Pith. sign in

REVIEW 29 cited by

Captum: A unified and generic model interpretability library for PyTorch

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2009.07896 v1 pith:ZZF64T4O submitted 2020-09-16 cs.LG cs.AIstat.ML

Narine Kokhlikyan , Vivek Miglani , Miguel Martin , Edward Wang , Bilal Alsallakh , Jonathan Reynolds , Alexander Melnikov , Natalia Kliushkina
show 3 more authors
Carlos Araya Siqi Yan Orion Reblitz-Richardson
This is my paper · ORCID
classification cs.LGcs.AIstat.ML
keywords libraryalgorithmscaptummodelallowsattributionbuiltextensibility
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

In this paper we introduce a novel, unified, open-source model interpretability library for PyTorch [12]. The library contains generic implementations of a number of gradient and perturbation-based attribution algorithms, also known as feature, neuron and layer importance algorithms, as well as a set of evaluation metrics for these algorithms. It can be used for both classification and non-classification models including graph-structured models built on Neural Networks (NN). In this paper we give a high-level overview of supported attribution algorithms and show how to perform memory-efficient and scalable computations. We emphasize that the three main characteristics of the library are multimodality, extensibility and ease of use. Multimodality supports different modality of inputs such as image, text, audio or video. Extensibility allows adding new algorithms and features. The library is also designed for easy understanding and use. Besides, we also introduce an interactive visualization tool called Captum Insights that is built on top of Captum library and allows sample-based model debugging and visualization using feature importance metrics.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 29 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 643 citations worldwide. Full citation record

  1. What is Missing? Explaining Neurons Activated by Absent Concepts

    cs.CV 2026-03 accept novelty 7.0 of 10

    DNNs encode concept absences that raise neuron activation; standard XAI misses them, but non-target attribution and minimization visualization uncover them and improve debiasing.

  2. Does Explainability Transfer? A Controlled Benchmark of Attribution Methods on Vision Transformers and CNNs

    cs.CV 2026-08 conditional novelty 6.0 of 10

    Attribution-method rankings are architecture-dependent: CNN conclusions do not reliably transfer to Vision Transformers, and apparent CAM localization wins under bounding boxes are largely metric saturation.

  3. Token-Level Diagnosis of Sycophancy in LLMs with Attribution-Guided Steering

    cs.CL 2026-07 reject novelty 6.0 of 10

    A token-attribution metric (ASI) and attribution-guided steering are proposed for LLM sycophancy, with large in-sample reductions reported.

  4. Scaling Time Series Classification via XAI-Driven Data Reduction

    cs.LG 2026-07 conditional novelty 6.0 of 10

    drXAI uses XAI attributions from a fast classifier to choose important channels/time points, achieving 80–90% data reduction with comparable classification accuracy.

  5. Imputation-free transformer learning enables robust Alzheimer's disease prediction and calibrated uncertainty quantification across heterogeneous clinical cohorts

    q-bio.NC 2026-07 conditional novelty 6.0 of 10

    An imputation-free transformer with masked and intersample attention predicts Alzheimer’s status and scores across cohorts with better calibration than tree ensembles.

  6. STAPO: Selective Trajectory-Aware Policy Optimization for LLM Agent Training

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Normalized entropy locates trajectory-neglect steps; STAPO selectively optimizes them with a trajectory-aware reward and trajectory-independent penalty, beating group-based RL baselines on agent tasks.

  7. Two Black Boxes, One Solver: Encoder Probing and Decoder Attribution for Neural Multi-Attribute VRP under Hard-Mask and Recourse Decoders

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Graph encoders and recourse training improve latent constraint representation and actionable decoder explanations in neural MAVRP solvers; make-feasible counterfactuals arise from the training regime, not the mask.

  8. Interpreto: An Explainability Library for Transformers

    cs.CL 2025-12 accept novelty 6.0 of 10

    Interpreto provides a unified, documented Python API for attribution and end-to-end concept-based explanation of HuggingFace classification and generation models.

  9. An Empirical Evaluation of Factors Affecting SHAP Explanation of Time Series Classification

    cs.AI 2025-09 conditional novelty 6.0 of 10

    Equal-length segmentation matches or beats seven more complex segmentation strategies for SHAP-based time series explanations, and length-weighted normalisation of segment attributions improves evaluation scores.

  10. Accept or Deny? Evaluating LLM Fairness and Performance in Loan Approval across Table-to-Text Serialization Approaches

    cs.LG 2025-08 conditional novelty 6.0 of 10

    Serialization format and in-context examples change both accuracy and gender fairness of LLM loan approvals, with finance-tuned models often showing larger disparities.

  11. DeepFaith: A Domain-Free and Model-Agnostic Unified Framework for Highly Faithful Explanations

    cs.LG 2025-08 unverdicted novelty 6.0 of 10

    A single trained explainer that unifies multiple faithfulness metrics and generates model-agnostic explanations with high measured faithfulness.

  12. POLARIS: Explainable Artificial Intelligence for Mitigating Power Side-Channel Leakage

    cs.CR 2025-07 conditional novelty 6.0 of 10

    An XAI-based masking-gate insertion flow is claimed to beat the state-of-the-art VALIANT in leakage reduction, speed, and overhead, but the comparison is weakened by test-set model selection and inference-only timing.

  13. TRACE: Training and Inference-Time Interpretability Analysis for Language Models

    cs.CL 2025-07 conditional novelty 6.0 of 10

    TRACE is a modular training-time interpretability toolkit that, combined with the ABSynth synthetic corpus generator, reveals phased emergence of syntax and semantics during transformer training.

  14. DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values

    cs.LG 2025-07 conditional novelty 6.0 of 10

    DeltaSHAP attributes a monitoring model's prediction change between time steps to individual features using sampled Shapley values over the latest observed measurements.

  15. Model Fusion via Retrofitting

    cs.LG 2025-06 conditional novelty 6.0 of 10

    A neuron-centric fusion method that clusters intermediate activations of independently trained models into importance-weighted centroids and fits the fused network to them, outperforming baselines in zero-shot non-IID...

  16. On the Necessity of Multi-Domain Explanation: An Uncertainty Principle Approach for Deep Time Series Models

    cs.LG 2025-06 conditional novelty 6.0 of 10

    A time-frequency uncertainty bound can tell when a model's time-domain and frequency-domain attributions point to different features, making multi-domain explanations necessary.

  17. MANTIS: Detection of Zero-Day Malicious Domains Leveraging Low Reputed Hosting Infrastructure

    cs.CR 2025-02 conditional novelty 6.0 of 10

    MANTIS detects new malicious domains by monitoring the low-reputation hosting infrastructure around known malicious domains, using a graph neural network, and reports finding about 19,000 new malicious domains daily.

  18. A Physics-Flavored Transformer Network for Parametrizing Contraction Dynamics of Engineered Skeletal Muscle Tissues

    cs.LG 2026-08 conditional novelty 5.0 of 10

    A physics-flavored CNN-transformer automatically fits stretched-exponential parameters to engineered skeletal muscle contraction curves, using synthetic pre-training and unsupervised real-data alignment.

  19. Automated Solar Radio Burst Detection Using Deep Learning on Augmented e-Callisto Data

    astro-ph.SR 2026-07 conditional novelty 5.0 of 10

    FlareSense, a ResNet detector trained on 304,750 e-Callisto spectrograms with SpecAugment and TimeWarp, reaches 93% precision and 73.15% recall, outperforming routine expert cataloging at matched precision.

  20. A Preliminary Study on Explaining Risk of Code Changes using LLM-Based Prediction Models

    cs.SE 2026-07 conditional novelty 5.0 of 10

    Top-2 attention-highlighted hunks cover expert-labeled outage root causes 53.85% of the time while requiring review of 26.28% of changed lines.

  21. Rashomon in the Streets: Explanation Ambiguity in Scene Understanding

    cs.LG 2025-09 conditional novelty 5.0 of 10

    Equally accurate models trained on the same driving scenes often point to different features as the cause of an action, so the explanation you get depends on which model you happen to train.

  22. Exploring Pose-based Sign Language Translation: Ablation Studies and Attention Insights

    cs.CV 2025-07 conditional novelty 5.0 of 10

    Pose normalization based on the signer's signing space substantially improves gloss-free sign language translation with a T5 model, while interpolation and augmentation give smaller, less certain gains.

  23. Crushing the Evidence: A Dual-Penalty Evasion Framework for Fooling White-Box Explainable AI Auditors

    cs.LG 2026-08 conditional novelty 4.0 of 10

    A dual-penalty training loss that suppresses trigger-feature gradients hides tabular backdoors from Integrated Gradients, DeepSHAP, and CAD-Detect while maintaining high attack success.

  24. Functional Groups are All you Need for Chemically Interpretable Molecular Property Prediction

    cs.LG 2025-09 conditional novelty 4.0 of 10

    FGR encodes molecules as functional-group bit vectors, embeds them with an autoencoder, and reports competitive or better benchmark accuracy while enabling chemical attribution.

  25. Uncovering Latent Connections in Indigenous Heritage: Semantic Pipelines for Cultural Preservation in Brazil

    cs.HC 2025-07 conditional novelty 4.0 of 10

    The paper introduces two embedding pipelines and a visualization tool that reveal latent clusters and label errors in the Museu Nacional dos Povos Indígenas digital collection.

  26. PyG 2.0: Scalable Learning on Real World Graphs

    cs.LG 2025-07 conditional novelty 4.0 of 10

    PyG 2.0 is presented as a modular, scalable graph-learning framework with heterogeneous and temporal graph support, compilation-based speedups, and explainability.

  27. Why Uncertainty Calibration Matters for Reliable Perturbation-based Explanations

    cs.LG 2025-06 conditional novelty 4.0 of 10

    Calibration of a model under the exact perturbations used by an explanation method improves explanation fidelity, and ReCalX achieves this with per-perturbation-strength temperature scaling.

  28. AnomalyExplainer Explainable AI for LLM-based anomaly detection using BERTViz and Captum

    cs.LG 2025-08 conditional novelty 3.0 of 10

    A chatbot that classifies HDFS log lines with fine-tuned RoBERTa and provides BERTViz/Captum visual explanations was benchmarked and tested with 13 users.

  29. ExplainBench: A Benchmark Framework for Local Model Explanations in Fairness-Critical Applications

    cs.LG 2025-05 reject novelty 3.0 of 10

    ExplainBench is a proposed open-source benchmarking suite for local explanation methods, but the paper provides no empirical evaluation.

Pith tools