# Feature attribution

6 related articles

Subscribe to updates

arXiv Nov 9, 2026
Survival models stop at a risk score: ODDS-DR rewrites outputs as additive log-odds and sweeps a full risk trajectory in two backward passes
Clinical survival models usually emit distribution parameters, abstract quantities that do not map onto the risk factors clinicians recognize. ODDS-DR reparameterizes the network outputs so they add up in log-odds, letting each output neuron be attributed directly to input features. On the SUPPORT2 and aortic coarctation cohorts its discrimination matches or exceeds DeepSurv, DeepHit and DSM, with a C-index of 0.83 for the log-logistic variant on SUPPORT2.
Clinical survival models usually emit distribution parameters, abstract quantities that do not map onto the risk factors clinicians recognize. ODDS-DR reparameterizes the network outputs so they add up in log-odds, letting each output neuron be attributed directly to input features. On the SUPPORT2 and aortic coarctation cohorts its discrimination matches or exceeds DeepSurv, DeepHit and DSM, with a C-index of 0.83 for the log-logistic variant on SUPPORT2.
Jakob Versnjak（First author）
ODDS-DR Distributional regression Time-to-event outcome prediction Feature attribution Layer-wise Relevance Propagation
arXiv Nov 5, 2026
Polymer property prediction drops chemistry-rule fragmentation: geometry-based fragments at a chosen radius, with roughly 5–10% gains on several properties
Polymer property prediction has long relied on molecular fragments cut by chemical rules, which may suit unfamiliar structures poorly and offer limited control over fragment size. A geometry-based fragmentation method casts fragmentation as a radius-constrained center-selection problem, where the number of fragments follows from a given radius rather than being fixed in advance, and changing the radius yields fragments at different scales. Feeding these multiscale fragments into a multimodal model that combines p-SMILES with 3D conformations lowers prediction error by roughly 5–10% relative to the baseline on several properties of a polymer benchmark, though the 3D modality's contribution varies by property.
Polymer property prediction has long relied on molecular fragments cut by chemical rules, which may suit unfamiliar structures poorly and offer limited control over fragment size. A geometry-based fragmentation method casts fragmentation as a radius-constrained center-selection problem, where the number of fragments follows from a given radius rather than being fixed in advance, and changing the radius yields fragments at different scales. Feeding these multiscale fragments into a multimodal model that combines p-SMILES with 3D conformations lowers prediction error by roughly 5–10% relative to the baseline on several properties of a polymer benchmark, though the 3D modality's contribution varies by property.
Makoto Takamoto（First author）
Geometry-based fragmentation Geometric fragment-based attention Polymer property prediction Feature attribution
arXiv Oct 8, 2026
Attribution maps stop guessing: optimizing insertion and deletion curves directly, Ra-NEM tops average faithfulness across four image models
Attribution maps for explainable AI were previously produced by heuristics, with faithfulness only measured afterward by insertion and deletion curves. Ra-NEM turns the area under those curves into a differentiable objective, using the Gumbel top-k trick for differentiable top-k feature selection inside the neural explanation mask framework. On ImageNet-pretrained ResNet50, ConvNeXt, VGG16 and ViT it reaches the highest average faithfulness and the fastest inference, though it ranks fifth on VGG16.
Attribution maps for explainable AI were previously produced by heuristics, with faithfulness only measured afterward by insertion and deletion curves. Ra-NEM turns the area under those curves into a differentiable objective, using the Gumbel top-k trick for differentiable top-k feature selection inside the neural explanation mask framework. On ImageNet-pretrained ResNet50, ConvNeXt, VGG16 and ViT it reaches the highest average faithfulness and the fastest inference, though it ranks fifth on VGG16.
Björn Möller（First author）
Ra-NEM Feature attribution Insertion and deletion curves
arXiv Oct 5, 2026
MEA uses reward-driven multi-agent optimization to raise explanation faithfulness by up to 34%
The work presents MEA, a multi-agent framework in which a Proposer agent selects and configures explanation tools based on the question and modality while an Actor agent is optimized end-to-end against faithfulness to turn outputs into natural-language explanations; the authors also introduce question types spanning feature attribution, counterfactual reasoning, and spurious feature detection, each paired with a perturbation-based faithfulness metric, find that frontier LLMs systematically produce unfaithful explanations, and report that MEA outperforms post hoc explainers, agentic, and closed-source baselines across six datasets, with faithfulness gains of +28% (tabular), +21% (text), and +34% (vision) over the untrained backbone.
The work presents MEA, a multi-agent framework in which a Proposer agent selects and configures explanation tools based on the question and modality while an Actor agent is optimized end-to-end against faithfulness to turn outputs into natural-language explanations; the authors also introduce question types spanning feature attribution, counterfactual reasoning, and spurious feature detection, each paired with a perturbation-based faithfulness metric, find that frontier LLMs systematically produce unfaithful explanations, and report that MEA outperforms post hoc explainers, agentic, and closed-source baselines across six datasets, with faithfulness gains of +28% (tabular), +21% (text), and +34% (vision) over the untrained backbone.
Vellore Institute of Technology University University of Virginia Indian Institute of Technology Patna
Yuyang Cheng（First author）
MEA Multi-agent system Perturbation-based faithfulness metric Feature attribution Counterfactual Reasoning
arXiv Sep 25, 2026
Nanjing University team releases DSEBench, the first test collection supporting both keyword-plus-example dataset search and field-level explanations
The work formalizes Dataset Search with Examples (DSE) and its explainable extension (ExDSE), builds DSEBench as the first test collection with both dataset-level and field-level human annotations (141 test cases, 7,415 triples, plus 5,699 training cases and 122,585 triples), uses a large language model to generate large-scale training annotations, and adapts and evaluates a range of retrieval, reranking, and explanation methods to establish baselines.
The work formalizes Dataset Search with Examples (DSE) and its explainable extension (ExDSE), builds DSEBench as the first test collection with both dataset-level and field-level human annotations (141 test cases, 7,415 triples, plus 5,699 training cases and 122,585 triples), uses a large language model to generate large-scale training annotations, and adapts and evaluates a range of retrieval, reranking, and explanation methods to establish baselines.
Nanjing University Nanjing University of Chinese Medicine
Qing Shi（First author）Gong Cheng（Corresponding author）
DSEBench Dataset Search with Examples Explainable Dataset Search Feature attribution
Scientific Reports Sep 14, 2026
Scaling vision models does not consistently improve localisation-based explanation quality
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
University of Antwerp University of Warsaw
Mateusz Cedro（First author）Marcin Chlebus（Corresponding author）
Dual-Polarity Precision Explainable AI Feature attribution Saliency Maps Grad-CAM

You’re all caught up

Survival models stop at a risk score: ODDS-DR rewrites outputs as additive log-odds and sweeps a full risk trajectory in two backward passes

Polymer property prediction drops chemistry-rule fragmentation: geometry-based fragments at a chosen radius, with roughly 5–10% gains on several properties

Attribution maps stop guessing: optimizing insertion and deletion curves directly, Ra-NEM tops average faithfulness across four image models

MEA uses reward-driven multi-agent optimization to raise explanation faithfulness by up to 34%

Nanjing University team releases DSEBench, the first test collection supporting both keyword-plus-example dataset search and field-level explanations

Scaling vision models does not consistently improve localisation-based explanation quality