Datasets:
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
id: string
text: string
label: int64
generator: string
domain: string
attack: string
prompt_id: string
attack_intensity: string
attack_n: double
attack_seed_strict: int64
to
{'id': Value('string'), 'text': Value('string'), 'label': Value('int64'), 'generator': Value('string'), 'domain': Value('string'), 'attack': Value('string'), 'prompt_id': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1779, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 299, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2321, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2249, in cast_table_to_schema
raise CastError(
datasets.table.CastError: Couldn't cast
id: string
text: string
label: int64
generator: string
domain: string
attack: string
prompt_id: string
attack_intensity: string
attack_n: double
attack_seed_strict: int64
to
{'id': Value('string'), 'text': Value('string'), 'label': Value('int64'), 'generator': Value('string'), 'domain': Value('string'), 'attack': Value('string'), 'prompt_id': Value('string')}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1347, in compute_config_parquet_and_info_response
parquet_operations = convert_to_parquet(builder)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 980, in convert_to_parquet
builder.download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 882, in download_and_prepare
self._download_and_prepare(
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 943, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1646, in _prepare_split
for job_id, done, content in self._prepare_split_single(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1832, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
id string | text string | label int64 | generator string | domain string | attack string | prompt_id string |
|---|---|---|---|---|---|---|
meldx/gpt-5.4-mini/d1f9fac71d61 | Background: Accurate marking of tumors and organs is a critical step in image-guided diagnosis and therapy, yet it remains highly challenging because of substantial inter- and intra-observer variability. Differences in clinical experience, image interpretation, and local anatomy can lead to inconsistent delineations, a... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0006/8cc907e393018570 |
meldx/gpt-5.4-mini/60159fc4f38e | The .uk domain provides a longitudinal record of the United Kingdom’s web presence and offers a unique basis for examining the growth, structure, and geographic distribution of national cyberspace. This study maps the national UK web presence from 1996 to 2010 by analyzing the .uk domain at multiple time points, with a... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0007/8d050eb93767f826 |
meldx/gpt-5.4-mini/b7c91fc8934a | Multi-target tracking with superpositional measurements arises in sensor modalities where individual target contributions are not separately resolved and the received signal depends on the aggregate effect of all targets. This setting is challenging because the measurement process is inherently nonlinear and does not p... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0009/b14096bed95b1425 |
meldx/gpt-5.4-mini/7ef7e3b2e881 | Piecewise-testable languages form an important subclass of regular languages characterized by finite Boolean combinations of conditions on scattered subwords. A central complexity parameter for such languages is their height, defined as the maximum length of the subwords needed to specify the language via inclusion and... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0000/8ea66b490f204aca |
meldx/gpt-5.4-mini/f401199e390b | Biochemical receptors in living cells and emerging diagnostic devices must detect extremely small changes in signaling molecule concentration while operating in noisy, dynamic environments. Achieving such sensitivity requires understanding how receptor architecture, ligand binding kinetics, and downstream signal transd... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0001/f5a3cedbaa5bd66e |
meldx/gpt-5.4-mini/562ac5135e43 | Most progress in semantic segmentation has been reported on daytime images acquired under favorable illumination, where high contrast and stable color cues simplify dense prediction. However, real-world deployment often requires robust performance across adverse conditions, including low light, shadows, and illuminatio... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0002/7283cb0e48d1d586 |
meldx/gpt-5.4-mini/a5067f1388d7 | Background: Anatomical and biophysical modeling of the left atrium (LA) and proximal pulmonary veins (PPVs) is increasingly important for understanding atrial electrophysiology and improving the clinical management of atrial fibrillation and other cardiac diseases. Patient-specific models can support risk stratificatio... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0010/bc9a887353e9d3f0 |
meldx/gpt-5.4-mini/e174885d8aa9 | Magnetically ordered insulators provide a paradigmatic setting for studying the interplay between exchange interactions, anisotropy, and collective spin excitations in systems where charge degrees of freedom are localized. In this report, we analyze a simple effective model for describing such materials, with the goal ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0011/26ae88345402b695 |
meldx/gpt-5.4-mini/c020dc9ab8f2 | We investigate a non-isothermal diffuse-interface model for two-phase incompressible flows in which thermocapillary stresses generate a Marangoni effect at the fluid interface. Such flows arise in material processing, coating, and microfluidic systems, where temperature gradients interact with interfacial tension to pr... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0004/85f3082ef4baee08 |
meldx/gpt-5.4-mini/870c6b404b38 | Domain adaptation has become an important strategy for medical image segmentation, where models trained on source-domain data often perform poorly when deployed on target-domain data acquired from different scanners, protocols, institutions, or patient populations. This distribution shift limits the clinical reliabilit... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0014/6b41f6bc7a6387d0 |
meldx/gpt-5.4-mini/8daf6b13ca4d | We introduce grafted hypersequents, a new Gentzen-style proof-theoretic framework that combines the structural flexibility of nested sequents with the componentwise expressiveness of hypersequents. The motivation is to obtain a uniform calculus in which local tree-like structure and global inter-component interaction c... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0016/f81c0bf83b32ddcc |
meldx/gpt-5.4-mini/1dc069e7d085 | Image-to-image (I2I) translation has become a central problem in computer vision, enabling the conversion of images from one domain to another while preserving content and transferring target-specific attributes. Recent advances in generative modeling, particularly adversarial training, have substantially accelerated p... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0003/6265574b25e2775a |
meldx/gpt-5.4-mini/5d997c2b2e5e | Many computer vision tasks require outputs to change predictably under image rotations, yet standard convolutional networks often learn only approximate rotational behavior from data. This mismatch can reduce sample efficiency, degrade generalization, and produce inconsistent predictions under test-time transformations... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0017/7af13d9e34627a78 |
meldx/gpt-5.4-mini/d5d55c6f787a | Image segmentation remains a fundamental yet challenging problem in computer vision, where accurate delineation of objects and scenes must be achieved under constraints of computational efficiency and annotation quality. Existing approaches often trade off precision for speed or rely on large amounts of supervised data... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0018/6a0a381d1a9d6044 |
meldx/gpt-5.4-mini/df6fc8de7769 | Semantic segmentation with encoder-decoder architectures has advanced rapidly, yet further gains in quality often require substantially deeper networks, higher memory consumption, or complex multi-branch designs. In this work, we investigate whether single encoder-decoder methodologies are approaching a practical upper... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0008/209991141dea46e3 |
meldx/gpt-5.4-mini/c01428369b02 | Deep neural networks have achieved remarkable performance across a range of image restoration and synthesis tasks, including image segmentation, super-resolution, coloration, and inpainting. These successes are driven by their ability to learn hierarchical feature representations directly from data, reducing reliance o... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0005/95d43c50e1208a9b |
meldx/gpt-5.4-mini/a2a456baf080 | We study the incompressible flow of a Newtonian fluid in a three-dimensional domain rotating uniformly about a vertical axis, subject to a vertically invariant horizontal body force. This configuration is relevant to rapidly rotating geophysical and engineering flows, where strong Coriolis effects induce anisotropy and... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0019/5b56924ec9c473f8 |
meldx/gpt-5.4-mini/62fcdaf53438 | Image segmentation has been advanced substantially by deep convolutional neural networks that learn dense image representations in which color, shape, and texture cues are jointly processed. Despite their strong performance, these methods often rely on increasingly complex architectures and large-scale supervision, mak... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0012/ffbb438c75f63d05 |
meldx/gpt-5.4-mini/41e55e977c0c | Ferrograph image segmentation is a critical step for extracting quantitative features from wear particles and enabling reliable condition monitoring of lubricated machinery. However, ferrograph images often exhibit uneven illumination, overlapping particles, staining artifacts, and blurred boundaries, making accurate s... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0020/4de482d104b0fd6d |
meldx/gpt-5.4-mini/9519c0b1c040 | Transfer learning has become a central strategy for training machine learning models, particularly when labeled data are limited, expensive to obtain, or drawn from specialized domains. By reusing representations learned from large source datasets, transfer learning can improve convergence, reduce sample complexity, an... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0024/10ec76b621380a69 |
meldx/gpt-5.4-mini/1a3dd50536dd | Many mathematical and computational models encounter singular limits, divergent series, unbounded domains, or iterated processes whose natural description extends to infinity. In such settings, direct analysis or numerical simulation beyond the limiting point is typically ill-posed, unstable, or computationally infeasi... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0026/6d1f34a192bbba4c |
meldx/gpt-5.4-mini/dbf072745a8d | Convolutional neural networks (CNNs) have become a dominant approach in computer vision because they can learn hierarchical feature representations directly from data and have achieved strong performance across tasks such as image classification and image segmentation. Despite their success, model performance remains s... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0027/211b09083e172b2f |
meldx/gpt-5.4-mini/531dfd5b0a55 | Background: Deep learning has achieved outstanding performance across a wide range of medical imaging and clinical prediction tasks, yet most state-of-the-art models depend on large annotated datasets that are costly, time-consuming, and often infeasible to obtain in clinical and health care settings. This limitation i... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0015/68328aa69ea18f6b |
meldx/gpt-5.4-mini/873c77e6ab91 | Deep learning has achieved remarkable success in medical image segmentation, but its performance typically depends on large datasets annotated with fine-grained pixel-wise masks. Because expert delineation of such masks is time-consuming, expensive, and prone to inter-observer variability, reducing annotation burden wh... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0025/9502ebbe4bd57b18 |
meldx/gpt-5.4-mini/d42e58de2268 | Accurate land use and land cover (LULC) mapping from high-resolution satellite imagery is essential for urban planning, environmental monitoring, and resource management. Conventional pixel-based and object-based approaches often struggle to capture complex spatial patterns and contextual information in heterogeneous e... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0029/286599657c41c8b4 |
meldx/gpt-5.4-mini/9e1bc7b76234 | Deep learning models have achieved remarkable performance across a wide range of prediction tasks, yet their reliability often degrades when deployed in environments that differ from the data used for training. This vulnerability, known as domain shift, arises from changes in acquisition conditions, populations, featur... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0013/4a4d58354c1767cf |
meldx/gpt-5.4-mini/a05db5dc6eba | We study a problem posed by Po-Shen Loh that is equivalent to a natural Ramsey-theoretic question, and we establish several new results toward its resolution. The problem asks for quantitative and structural guarantees on monochromatic configurations in edge-colored complete graphs under additional constraints, placing... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0032/3add674b2e9244a6 |
meldx/gpt-5.4-mini/a0747b695b59 | Few-shot segmentation aims to delineate novel object classes from only a handful of annotated examples, yet it remains challenging due to severe data scarcity, large intra-class variation, and ambiguity in class appearance across scenes. In this work, we address the challenging task of few-shot segmentation by introduc... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0021/81e1b95b93f10494 |
meldx/gpt-5.4-mini/0959f4f35088 | We study the asymptotic behavior of the nonparametric maximum likelihood estimator of a one-dimensional log-concave density, where the true density has the form f0 = exp(φ0) with φ0 concave on R. Log-concave density estimation provides a fully automatic shape-constrained alternative to classical smoothing methods, but ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0034/a9c2ef16858307be |
meldx/gpt-5.4-mini/b91fc46943a2 | Electron microscopic connectomics aims to reconstruct comprehensive brain connectivity maps at synaptic resolution by combining high-throughput nano-scale imaging with computational analysis. This approach promises unprecedented insight into circuit organization, yet remains limited by the enormous data volumes, imagin... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0035/a5916232a0d77a64 |
meldx/gpt-5.4-mini/f59cd812d109 | Background: The Jaccard index, also known as the intersection-over-union (IoU) score, is widely used to evaluate image segmentation because it provides an intuitively meaningful measure of spatial overlap, is scale invariant, and is less sensitive to class imbalance than pixelwise accuracy. Despite its popularity, Jacc... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0036/0a05bc103fcd06d1 |
meldx/gpt-5.4-mini/2573c7dd1a7f | Neural networks have emerged as a powerful family of methods for analyzing physiological time-series, driven by the growing availability of high-resolution biosignals and the limitations of traditional feature-engineering approaches. However, their practical performance across heterogeneous clinical signals and varying... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0022/fb676b8cc6c6f47c |
meldx/gpt-5.4-mini/42d3a69c598a | Semantic image segmentation is a fundamental task in medical image analysis, enabling precise delineation of anatomical structures and pathological regions for diagnosis, treatment planning, and quantitative assessment. Despite major progress with deep learning, medical segmentation remains challenging due to limited l... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0030/fa47706531ba9d29 |
meldx/gpt-5.4-mini/94e97ca75fb1 | Discrete-time quantum walks (DTQWs) have become a central model for studying coherent transport, quantum algorithms, and asymptotic limit phenomena. A key feature of DTQWs is that, under suitable scaling, their position distributions often converge to non-Gaussian limit laws, in contrast to the classical central limit ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0031/6bac76488727f623 |
meldx/gpt-5.4-mini/44ac339aad5a | Games such as go, chess, and checkers exhibit extensive state-space redundancy, in which multiple board configurations correspond to the same underlying game position because of symmetries, move-order equivalences, or reversible transformations. This multiplicity can inflate search, learning, and storage costs in game-... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0028/74b7f6a5afa1a3a9 |
meldx/gpt-5.4-mini/046ace16c919 | Vision Transformers (ViT) have emerged as a powerful alternative to convolutional neural networks for a broad range of vision tasks, including image classification, object detection, and semantic segmentation. Unlike convolutional architectures, ViTs model global dependencies directly through self-attention, enabling f... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0023/8ae68d6fcf2ad059 |
meldx/gpt-5.4-mini/59e4aadf0701 | Image segmentation is a fundamental problem in computer vision that partitions an image into semantically meaningful regions and underpins a wide range of automated image processing applications, including medical analysis, remote sensing, autonomous navigation, and industrial inspection. Despite substantial progress, ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0037/29d1cf28cfa025d3 |
meldx/gpt-5.4-mini/c5463c9a6322 | Background: Segmentation of 3D biomedical images is a fundamental task in image analysis, enabling quantitative characterization of anatomical structures, disease biomarkers, and treatment response. However, accurate segmentation remains challenging due to low contrast boundaries, image noise, intensity inhomogeneity, ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0033/438d3aad3f140f19 |
meldx/gpt-5.4-mini/2fd9af3716a6 | Few-shot semantic segmentation (FSS) offers a promising framework for medical image analysis, where dense pixel-level annotations are expensive and expert labeling is often limited. This study investigates the applicability of FSS to medical imaging and proposes a task-adaptive segmentation framework designed to improv... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0039/5b41bc3947d7aaf2 |
meldx/gpt-5.4-mini/1b0830ec7413 | Direction-of-arrival (DOA) estimation is a fundamental problem in array signal processing for localizing multiple acoustic or electromagnetic sources from a limited number of sensor measurements. Conventional high-resolution methods often rely on dense arrays, accurate covariance estimates, and prior knowledge of sourc... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0040/c101672c606c8d7e |
meldx/gpt-5.4-mini/345e3731c2fc | Background: Automatic parsing of anatomical objects in X-ray images is essential for a range of clinical applications, including image-guided intervention, procedure planning, and workflow automation. However, X-ray interpretation remains challenging because of low soft-tissue contrast, object overlap, variable acquisi... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0045/74c93452e2e0e2d2 |
meldx/gpt-5.4-mini/2ea5e8ca08b0 | Adversarial attacks are typically crafted in a task-specific manner, with perturbation strategies tailored separately for image classification, object detection, semantic segmentation, or other downstream objectives. This specialization limits transferability and complicates the development of unified defenses, motivat... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0043/c258fdd81dc98718 |
meldx/gpt-5.4-mini/a61373e317d8 | The direct conversion of ambient thermal radiation into electrical work at a uniform temperature remains a longstanding challenge because equilibrium systems are constrained by detailed balance and the second law of thermodynamics. Here we investigate whether the photoelectric effect driven by blackbody radiation can p... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0041/65004f5fe2070202 |
meldx/gpt-5.4-mini/0f346e4bfbb3 | Deep neural networks have achieved remarkable performance in medical image segmentation; however, their success typically depends on large-scale, expert-annotated datasets, which are costly and time-consuming to acquire. This limitation is especially pronounced in clinical settings where data scarcity, label variabilit... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0047/dde92ed8fffb4ea5 |
meldx/gpt-5.4-mini/851e6b1bcdcf | Arbitrary style transfer aims to synthesize a novel image that preserves the semantic content of a source image while adopting the appearance characteristics of a separate style image. This problem has attracted substantial interest because it enables flexible image editing without retraining for each new style. Howeve... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0042/3e68367ef781b093 |
meldx/gpt-5.4-mini/ca21b1da7b98 | Generative adversarial networks (GANs) have emerged as a powerful class of deep generative models, achieving impressive results in tasks such as image synthesis, super-resolution, and inpainting. Despite their success, training instability, limited diversity, and weak structural consistency remain persistent challenges... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0038/af3b5e01f9e8709a |
meldx/gpt-5.4-mini/d81a6cd94c21 | Accurate segmentation of neuron membranes in 2D electron microscopy (EM) imagery is essential for reconstructing neural circuits and quantifying cellular ultrastructure, yet remains challenging due to low contrast, heterogeneous texture, and frequent membrane discontinuities. We present a method for automated membrane ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0046/ce060fdae79e1d6b |
meldx/gpt-5.4-mini/405cb7492a14 | Multiclass semi-supervised classification on graphs has become a powerful framework for learning from limited labels by combining smoothness priors with the discrete structure of relational data. Diffuse interface methods, originally developed for image segmentation and later adapted to graph-based learning, offer an e... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0048/aeb2fe25a7f929d3 |
meldx/gpt-5.4-mini/066412391338 | The search for neutrinoless double-beta decay (0νββ) demands exceptionally low-background detectors and a quantitative understanding of how residual backgrounds populate the region of interest around the decay Q-value. We present a comprehensive background study motivated by next-generation 0νββ experiments, combining ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0049/76477949f61189a0 |
meldx/gpt-5.4-mini/ac284f7e9082 | Background: The rapid expansion of the literature has made manual screening of scientific papers increasingly difficult, motivating automated methods for identifying relevant studies from bibliographic records. This work investigates whether lightweight natural language processing and machine learning can improve abstr... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0044/cf4e0d67f7471d57 |
meldx/gpt-5.4-mini/1db9a0f9b1c1 | The Hat Problem is a classical problem in combinatorial game theory and distributed reasoning, in which players must infer hidden information from limited observations and a prescribed strategy. Ebert’s Hat Problem is a particularly well-studied version involving a small number of players and binary colors, where the g... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0051/f3a99dd3023a45ba |
meldx/gpt-5.4-mini/a6de05d5969d | Poisson image denoising remains challenging in photon-limited imaging because the noise variance depends on the underlying signal, making conventional algorithms less effective at high noise levels. Traditional methods, including variance-stabilizing transforms and model-based regularization, often provide competitive ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0052/09f870ea8231bbaa |
meldx/gpt-5.4-mini/9d93fe5052be | Intensity-fluctuation spectroscopy is a sensitive tool for probing dynamical processes in optical media, but its practical implementation is often limited by extraneous noise, particularly detector shot noise and technical intensity fluctuations. We present a measurement system designed to obtain accurate power spectra... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0055/b273e4e6ba99be5a |
meldx/gpt-5.4-mini/7287925baae6 | Background: Deep learning has emerged as a powerful approach for medical image segmentation, achieving high accuracy when trained on large, representative datasets with expert annotations. However, in many clinical settings, annotated data are limited, heterogeneous, and costly to obtain, which can restrict model perfo... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0057/a0f03aac5a028bb7 |
meldx/gpt-5.4-mini/5952f7400b05 | We construct the U(1) gauge theory on a spatial manifold endowed with Lie-type noncommutativity, where coordinate commutators close linearly on the coordinates and the deformation is controlled by a non-Abelian Lie algebra. Such spaces provide a natural generalization of canonical noncommutative geometry and are releva... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0053/0bcb254ae1787dca |
meldx/gpt-5.4-mini/b01733a902fc | Classical supervised learning methods have achieved strong performance across many pattern recognition tasks, but their practical deployment is often limited by the need for large annotated datasets and by poor transferability to unseen data distributions. These limitations are especially pronounced in real-world setti... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0056/582f404fcda6d56c |
meldx/gpt-5.4-mini/3fd5373330b4 | Shape analysis is a fundamental problem in computer vision and geometric learning, with applications in object recognition, retrieval, segmentation, and 3D modeling. However, many existing deep architectures are designed primarily for Euclidean grid data and do not directly exploit the intrinsic structure of shapes. In... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0050/9c3db86cc5aac387 |
meldx/gpt-5.4-mini/ca938c05e20c | Background: Federated learning has emerged as a promising framework for collaborative model development across distributed medical institutions while preserving patient privacy and institutional data sovereignty. By enabling local training and selective sharing of model updates rather than raw data, federated approache... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0054/b025ac4f27183e06 |
meldx/gpt-5.4-mini/9def6ce4cbe8 | Automated segmentation in medical image analysis remains a challenging problem because accurate delineation of anatomical structures or lesions often requires large quantities of manually labeled data, which are costly and time-consuming to obtain. This limitation is especially pronounced in clinical settings where exp... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0058/e27e6543d56888e2 |
meldx/gpt-5.4-mini/f45eae8cfd99 | Efficient phase-matched resonant four-wave mixing (FWM) is central to frequency conversion, quantum photonics, and integrated nonlinear optics, yet it typically requires precise dispersion control over extended propagation lengths. Here we propose and demonstrate localized mode coupling as a viable dispersion engineeri... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0059/e62d7f6f2d819ef7 |
meldx/gpt-5.4-mini/7ed2bd00c552 | We establish the first positive solvability results for boundary value problems in the upper half-space for second-order parabolic systems under minimal structural assumptions on the coefficients. Specifically, we consider divergence-form systems whose coefficients are merely measurable in the time and normal direction... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0065/4ace4bbd9129fd6b |
meldx/gpt-5.4-mini/18499943f167 | We study purely loxodromic free Kleinian groups generated by two non-commuting isometries of hyperbolic 3-space. Let $\xi$ and $\eta$ be isometries of $\mathbb{H}^3$ such that $\Gamma=\langle \xi,\eta\rangle$ is free and contains no parabolic or elliptic elements. The geometry of such groups is closely tied to the disp... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0063/f1765b326fdb5035 |
meldx/gpt-5.4-mini/01c08e5e7a47 | Query evaluation on probabilistic databases is generally intractable and, for many natural classes of queries, #P-hard. This complexity barrier limits the practical use of probabilistic data management in applications that require exact uncertainty-aware answers. In this work, we study the computational landscape of qu... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0061/863da72dcd348ffd |
meldx/gpt-5.4-mini/96342318cd9b | Background: Difference-in-differences (DID) is a widely used quasi-experimental approach for estimating treatment effects when randomized assignment is not feasible. By comparing changes in outcomes over time between treated and untreated groups, DID can remove time-invariant differences between groups and common tempo... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0064/34016efb14478746 |
meldx/gpt-5.4-mini/54a8b36df17e | Object-centric generative models (OCGMs) have emerged as a powerful paradigm for learning structured scene representations from unlabeled visual data, enabling unsupervised object segmentation and interpretable scene generation. Despite substantial progress, existing methods vary widely in how they represent objects, i... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0062/e89fcd86649fb775 |
meldx/gpt-5.4-mini/ab3bd66105db | Deep convolutional neural networks (DCNNs) have emerged as a dominant approach for high-level vision tasks, achieving state-of-the-art performance in image classification, object detection, and related recognition problems. Their success is largely attributed to hierarchical feature learning, which enables the automati... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0060/056371671f296390 |
meldx/gpt-5.4-mini/f9ebf7f9663b | The Riemann zeta function, introduced by Riemann in 1859, lies at the center of analytic number theory through its deep connection with the distribution of prime numbers. This paper briefly reviews major accomplishments in the study of the zeta function and the Riemann hypothesis from Riemann’s original work to the pre... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0066/3f3347d289efcf4d |
meldx/gpt-5.4-mini/27cebf1fef3c | Pixel-wise segmentation remains one of the most data- and annotation-intensive tasks in computer vision, largely because accurate delineation of object boundaries requires dense expert supervision. This requirement limits scalability, increases labeling cost, and hinders deployment in domains where annotated data are s... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0068/9ad83b0abe4368d2 |
meldx/gpt-5.4-mini/3d4ded2bc14e | Medical image segmentation is a foundational component of computer-assisted diagnosis and therapy, enabling lesion delineation, organ quantification, treatment planning, and longitudinal disease monitoring. Despite substantial progress with deep learning, accurate segmentation remains challenging because of limited ann... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0067/4d18cd550a704b18 |
meldx/gpt-5.4-mini/6ae1e3fce7b2 | We investigate weighted Sobolev spaces on metric measure spaces \((X,d,m)\), motivated by the need to extend variational and analytic tools from smooth settings to spaces with potentially singular geometry and non-Euclidean measure structure. Weighted Sobolev frameworks are particularly relevant when the underlying mea... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0069/4f2755017cb98376 |
meldx/gpt-5.4-mini/28f22ff9a930 | Medical image segmentation is a fundamental step in medical image analysis, enabling the delineation of anatomical structures and pathological regions for diagnosis, treatment planning, and quantitative assessment. However, accurate segmentation remains challenging due to low contrast, noise, shape variability, and lim... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0071/5eca09a8fdf54d34 |
meldx/gpt-5.4-mini/b0e19d48b329 | We study relative Chow and relative K-stability for polarized toric manifolds in the toric setting, with respect to the action of the maximal torus and its extremal affine functions. Motivated by the link between algebro-geometric stability and canonical metrics, we compare these two notions of stability through the co... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0070/d2b1626111592109 |
meldx/gpt-5.4-mini/a0c954537c01 | Background: Accurate segmentation is a foundational step in dental image analysis, enabling diagnosis, treatment planning, implant placement, orthodontic assessment, and quantitative evaluation of oral structures. Dental imaging modalities such as cone-beam computed tomography (CBCT), panoramic radiography, intraoral r... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0072/4fa7a62c3e0492a0 |
meldx/gpt-5.4-mini/5ae6cfec6337 | Background subtraction is a classical and widely used approach for detecting moving objects in video sequences, particularly in surveillance and traffic monitoring applications such as vehicles traveling on freeways. The method relies on estimating a representation of the static scene and identifying pixels or regions ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0073/b3da2b8b03895513 |
meldx/gpt-5.4-mini/049cb7ef9c30 | We study a one-dimensional reaction-diffusion equation with a bistable reaction term and a nonlocal space-fractional diffusion operator of Riesz-Feller type. Such models arise in anomalous transport, interface propagation, and pattern formation when dispersal is asymmetric and long-ranged. The combination of bistabilit... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0074/f96293608f2438ee |
meldx/gpt-5.4-mini/f5465f7525b8 | Video-based eye tracking has become an important noninvasive method for quantifying gaze behavior across psychology, neuroscience, human-computer interaction, clinical research, education, and marketing. Unlike head-mounted or scleral techniques, video-based systems can capture eye movements in naturalistic settings wi... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0075/52b06690f79ccbeb |
meldx/gpt-5.4-mini/4eb29114d2a1 | Information geometry provides a differential-geometric framework for families of probability distributions, but its classical finite-dimensional formulation does not directly extend to statistical models of arbitrary complexity. In particular, when the space of probability densities is infinite-dimensional, standard ta... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0078/e948e5f53595969f |
meldx/gpt-5.4-mini/ef048f33ce5d | Isolated gravitating systems in asymptotically flat spacetimes are naturally described at null infinity by the Bondi-Metzner-Sachs (BMS) symmetry group, whose associated conserved charges encode the radiative and Coulombic content of the spacetime. These charges provide a covariant framework for characterizing mass, mo... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0077/5215911e65e9be7c |
meldx/gpt-5.4-mini/7ce3aef2d0f0 | Chromospheric jets are ubiquitous dynamic structures that provide important diagnostics of mass and energy transport in the solar atmosphere. However, their transverse motions and recurrence properties remain insufficiently understood. In this study, we investigate transverse oscillatory motions and recurrent behavior ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0079/ce3ced1781bccebc |
meldx/gpt-5.4-mini/5fcd68999352 | Accurate medical image segmentation remains a foundational task in quantitative imaging, surgical planning, and treatment response assessment, yet contemporary deep learning-based methods continue to depend on large volumes of pixel-wise annotations that require hours of labor from domain experts. This annotation burde... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0081/c3081e37476ff40f |
meldx/gpt-5.4-mini/7246be6764fb | Laser-induced magnetization dynamics in ultrathin ferromagnetic heterostructures are of interest for understanding ultrafast spin–lattice energy transfer and for developing optically driven magnetic devices. Here we investigate the laser-induced magnetization precession of a thick Pt/Co/Pt film exhibiting perpendicular... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0080/547faf82e614a3b1 |
meldx/gpt-5.4-mini/43be7b605b19 | Convolutional neural networks (CNNs) have become the dominant paradigm in modern computer vision because of their strong ability to learn hierarchical feature representations directly from image data. Despite their success, understanding how architectural design choices influence feature extraction and downstream perfo... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0082/9e3245f4e8d9f820 |
meldx/gpt-5.4-mini/33f9d4ed2b96 | Accurate evaluation of medical image segmentation is essential for comparing algorithms and supporting clinical translation. The Dice similarity coefficient and Jaccard index are among the most widely used overlap-based metrics for this purpose, yet their mathematical relationship, sensitivity to object size, and behav... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0076/a5e4102240f29ebf |
meldx/gpt-5.4-mini/80a20469d3ba | We determine the Seiberg-Witten geometry of four-dimensional mass-deformed $\mathcal N=2$ superconformal quiver gauge theories of ADE type. These theories arise from conformal quiver constructions whose gauge and matter content are encoded by simply laced Dynkin diagrams, and their Coulomb-branch dynamics are expected ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0083/cad6d4e3b24a6aff |
meldx/gpt-5.4-mini/bb094fba9c0a | Nonlinear interferometry is a powerful tool for enhancing phase sensitivity and probing weak perturbations, yet most implementations rely on explicit beam splitting and recombination. Here we investigate an alternative interferometric architecture in which interference arises through diffraction from a nonlinear medium... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0085/221f6fb1e26c5606 |
meldx/gpt-5.4-mini/ef2975bec6d1 | Accurate and efficient image segmentation remains a central challenge in medical and scientific image analysis, particularly for complex 2D structures and volumetric datasets. The Live-Wire method, which formulates segmentation as an interactive shortest-path problem on image-derived cost maps, offers a promising frame... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0084/1c56fe9e2f3da63a |
meldx/gpt-5.4-mini/6f2927ac4782 | The unknotting number is a fundamental measure of knot complexity, defined as the minimum number of crossing changes required to transform a knot into the unknot. Despite its simple definition, determining the unknotting number remains a difficult problem in low-dimensional topology, with exact values known for only a ... | 1 | gpt-5.4-mini | abstracts | none | meldx/abstracts/0086/3854642bc237a4cf |
End of preview.
MELD-eval
Held-out evaluation pool for AI-text detectors.
Paired across four current-generation chat models, eight RAID-style English domains, and six surface attacks, plus the matched human seeds each AI row was conditioned on.
At a glance
| Quantity | Count |
|---|---|
| Total rows | 227,998 |
| Paired human seeds | 7,862 |
| Clean AI rows | 31,448 |
| Attacked AI rows | 188,688 |
| Generators | 4 |
| Domains | 8 |
Attack types (incl. none) |
7 |
Generators: gpt-5.4-mini, gemini-3-flash, claude-haiku,
qwen-3.6-plus.
Domains: books, news, abstracts, recipes, reddit,
reviews, wiki, poetry.
Attacks: zero_width_space, homoglyph, whitespace, synonym,
upper_lower, number, plus none for clean rows.
Files
meld_eval.jsonl— one JSON object per line.
Schema
{
"id": "meldx/gpt-5.4-mini/d1f9fac71d61",
"text": "Background: Accurate marking of tumors and organs ...",
"label": 1,
"generator": "gpt-5.4-mini",
"domain": "abstracts",
"attack": "none",
"prompt_id": "meldx/abstracts/0006/8cc907e393018570"
}
| Field | Type | Description |
|---|---|---|
id |
string | Globally unique row id. |
text |
string | The full document. |
label |
int | 1 if AI-generated, 0 if human. |
generator |
string | One of the four generators above, or "human". |
domain |
string | One of the eight RAID-style domains. |
attack |
string | One of the six attack types, or "none". |
prompt_id |
string | Pairing key: every AI row shares its prompt_id with one human seed. |
Loading
import json
rows = [json.loads(l) for l in open("meld_eval.jsonl")]
Or with the datasets library:
from datasets import load_dataset
ds = load_dataset("<this-repo>", data_files="meld_eval.jsonl", split="train")
Intended use
Zero-shot generalization benchmark for AI-text detectors against current-generation chat models, with per-attack and per-generator breakdowns. The benchmark is not a training corpus.
License
CC-BY-4.0 for annotations and metadata. The text field combines (a)
human passages from publicly released prior corpora (notably RAID) under
their respective licenses, and (b) machine completions from commercial
chat APIs in 2026 under each provider's terms of service. Users are
responsible for upstream license compliance.
Citation
Citation information will be added in a future release.
- Downloads last month
- 33