abstracts listlengths 2 2 | id_1 stringclasses 500
values | id_2 stringclasses 500
values | pair_id stringlengths 21 24 |
|---|---|---|---|
[
"This paper introduces **IDGP (Instance-Dependent Generative Process)**, a novel approach for Partial Label Learning (PLL) that specifically addresses the more realistic and challenging scenario of instance-dependent candidate labels. Unlike many existing PLL methods that assume incorrect labels are randomly picked... | lKOfilXucGB | NPrsUQgMjKK | lKOfilXucGB_NPrsUQgMjKK |
[
"This paper introduces **SENet** (Sensitivity-driven Efficient Neural Network), a novel three-stage training methodology designed to linearize Deep Neural Networks (DNNs) by significantly reducing the number of ReLU non-linearity operations. This reduction is crucial for efficient and secure Private Inference (PI),... | BGF9IeDfmlH | a6rCdfABJXg | BGF9IeDfmlH_a6rCdfABJXg |
[
"This paper introduces **Proto-Value Networks (PVNs)**, a novel approach for scaling representation learning in deep reinforcement learning (RL) using auxiliary tasks. The core idea is to leverage the concept of the successor measure to generate a rich, potentially infinite, source of auxiliary tasks that can effec... | oGDKSt9JrZi | Ki4ocDm364 | oGDKSt9JrZi_Ki4ocDm364 |
[
"The paper introduces **RandProx**, a novel primal-dual optimization algorithm designed for solving large-scale, nonsmooth optimization problems, particularly those prevalent in machine learning. It builds upon the deterministic Primal-Dual Davis-Yin (PDDY) algorithm by introducing randomization into its dual updat... | cB4N3G5udUS | uzFQpkEzOo | cB4N3G5udUS_uzFQpkEzOo |
[
"This paper introduces **Federated Nearest Neighbor (FedNN)**, a novel framework for Neural Machine Translation (NMT) that addresses the high communication and synchronization overheads of traditional Federated Learning (FL) approaches like FedAvg, while also ensuring user data privacy.\n\n**Problem:**\nTraditional... | R1U5G2spbLd | kIAx30hYi_p | R1U5G2spbLd_kIAx30hYi_p |
[
"This paper introduces a novel deep learning model for predicting the activation of olfactory receptors (ORs) by odorant molecules, a crucial step in understanding the mammalian sense of smell. The authors address the challenge of predicting OR-molecule interactions, especially given that many ORs lack identified a... | q9VherQJd8_ | 3F6I-0-57SC | q9VherQJd8__3F6I-0-57SC |
[
"This paper, \"ON THE EFFECTIVENESS OF OUT-OF-DISTRIBUTION DATA IN SELF-SUPERVISED LONG-TAIL LEARNING,\" addresses the significant challenge of poor generalization of Self-Supervised Learning (SSL) methods on long-tailed datasets. Existing approaches often rely on collecting additional in-domain (ID) data to re-bal... | v8JIQdiN9Sh | q-PbpHD3EOk | v8JIQdiN9Sh_q-PbpHD3EOk |
[
"The paper introduces **AgentNet**, a novel Graph Neural Network (GNN) architecture specifically designed for **graph-level tasks**. It departs fundamentally from traditional message-passing GNNs by adopting an **agent-based approach** inspired by sublinear algorithms, aiming for a computational complexity that is ... | 8WTAh0tj2jC | 77lSWa-Tm3Z | 8WTAh0tj2jC_77lSWa-Tm3Z |
[
"This paper introduces **Masked Unsupervised Self-Training (MUST)**, a novel and effective method for label-free image classification. MUST aims to improve the performance of pre-trained zero-shot classifiers, such as CLIP, by unsupervised finetuning on abundant unlabeled data from a target domain.\n\n**Problem Add... | ZAKkiVxiAM9 | ZrEbzL9eQ3W | ZAKkiVxiAM9_ZrEbzL9eQ3W |
[
"This paper, \"Language Models Can Teach Themselves to Program Better,\" introduces a novel self-improvement pipeline that enables Language Models (LMs) to enhance their programming abilities by generating and verifying their own instructive programming problems and solutions. Inspired by self-play in games like Go... | SaRj2ka1XZ3 | _hb4vM3jspB | SaRj2ka1XZ3__hb4vM3jspB |
[
"This paper introduces a novel framework for **Causal Imitation Learning (CIL) via Inverse Reinforcement Learning (IRL)**, addressing the critical challenges of unobserved confounders (UCs), sensory mismatches between expert and imitator, and latent reward signals in imitation learning.\n\n**Problem Statement:**\nT... | B-z41MBL_tH | FYZCHEtt6H0 | B-z41MBL_tH_FYZCHEtt6H0 |
[
"This paper introduces a scalable sampling-based Bayesian inference method for large linear models, with a primary application to the linearized Laplace method for neural network uncertainty quantification. The core problem addressed is the computational intractability of traditional Bayesian inference for linear m... | aoDyX6vSqsd | v8JIQdiN9Sh | aoDyX6vSqsd_v8JIQdiN9Sh |
[
"This paper, \"WORDS ARE ALL YOU NEED? LANGUAGE AS AN APPROXIMATION FOR HUMAN SIMILARITY JUDGMENTS,\" addresses the significant cost and scalability challenges associated with collecting human similarity judgments (which typically require O(N^2) judgments for N stimuli). While pre-trained Deep Neural Networks (DNNs... | O-G91-4cMdv | uu1GBD9SlLe | O-G91-4cMdv_uu1GBD9SlLe |
[
"This paper investigates and proposes a method for tuning the frequency bias observed in Neural Network (NN) training, particularly when dealing with **nonuniform data distributions**, a common scenario in practice that existing theories often overlook.\n\n**Method Used:**\n\n1. **Reinterpreting Empirical Loss via... | oLIZ2jGTiv | g4OTKRKfS7R | oLIZ2jGTiv_g4OTKRKfS7R |
[
"This paper introduces **Textual Inversion**, a novel method for personalizing text-to-image generation models to incorporate user-provided concepts (objects or styles) using only a few example images.\n\n**Method:**\nThe core idea is to represent a new, unique concept as a \"pseudo-word\" (denoted as S*) within th... | NAQvF08TcyG | zt53IDUR1U | NAQvF08TcyG_zt53IDUR1U |
[
"This paper introduces **Decomposed Prompting (DECOMP)**, a modular approach designed to enhance Large Language Models (LLMs) in solving complex tasks, particularly where traditional few-shot prompting or Chain-of-Thought (CoT) methods struggle due to task complexity, difficulty in learning individual reasoning ste... | _nGgzQjzaRy | 0paCJSFW7j | _nGgzQjzaRy_0paCJSFW7j |
[
"This paper introduces **Meta Knowledge Condensation for Federated Learning (FedMK)**, a novel approach designed to significantly reduce communication costs and mitigate data heterogeneity in Federated Learning (FL) while achieving superior model performance.\n\n**The Core Problem Addressed:**\nTraditional FL parad... | TDf-XFAwc79 | ULnHxczCBaE | TDf-XFAwc79_ULnHxczCBaE |
[
"This paper addresses the challenge of optimizing neural network architectures for equivariance, particularly when the underlying symmetries in the data are imperfectly known or not perfectly explicit. While strong equivariance constraints can improve performance and generalization, they can also be overly restrict... | a6rCdfABJXg | hLbeJ6jObDD | a6rCdfABJXg_hLbeJ6jObDD |
[
"This paper, \"Layer Grafted Pre-training: Bridging Contrastive Learning and Masked Image Modeling for Label-Efficient Representations,\" proposes a novel self-supervised pre-training method that effectively combines the strengths of Masked Image Modeling (MIM) and Contrastive Learning (CL).\n\n**Problem and Key In... | jwdqNwyREyh | 0f-0I6RFAch | jwdqNwyREyh_0f-0I6RFAch |
[
"The paper introduces **Simple and Scalable Nearest Neighbor Machine Translation (SK-MT)**, an approach designed to address the significant storage and computational overheads of traditional kNN-MT models while maintaining or improving translation performance.\n\n**Problem Addressed:**\nTraditional kNN-MT (Khandelw... | uu1GBD9SlLe | yyBis80iUuU | uu1GBD9SlLe_yyBis80iUuU |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 9