Research map / Label-efficient and robust learning / Recognition and long-tail
Post-hoc explanation analysis
Evaluates and improves post-hoc explanations to expose spurious features and user needs.
86 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Recognition and long-tail, keyword cluster “explanations · concept · interpretable”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
25 papers in 2023–24 and 28 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 2.0% to 2.1%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Examples are not enough, learn to criticize! Criticism for Interpretability (NeurIPS 2016 · 525 citations)
- 2017 · From Red Wine to Red Tomato: Composition With Context (CVPR 2017 · 247 citations)
- 2018 · Model Agnostic Supervised Local Explanations (NeurIPS 2018 · 124 citations)
- 2019 · Towards Automatic Concept-based Explanations (NeurIPS 2019 · 86 citations)
- 2020 · How Can I Explain This to You? An Empirical Study of Deep Neural Network Explanation Methods (NeurIPS 2020 · 94 citations)
- 2021 · On Generating Plausible Counterfactual and Semi-Factual Explanations for Deep Learning (AAAI 2021 · 95 citations)
- 2022 · OpenXAI: Towards a Transparent Evaluation of Model Explanations (NeurIPS 2022 · 81 citations)
- 2023 · PIP-Net: Patch-Based Intuitive Prototypes for Interpretable Image Classification (CVPR 2023 · 107 citations)
- 2024 · SLICE: Stabilized LIME for Consistent Explanations for Image Classification (CVPR 2024 · 17 citations)
- 2025 · Interactive Medical Image Analysis with Concept-based Similarity Reasoning (CVPR 2025 · 5 citations)
- 2026 · Beyond Patches: Mining Interpretable Part-Prototypes for Explainable AI (AAAI 2026 · 1 citation)
Earliest papers here
- Examples are not enough, learn to criticize! Criticism for Interpretability (NeurIPS 2016 · 525 citations)
- From Red Wine to Red Tomato: Composition With Context (CVPR 2017 · 247 citations)
- Practical TBox Abduction Based on Justification Patterns (AAAI 2017 · 15 citations)
What has been tried
72 papers here report 658 measured changes (a component added, removed or swapped, compared with a baseline): helped 64%, no clear effect 9%, hurt 23%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add Ada-Blur to SLICE FE | Oxford-IIIT Pets and PASCAL VOC 2007 with InceptionV3 and ResNet50 · consistency (CCM) | Helped | SLICE: Stabilized LIME for Consistent Explanations for Image Classification (CVPR 2024 · 17 citations) | |
| Show positive and negative features from one model instead of explanations from disagreeing models | Deceptive hotel reviews · Overreliance | Hurt | Dissenting Explanations: Leveraging Disagreement to Reduce Model Overreliance (AAAI 2024 · 10 citations) | |
| Refine the concept atlas with doctor review | TBX11K · Macro F1-score | 94.4 -> 94.0 | Hurt | Interactive Medical Image Analysis with Concept-based Similarity Reasoning (CVPR 2025 · 5 citations) |
| Tailor selected examples to each user's expertise rather than using examples selected for a different user's expertise | Synthetic dataset · simulatability accuracy | Helped | I-CEE: Tailoring Explanations of Image Classification Models to User Expertise (AAAI 2024 · 2 citations) | |
| Add inference-time finetuning to pBEE | ImageNet validation · INS | 51.15 -> 51.73 | Helped | BEE: Metric-Adapted Explanations via Baseline Exploration-Exploitation (AAAI 2025 · 1 citation) |
| Add iterative local-importance recomputation and reclustering | Synthetic Patient Classification · Cohort Locality | 0.0157 ± 0.0050 -> 0.0052 ± 0.0088 | Helped | CohEx: A Generalized Framework for Cohort Explanation (AAAI 2025 · 1 citation) |
| Set window size w to 1 instead of w = 3 | 150-sentence double-negative dataset · F1 | 0.9155 -> 0.7541 | Hurt | NOMATTERXAI: Generating “No Matter What” Alterfactual Examples for Explaining Black-Box Text Classification Models (AAAI 2025 · 1 citation) |
| Replace L1 mask regularization with annealing L1 regularization | DSP · balanced accuracy | 90.75 +/- 0.01 -> 87.76 +/- 3.12 | Hurt | Intrinsic User-Centric Interpretability through Global Mixture of Experts (ICLR 2025 · 1 citation) |
| Refine concept embeddings with CCR instead of using the unrefined CCR baseline | CIFAR-10, CIFAR-100, ImageNet, CUB-200, Places365 · test accuracy | Helped | Enhancing Performance of Explainable AI Models with Constrained Concept Refinement (ICML 2025 · 0 citations) | |
| Add Gaussian noise N(0, 0.1) to training inputs | Boolean logic gates · Total mappings | 9.86 -> 9.89; p-value 0.959 | No clear effect | Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable? (ICLR 2025 · 0 citations) |
| Use robust masking instead of baseline masking | MNIST · size (%) | 23.69 -> 1.42 | Helped | Explain Yourself, Briefly! Self-Explaining Neural Networks with Concise Sufficient Reasons (ICLR 2025 · 0 citations) |
| Use raw features rather than learned data representations | Image classification datasets, Horizontal Flip · Hit Rate | Hurt | Data-centric Prediction Explanation via Kernelized Stein Discrepancy (ICLR 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 20 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 4 of 20 papers
- ImageNet · 4 of 20 papers
- MNIST · 3 of 20 papers
- CIFAR-100 · 2 of 20 papers
Baselines
- Grad-CAM · 2 of 20 papers
- Integrated Gradients · 2 of 20 papers
- SHAP · 2 of 20 papers
- LIME · 2 of 20 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 53 papers; from OpenAlex and the papers' affiliations).
- Meta · 3 papers
- CNRS · 2 papers
- Johns Hopkins University · 2 papers
- University College Dublin · 2 papers
- Insight Centre for Data Analytics · 2 papers
- Rice University · 2 papers
- Purdue University · 2 papers
- University of Washington · 2 papers
- Princeton University · 2 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Self-explaining and visual explanation networks (108 papers)
- Prototype and concept dictionary models (104 papers)
- Feature attribution methods (51 papers)
- Concept bottleneck models (30 papers)
Elsewhere in the topic
- Multilingual language representation pretraining (507 papers)
- Contrastive representation theory (495 papers)
- Dense features and foundation priors (482 papers)
- Correspondence and causal representation learning (447 papers)
- Data augmentation and generalization bias (409 papers)
- Partial and multi-label learning (345 papers)
- Loss design and classifier theory (273 papers)
- Group-robust classification (251 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).