose a Knowledge Distillation based Blind-SR network (KDSR). It consists of a knowledge distillation based implicit degradation estimator network (KD-IDE) and an efficient SR network. To learn the KDSR model, we first train a teacher network: KD-IDE$_{T}$. It takes paired HR and LR patches as inputs and is optimized with the SR network jointly. Then, we further train a student network KD-IDE$_{S}$, which only takes LR images as input and learns to extract the same implicit degradation representation (IDR) as KD-IDE$_{T}$. In addition, to fully use extracted IDR, we design a simple, strong, and efficient IDR based dynamic convolution residual block (IDR-DCRB) to build an SR network. We conduct extensive experiments under classic and real-world degradation settings. The results show that KDSR achieves SOTA performance and can generalize to various degradation processes. The source codes and pre-trained models will be released.
We propose a knowledge distillation based blind super-resolution network, which can generalize to all degradation processes and achieve SOTA performance efficiently.
Graph generative models become increasingly effective for data distribution approximation and data augmentation. Although still in sandboxes, they have aroused public concerns about their malicious misuses or misinformation broadcasts, just as what Deepfake visual and auditory media has been delivering to society. It is never too early to regulate the prevalence of generated graphs. As a preventive response, we pioneer to formulate the generated graph detection problem to distinguish generated graphs from real ones. We propose the first framework to systematically investigate a set of sophisticated models and their performance in four classification scenarios. Each scenario switches between seen and unseen datasets/generators during testing to get closer to real world settings and progressively challenge the classifiers. Extensive experiments evidence that all the models are qualified for generated graph detection, with specific models having advantages in specific scenarios. Resulting from the validated generality and oblivion of the classifiers to unseen datasets/generators, we draw a safe conclusion that our solution can sustain for a decent while to curb generated graph misuses.
We propose a general framework to detect generated graphs using GNN-based methods.
Over the past few years afterward the birth of ResNet, skip connection has become the defacto standard for the design of modern architectures due to its widespread adoption, easy optimization, and proven performance. Prior work has explained the effectiveness of the skip connection mechanism from different perspectives. In this work, we deep dive into the model's behaviors with skip connections which can be formulated as a learnable Markov chain. An efficient Markov chain is preferred as it always maps the input data to the target domain in a better way. However, while a model is explained as a Markov chain, it is not guaranteed to be optimized following an efficient Markov chain by existing SGD-based optimizers prone to getting trapped in local optimal points. In order to move towards a more efficient Markov chain, we propose a simple routine of penal connection to make any residual-like model become a learnable Markov chain. Aside from that, the penal connection can also be viewed as a particular model regularization and can be easily implemented with one line of code in the most popular deep learning frameworks. The encouraging experimental results in multi-modal translation and image recognition empirically confirm our conjecture of the learnable Markov chain view and demonstrate the superiority of the proposed penal connection.
Penal connection only introduces negligible computational burden and can be implemented with one line of code under most popular deep learning frameworks.
A common strategy in curriculum generation for reinforcement learning is to train a teacher network to generate tasks that fall within a student network's ``zone of proximal development'' (ZPD). These are tasks that are not too easy and not too hard for the student. Albeit intuitive, ZPD is not well understood computationally. We propose ZONE, a novel computational framework that operationalizes ZPD. It formalizes ZPD through the language of Bayesian probability theory, revealing that tasks should be selected by difficulty (the student's success probability on the task) and learning progression (the degree of change in the student's model parameters). ZONE operationalizes ZPD with two techniques that we apply on top of existing algorithms. One is REJECT, which rejects tasks outside a difficulty scope and the other is GRAD, which prioritizes tasks that maximize the student's gradient norm. Compared to the original algorithms, the ZONE techniques improve the student’s generalization performance on discrete Minigrid environments and continuous control Mujoco domains with up to $9 \times$ higher success. ZONE also accelerates the student's learning by training on up to $10\times$ less data.
This work proposes a Bayesian computational framework to operationalize ``the zone of proximal development'' and to improve existing curriculum generation algorithms.
To provide rigorous uncertainty quantification for online learning models, we develop a framework for constructing uncertainty sets that provably control risk---such as coverage of confidence intervals, false negative rate, or F1 score---in the online setting. This extends conformal prediction to apply to a larger class of online learning problems. Our method guarantees risk control at any user-specified level even when the underlying data distribution shifts drastically, even adversarially, over time in an unknown fashion. The technique we propose is highly flexible as it can be applied with any base online learning algorithm (e.g., a deep neural network trained online), requiring minimal implementation effort and essentially zero additional computational cost. We further extend our approach to control multiple risks simultaneously, so the prediction sets we generate are valid for all given risks. To demonstrate the utility of our method, we conduct experiments on real-world tabular time-series data sets showing that the proposed method rigorously controls various natural risks. Furthermore, we show how to construct valid intervals for an online image-depth estimation problem that previous sequential calibration schemes cannot handle.
A flexible tool for constructing uncertainty estimates with a rigorous long-range risk control (such as coverage, false negative rate, or F1 score) in an online learning setting, where the distribution can vary greatly over time.
Intent detection with semantically similar fine-grained intents is a challenging task. To address it, we reformulate intent detection as a question-answering retrieval task by treating utterances and intent names as questions and answers. To that end, we utilize a question-answering retrieval architecture and adopt a two stages training schema with batch contrastive loss. In the pre-training stage, we improve query representations through self-supervised training. Then, in the fine-tuning stage, we increase contextualized token-level similarity scores between queries and answers from the same intent. Our results on three few-shot intent detection benchmarks achieve state-of-the-art performance.
Our method achieve SOTA results on few-shot intent detection by combining Question-Answering architecture, Contrastive Learning techniques and use of the intent name as answer.
In semi-supervised learning, student-teacher distribution matching has been successful in improving performance of models using unlabeled data in conjunction with few labeled samples. In this paper, we aim to replicate that success in the self-supervised setup where we do not have access to any labeled data during pre-training. We show it is possible to induce the student-teacher distributions without any knowledge of downstream classes by using a queue of embeddings of samples from the unlabeled dataset. We show that Q-Match outperforms previous self-supervised learning techniques on tabular datasets when measuring downstream classification performance. Furthermore, we show that our method is sample efficient, both in terms of labels required for both downstream task training and amount of unlabeled data required for pre-training.
A self-supervised method to train models by minimizing the cross-entropy loss between student-teacher distributions generated using a queue of embeddings. This results in better downstream task performance with less labeled data.
Predicting discrete events in time and space has many scientific applications, such as predicting hazardous earthquakes and outbreaks of infectious diseases. History-dependent spatio-temporal Hawkes processes are often used to mathematically model these point events. However, previous approaches have faced numerous challenges, particularly when attempting to forecast multiple future events. In this work, we propose a new neural architecture for multi-event forecasting of spatio-temporal point processes, utilizing transformers, augmented with normalizing flows and probabilistic layers. Our network makes batched predictions of complex history-dependent spatio-temporal distributions of future discrete events, achieving state-of-the-art performance on a variety of benchmark datasets including the South California Earthquakes, Citibike, Covid19, and Hawkes synthetic Pinwheel datasets. More generally, we illustrate how our network can be applied to any dataset of discrete events with associated markers, even when no underlying physics is known.
In this work, we introduce a novel neural network that is capable of simultaneous multi-event forecasting of spatio-temporal distributions associated with stochastic discrete events.
Traditional machine learning models focus on achieving good performance on the overall training distribution, but they often underperform on minority groups. Existing methods can improve the worst-group performance, but they can have several limitations: (i) they require group annotations, which are often expensive and sometimes infeasible to obtain, and/or (ii) they are sensitive to outliers. Most related works fail to solve these two issues simultaneously as they focus on conflicting perspectives of minority groups and outliers. We address the problem of learning group annotations in the presence of outliers by clustering the data in the space of gradients of the model parameters. We show that data in the gradient space has a simpler structure while preserving information about minority groups and outliers, making it suitable for standard clustering methods like DBSCAN. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art both in terms of group identification and downstream worst-group performance.
We propose to perform clustering in the gradient space for outlier-robust group identification, thereby learning distributionally and outlier robust models when group labels are unavailable.
The recent rapid advances in the development and deployment of machine learning technologies largely depend on the vast richness of data available today, in terms of both the quantity and the rich content contained within. For example, biometric data such as images and voices could reveal people's attributes like age, gender, sentiment, and origin, whereas location/motion data could be used to infer people's activity levels, transportation modes, and life habits. Along with the new services and applications enabled by such technological advances, various governmental policies are put in place to regulate such data usage and protect people's privacy and rights. As a result, data owners often opt for simple data obfuscation (e.g., blur people's faces in images) or withholding data altogether, which leads to severe data quality degradation and greatly limits the data's potential utility. Aiming for a sophisticated mechanism which gives data owners fine-grained control while retaining the maximal degree of data utility, we propose Multi-attribute Selective Suppression, or MaSS, a general framework for performing precisely targeted data surgery to simultaneously suppress any selected set of attributes while preserving the rest for downstream machine learning tasks. MaSS learns a data modifier through adversarial games between two sets of networks, where one is aimed at suppressing selected attributes, and the other ensures the retention of the rest of the attributes via general contrastive loss as well as explicit classification metrics. We carried out an extensive evaluation of our proposed method using multiple datasets from different domains including facial images, voice audio, and video clips, and obtained highly promising results in MaSS' generalizability and capability of drastically suppressing targeted attributes (e.g., reducing inference on such attributes to random guess) while imposing virtually no impact on the data's usability in other downstream ML tasks.
Selectively suppress the attributes while preserving the rest of attributes
Recently, transformers have shown strong ability as visual feature extractors, surpassing traditional convolution-based models in various scenarios. However, the success of vision transformers largely owes to their capacity to accommodate numerous parameters. As a result, new challenges for adapting a well-trained transformer to downstream tasks arise. On the one hand, classic fine-tuning tunes all parameters in a huge model for every downstream task and thus easily falls into an overfitting situation, leading to inferior performance. On the other hand, on resource-limited devices, fine-tuning stores a full copy of all parameters and thus is usually impracticable for the shortage of storage space. However, few works have focused on how to efficiently and effectively transfer knowledge in a vision transformer. Existing methods did not dive into the properties of visual features, leading to inferior performance. Moreover, some of them bring heavy inference cost though benefiting storage. To tackle these problems, we propose consolidator to achieve efficient transfer learning for large vision models. Our consolidator modifies the pre-trained model with the addition of a small set of tunable parameters to temporarily store the task-specific knowledge while freezing the backbone model during adaptation. Motivated by the success of group-wise convolution, we adopt grouped connections across the features extracted by fully connected layers to construct tunable parts in a consolidator. To further enhance the model's capacity to transfer knowledge under a constrained storage budget and keep inference efficient, we consolidate the parameters in two stages: 1. between adaptation and storage, and 2. between loading and inference. On a series of downstream visual tasks, our consolidator can reach up to 7.56 better accuracy than full fine-tuning with merely 0.35% parameters, and outperform state-of-the-art parameter-efficient tuning methods by a clear margin. Code is available at github.
We propose a module named consolidator to achieve both parameter- and inference-efficient transfer learning for vision transformers
Most factorized bilinear pooling (FBiP) employs Hadamard product-based bilinear projection to learn appropriate projecting directions to reduce the dimension of bilinear features. However, in this paper, we reveal that the Hadamard product-based bilinear projection makes FBiP miss a lot of possible projecting directions, which will significantly harm the performance of outputted compact bilinear features, including compactness and effectiveness. To address this issue, we propose a general matrix-based bilinear projection based on the rank-$k$ matrix base decomposition, where the Hadamard-based bilinear projection is a special case of our proposed one. Using the proposed bilinear projection, we design a novel low-rank factorized bilinear pooling (named RK-FBP), which does not miss any projecting directions. Thus, our RK-FBP can generate better compact bilinear features. To leverage high-order information in local features, we nest several RK-FBP modules together to formulate a multi-linear pooling that outputs compact multi-linear features. At last, we conduct experiments on several fine-grained image tasks to evaluate our models. The experiments show that our models achieve new state-of-the-art classification accuracy by the lowest dimension.
We proposed a general bilinear projection based on complete matrix bases, and then we design a compact bilinear pooling algorithm by using the proposed general bilinear pooling.
This paper investigates a missing feature imputation problem for graph learning tasks. Several methods have previously addressed learning tasks on graphs with missing features. However, in cases of high rates of missing features, they were unable to avoid significant performance degradation. To overcome this limitation, we introduce a novel concept of channel-wise confidence in a node feature, which is assigned to each imputed channel feature of a node for reflecting the certainty of the imputation. We then design pseudo-confidence using the channel-wise shortest path distance between a missing-feature node and its nearest known-feature node to replace unavailable true confidence in an actual learning process. Based on the pseudo-confidence, we propose a novel feature imputation scheme that performs channel-wise inter-node diffusion and node-wise inter-channel propagation. The scheme can endure even at an exceedingly high missing rate (e.g., 99.5\%) and it achieves state-of-the-art accuracy for both semi-supervised node classification and link prediction on various datasets containing a high rate of missing features. Codes are available at https://github.com/daehoum1/pcfi.
For graphs with missing features, we define a novel concept of confidence and propose a pseudo-confidence-based feature imputation (PCFI) scheme.
We present a comparative study on how and why contrastive learning (CL) and masked image modeling (MIM) differ in their representations and in their performance of downstream tasks. In particular, we demonstrate that self-supervised Vision Transformers (ViTs) have the following properties: (1) CL trains self-attentions to capture longer-range global patterns than MIM, such as the shape of an object, especially in the later layers of the ViT architecture. This CL property helps ViTs linearly separate images in their representation spaces. However, it also makes the self-attentions collapse into homogeneity for all query tokens and heads. Such homogeneity of self-attention reduces the diversity of representations, worsening scalability and dense prediction performance. (2) CL utilizes the low-frequency signals of the representations, but MIM utilizes high-frequencies. Since low- and high-frequency information respectively represent shapes and textures, CL is more shape-oriented and MIM more texture-oriented. (3) CL plays a crucial role in the later layers, while MIM mainly focuses on the early layers. Upon these analyses, we find that CL and MIM can complement each other and observe that even the simplest harmonization can help leverage the advantages of both methods. The code is available at https://github.com/naver-ai/cl-vs-mim.
We show that (1) CL primarily captures global patterns compared with MIM, (2) CL is more shape-oriented whereas MIM is more texture-oriented, and (3) CL plays a key role in the later layers while MIM focuses on the early layers.
Pretraining methods are typically compared by evaluating the accuracy of linear classifiers, transfer learning performance, or visually inspecting the representation manifold's (RM) lower-dimensional projections. We show that the differences between methods can be understood more clearly by investigating the RM directly, which allows for a more detailed comparison. To this end, we propose a framework and new metric to measure and compare different RMs. We also investigate and report on the RM characteristics for various pretraining methods. These characteristics are measured by applying sequentially larger local alterations to the input data, using white noise injections and Projected Gradient Descent (PGD) adversarial attacks, and then tracking each datapoint. We calculate the total distance moved for each datapoint and the relative change in distance between successive alterations. We show that self-supervised methods learn an RM where alterations lead to large but constant size changes, indicating a smoother RM than fully supervised methods. We then combine these measurements into one metric, the Representation Manifold Quality Metric (RMQM), where larger values indicate larger and less variable step sizes, and show that RMQM correlates positively with performance on downstream tasks.
We introduce the Representation manifold quality metric (RMQM), which measures the structure of the learned representation manifold, where we then show that RMQM correlates positively to the generalisation of neural networks to downstream tasks.
Recently, 2D semantic segmentation has witnessed a significant advancement thanks to the huge amount of 2D image datasets available. Therefore, in this work, we propose the first 2D-to-3D knowledge distillation strategy to enhance 3D semantic segmentation model with knowledge embedded in the latent space of powerful 2D models. Specifically, unlike standard knowledge distillation, where teacher and student models take the same data as input, we use 2D panoramas properly aligned with corresponding 3D rooms to train the teacher network and use the learned knowledge from 2D teacher to guide 3D student. To facilitate our research, we create a large-scale, fine-annotated 3D semantic segmentation benchmark, containing voxel-wise semantic labels and aligned panoramas of 5175 scenes. Based on this benchmark, we propose a 3D volumetric semantic segmentation network, which adapts Video Swin Transformer as backbone and introduces a skip connected linear decoder. Achieving a state-of-the-art performance, our 3D Segmenter is computationally efficient and only requires $3.8\%$ of the parameters compared to the prior art. Our code and data will be released upon acceptance.
Distill knowledge from 2D strong model to enhance 3D semantic segmentation
Unlike conventional grid and mesh based methods for solving partial differential equations (PDEs), neural networks have the potential to break the curse of dimensionality, providing approximate solutions to problems where using classical solvers is difficult or impossible. While global minimization of the PDE residual over the network parameters works well for boundary value problems, catastrophic forgetting impairs applicability to initial value problems (IVPs). In an alternative local-in-time approach, the optimization problem can be converted into an ordinary differential equation (ODE) on the network parameters and the solution propagated forward in time; however, we demonstrate that current methods based on this approach suffer from two key issues. First, following the ODE produces an uncontrolled growth in the conditioning of the problem, ultimately leading to unacceptably large numerical errors. Second, as the ODE methods scale cubically with the number of model parameters, they are restricted to small neural networks, significantly limiting their ability to represent intricate PDE initial conditions and solutions. Building on these insights, we develop Neural-IVP, an ODE based IVP solver which prevents the network from getting ill-conditioned and runs in time linear in the number of parameters, enabling us to evolve the dynamics of challenging PDEs with neural networks.
We develop Neural-IVP, a new method for solving initial value PDEs with Neural Networks that is both stable and scalable.
Recently, there has been considerable interests in understanding pretrained language models. This work studies the hidden geometry of the representation space of language models from a unique topological perspective. We hypothesize that there exist a network of latent anchor states summarizing the topology (neighbors and connectivity) of the representation space. we infer this latent network in a fully unsupervised way using a structured variational autoencoder. We show that such network exists in pretrained representations, but not in baseline random or positional embeddings. We connect the discovered topological structure to their linguistic interpretations. In this latent network, leave nodes can be grounded to word surface forms, anchor states can be grounded to linguistic categories, and connections between nodes and states can be grounded to phrase constructions and syntactic templates. We further show how such network evolves as the embeddings become more contextualized, with observational and statistical evidence demonstrating how contextualization helps words “receive meaning” from their topological neighbors via the anchor states. We demonstrate these insights with extensive experiments and visualizations.
We discover the hidden topology within the representation space of contextualized representations
We consider the task of learning to recover clean signals given only access to noisy data. Recent work in computer vision has addressed this problem in the context of images using denoising autoencoders (DAEs). However, to date DAEs for learning from noisy data have not been explored in the context of time-series data. DAEs for denoising images often rely on assumptions unlikely to hold in the context of time series, \textit{e.g.}, multiple noisy samples of the same example. Here, we adapt DAEs to cleaning time-series data with noisy samples only. To recover the clean target signal when only given access to noisy target data, we leverage a noise-free auxiliary time-series signal that is related to the target signal. In addition to leveraging the relationship between the target signal and auxiliary signal, we iteratively filter and learn from clean samples using an approach based on co-teaching. Applied to the task of recovering carbohydrate values for blood glucose management, our approach reduces noise (MSE) in patient-reported carbohydrates from 72$g^2$ (95\% CI: 54,93) to 18$g^2$ (13,25), outperforming the best baseline (MSE = 33$g^2$ (27,43)). We demonstrate strong time-series denoising performance, extending the applicability of DAEs to a previously under-explored setting.
We combine Co-teaching and De-noising Autoencoders to recover clean signals from only noisy data in a time series setting.
Lexicographic multi-objective problems, which impose a lexicographic importance order over the objectives, arise in many real-life scenarios. Existing Reinforcement Learning work directly addressing lexicographic tasks has been scarce. The few proposed approaches were all noted to be heuristics without theoretical guarantees as the Bellman equation is not applicable to them. Additionally, the practical applicability of these prior approaches also suffers from various issues such as not being able to reach the goal state. While some of these issues have been known before, in this work we investigate further shortcomings, and propose fixes for improving practical performance in many cases. We also present a policy optimization approach using our Lexicographic Projection Optimization (LPO) algorithm that has the potential to address these theoretical and practical concerns. Finally, we demonstrate our proposed algorithms on benchmark problems.
We investigate reinforcement learning for thresholded lexicographic ordered multi-objective settings.
It is unclear how changing the learning rule of a deep neural network alters its learning dynamics and representations. To gain insight into the relationship between learned features, function approximation, and the learning rule, we analyze infinite-width deep networks trained with gradient descent (GD) and biologically-plausible alternatives including feedback alignment (FA), direct feedback alignment (DFA), and error modulated Hebbian learning (Hebb), as well as gated linear networks (GLN). We show that, for each of these learning rules, the evolution of the output function at infinite width is governed by a time varying effective neural tangent kernel (eNTK). In the lazy training limit, this eNTK is static and does not evolve, while in the rich mean-field regime this kernel's evolution can be determined self-consistently with dynamical mean field theory (DMFT). This DMFT enables comparisons of the feature and prediction dynamics induced by each of these learning rules. In the lazy limit, we find that DFA and Hebb can only learn using the last layer features, while full FA can utilize earlier layers with a scale determined by the initial correlation between feedforward and feedback weight matrices. In the rich regime, DFA and FA utilize a temporally evolving and depth-dependent NTK. Counterintuitively, we find that FA networks trained in the rich regime exhibit more feature learning if initialized with smaller correlation between the forward and backward pass weights. GLNs admit a very simple formula for their lazy limit kernel and preserve conditional Gaussianity of their preactivations under gating functions. Error modulated Hebb rules show very small task-relevant alignment of their kernels and perform most task relevant learning in the last layer.
A theoretical analysis of deep networks and their representations when trained with a variety of learning rules.
Multi-vector retrieval models improve over single-vector dual encoders on many information retrieval tasks. In this paper, we cast the multi-vector retrieval problem as sparse alignment between query and document tokens. We propose ALIGNER, a novel multi-vector retrieval model that learns sparsified pairwise alignments between query and document tokens (e.g. `dog' vs. `puppy') and per-token unary saliences reflecting their relative importance for retrieval. We show that controlling the sparsity of pairwise token alignments often brings significant performance gains. While most factoid questions focusing on a specific part of a document require a smaller number of alignments, others requiring a broader understanding of a document favor a larger number of alignments. Unary saliences, on the other hand, decide whether a token ever needs to be aligned with others for retrieval (e.g. `kind' from `what kind of currency is used in new zealand'). With sparsified unary saliences, we are able to prune a large number of query and document token vectors and improve the efficiency of multi-vector retrieval. We learn the sparse unary saliences with entropy-regularized linear programming, which outperforms other methods to achieve sparsity. In a zero-shot setting, ALIGNER scores 51.1 nDCG@10, achieving a new retriever-only state-of-the-art on 13 tasks in the BEIR benchmark. In addition, adapting pairwise alignments with a few examples (<= 8) further improves the performance up to 15.7 points nDCG@10 for argument retrieval tasks. The unary saliences of ALIGNER helps us to keep only 20% of the document token representations with minimal performance loss. We further show that our model often produces interpretable alignments and significantly improves its performance when initialized from larger language models.
We propose a novel multi-vector retrieval model with pairwise alignment and unary salience.
Video scene graph generation (VidSGG) aims to generate a sequence of graph-structure representations for the given video. However, all existing VidSGG methods are fully-supervised, i.e., they need dense and costly manual annotations. In this paper, we propose the first weakly-supervised VidSGG task with only single-frame weak supervision: SF-VidSGG. By ``weakly-supervised", we mean that SF-VidSGG relaxes the training supervision from two different levels: 1) It only provides single-frame annotations instead of all-frame annotations. 2) The single-frame ground-truth annotation is still a weak image SGG annotation, i.e., an unlocalized scene graph. To solve this new task, we also propose a novel Pseudo Label Assignment based method, dubbed as PLA. PLA is a two-stage method, which generates pseudo visual relation annotations for the given video at the first stage, and then trains a fully-supervised VidSGG model with these pseudo labels. Specifically, PLA consists of three modules: an object PLA module, a predicate PLA module, and a future predicate prediction (FPP) module. Firstly, in the object PLA, we localize all objects for every frame. Then, in the predicate PLA, we design two different teachers to assign pseudo predicate labels. Lastly, in the FPP module, we fusion these two predicate pseudo labels by the regularity of relation transition in videos. Extensive ablations and results on the benchmark Action Genome have demonstrated the effectiveness of our PLA.
We propose a novel method for weakly-supervised VidSGG task with only single-frame weak supervision.
Semi-supervised learning aims to train a model using limited labels. State-of-the-art semi-supervised methods for image classification such as PAWS rely on self-supervised representations learned with large-scale unlabeled but curated data. However, PAWS is often less effective when using real-world unlabeled data that is uncurated, e.g., contains out-of-class data. We propose RoPAWS, a robust extension of PAWS that can work with real-world unlabeled data. We first reinterpret PAWS as a generative classifier that models densities using kernel density estimation. From this probabilistic perspective, we calibrate its prediction based on the densities of labeled and unlabeled data, which leads to a simple closed-form solution from the Bayes' rule. We demonstrate that RoPAWS significantly improves PAWS for uncurated Semi-iNat by +5.3% and curated ImageNet by +0.4%.
We propose a robust semi-supervised learning method for uncurated data derived from a novel probabilistic view of learned representations
Index structure is a fundamental component in database and facilitates broad data retrieval applications. Recent learned index methods show superior performance by learning hidden yet useful data distribution with the help of machine learning, and provide a guarantee that the prediction error is no more than a pre-defined $\epsilon$. However, existing learned index methods adopt a fixed $\epsilon$ for all the learned segments, neglecting the diverse characteristics of different data localities. In this paper, we propose a mathematically-grounded learned index framework with dynamic $\epsilon$, which is efficient and pluggable to existing learned index methods. We theoretically analyze prediction error bounds that link $\epsilon$ with data characteristics for an illustrative learned index method. Under the guidance of the derived bounds, we learn how to vary $\epsilon$ and improve the index performance with a better space-time trade-off. Experiments with real-world datasets and several state-of-the-art methods demonstrate the efficiency, effectiveness and usability of the proposed framework.
Based on the theoretically derived prediction error bounds, we propose a mathematically-grounded learned index framework with dynamic $\epsilon$, which is efficient and pluggable to several state-of-the-art learned index methods.
Distributional reinforcement learning~(RL) is a class of state-of-the-art algorithms that estimate the entire distribution of the total return rather than only its expectation. The empirical success of distributional RL is determined by the representation of return distributions and the choice of distribution divergence. In this paper, we propose a new class of \textit{Sinkhorn distributional RL~(SinkhornDRL)} algorithm that learns a finite set of statistics, i.e., deterministic samples, from each return distribution and then uses Sinkhorn iterations to evaluate the Sinkhorn distance between the current and target Bellmen distributions. Sinkhorn divergence features as the interpolation between the Wasserstein distance and Maximum Mean Discrepancy~(MMD). SinkhornDRL finds a sweet spot by taking advantage of the geometry of optimal transport-based distance and the unbiased gradient estimate property of MMD. Finally, compared to state-of-the-art algorithms, SinkhornDRL's competitive performance is demonstrated on the suit of 55 Atari games.
We designed a new class of distributional RL algorithm based on Sinkhorn divergence.
The compute effort required to perform inference on state-of-the-art deep learning models is ever growing. Practical applications are commonly limited to a certain cost per inference. Cascades of pretrained models with conditional execution address these requirements based on the intuition that some inputs are easy enough that they can be processed correctly by a small model allowing for an early exit. If the small model is not sufficiently confident in its prediction, the input is passed on to a larger model. The selection of the confidence threshold allows to trade off compute effort against accuracy. In this work, we explore the effective design of model cascades, and thoroughly evaluate the impact on the accuracy-compute trade-off. We find that they not only interpolate favorably between pretrained models, but that this trade-off curve commonly outperforms single models. This allows us to redefine most of the ImageNet Pareto front already with 2-model cascades, achieving an average reduction in compute effort at equal accuracy of almost 3.1x above 86% and more than 1.9x between 80% and 86% top-1 accuracy. We confirm the wide applicability and effectiveness of the method on the GLUE benchmark. We release the code to reproduce our experiments in the supplementary material and use only publicly available models and datasets.
We show how to combine pairs of pretrained models to improve the entire ImageNet accuracy-compute Pareto front.
Many applications of quantum computing in the near term rely on variational quantum circuits (VQCs). They have been showcased as a promising model for reaching a quantum advantage in machine learning with current noisy intermediate scale quantum computers (NISQ). It is often believed that the power of VQCs relies on their exponentially large feature space, and extensive works have explored the expressiveness and trainability of VQCs in that regard. In our work, we propose a classical sampling method that can closely approximate most VQCs with Hamiltonian encoding, given only the description of their architecture. It uses the seminal proposal of Random Fourier Features (RFF) and the fact that VQCs can be seen as large Fourier series. We show theoretically and experimentally that models built from exponentially large quantum feature space can be classically reproduced by sampling a few frequencies to build an equivalent low dimensional kernel. Precisely, we show that the number of required samples grows favourably with the size of the quantum spectrum. This tool therefore questions the hope for quantum advantage from VQCs in many cases, but conversely helps to narrow the conditions for their potential success. We expect VQCs with various and complex encoding Hamiltonians, or with large input dimension, to become more robust to classical approximations.
We show theoretically and experimentally that models built from exponentially large quantum feature space can be classically reproduced by sampling a few frequencies to build an equivalent low dimensional kernel
This paper addresses a regression problem in which output label values represent the results of sensing the magnitude of a phenomenon. A low value of such labels can either mean that the actual magnitude of the phenomenon has been low or that the sensor has made an incomplete observation. This leads to a bias toward lower values in labels and its resultant learning because labels for incomplete observations are recorded as lower than those for typical observations, even if both have monitored similar phenomena. Moreover, because an incomplete observation does not provide any tags indicating incompleteness, we cannot eliminate or impute them. To address this issue, we propose a learning algorithm that explicitly models the incomplete observations to be corrupted with an asymmetric noise that always has a negative value. We show that our algorithm is unbiased with a regression learned from the uncorrupted data that does not involve incomplete observations. We demonstrate the advantages of our algorithm through numerical experiments.
This paper addresses a regression problem for sensor magnitude in which a low value of labels can also mean incomplete observation. We derive an unbiased learning algorithm with a regression learned from data without incomplete observations.
README.md exists but content is empty. Use the Edit dataset card button to edit it.
Downloads last month
0
Edit dataset card