mlsanityData to 14 Oct 2026

Are Compact Rationales Free? Measuring Tile Selection Headroom in Frozen WSI-MIL

arXiv 2605.12575 ↗First posted 12 May 2026eess.IV cs.AI cs.CVcited by 1

Not on the map. Why?

Abstract (arXiv)

Whole-slide image (WSI) multiple instance learning (MIL) classifiers can achieve strong slide-level AUC while leaving the full-bag prediction opaque. Attention scores are widely reused as post-hoc explanations, but high attention can reflect aggregation preference rather than a compact, model-sufficient rationale. We study post-hoc rationale highlighting for frozen WSI-MIL: given a trained classifier, can its slide-level prediction be recovered from a compact, output-consistent tile subset without retraining the backbone? We instantiate this with Finding Optimal Contextual Instances (FOCI), a lightweight rationale-readout layer over a frozen MIL backbone. FOCI is trained with model-output sufficiency and exclusion objectives over keep/drop tile subsets, evaluated with an insertion-style Sequential Reveal Protocol (SRP) adapted to WSI-MIL, and summarized by the Selection Headroom Index (SHI). Across three WSI benchmarks and seven MIL backbones, FOCI reveals that compact rationales are selection-headroom dependent: transformer and multi-branch attention aggregators can admit compact rationales, near-minimal attention-pooling baselines enter a selection-saturation regime, and hard-selection backbones can conflict with an external readout. For TransMIL, relative to its documented CLS-proxy ranking, FOCI reduces the Minimum Sufficient K (MSK) tile count by 32-56% across benchmarks, while ACMIL+FOCI attains the highest mean SHI (+0.465). Deletion-based perturbation and selected-only downstream evaluation provide complementary checks. These results position FOCI as a model-level interpretability and audit layer: selected tiles are not claims of clinical or pathologist-level diagnostic sufficiency, but candidate rationales that offer a compact, reviewable view of when a frozen MIL prediction can be localized to a small output-consistent subset.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 30 of its 40 references that are in the map. First posted on arXiv 12 May 2026.

Builds on

30 of 40 references.

References · 30

  1. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  2. 2016Categorical Reparameterization with Gumbel-Softmaxcited by 2,739
  3. 2013Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computationcited by 2,079
  4. 2016The Concrete Distribution: A Continuous Relaxation of Discrete Random Variablescited by 1,148
  5. 2018Attention-based Deep Multiple Instance Learningcited by 1,095
  6. 2020Concept Bottleneck Modelscited by 1,032
  7. 2017Selective Classification for Deep Neural Networkscited by 852
  8. 2020Data Efficient and Weakly Supervised Computational Pathology on Whole Slide Imagescited by 745
  9. 2017BranchyNet: Fast Inference via Early Exiting from Deep Neural Networkscited by 596
  10. 2021TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classificationcited by 555
  11. 2015Evaluating the visualization of what a Deep Neural Network has learnedcited by 419
  12. 2019Is Attention Interpretable?cited by 389
18 more
  1. 2022Scaling Vision Transformers to Gigapixel Images via Hierarchical Self-Supervised Learningcited by 359
  2. 2016Rationalizing Neural Predictionscited by 281
  3. 2019Interpretable Neural Predictions with Differentiable Binary Variablescited by 126
  4. 2019Learning to Deceive with Attention-Based Explanationscited by 111
  5. 2019Rethinking Cooperative Rationalization: Introspective Extraction and Complement Controlcited by 96
  6. 2023Attention-Challenging Multiple Instance Learning for Whole Slide Image Classificationcited by 61
  7. 2023Multiple Instance Learning Framework with Masked Hard Instance Mining for Whole Slide Image Classificationcited by 60
  8. 2022Additive MIL: Intrinsically Interpretable Multiple Instance Learning for Pathologycited by 58
  9. 2022Multiple Instance Learning for Digital Pathology: A Review on the State-of-the-Art, Limitations & Future Potentialcited by 54
  10. 2019Deep Learning for Whole Slide Image Analysis: An Overviewcited by 51
  11. 2023SI-MIL: Taming Deep MIL for Self-Interpretability in Gigapixel Histopathologycited by 25
  12. 2023Rationalization for Explainable NLP: A Surveycited by 22
  13. 2024xMIL: Insightful Explanations for Multiple Instance Learning in Histopathologycited by 16
  14. 2023Rethinking Multiple Instance Learning for Whole Slide Image Classification: A Bag-Level Classifier is a Good Instance-Level Teachercited by 9
  15. 2025How Effective Can Dropout Be in Multiple Instance Learning ?cited by 7
  16. 2026ASMIL: Attention-Stabilized Multiple Instance Learning for Whole Slide Imagingcited by 6
  17. 2025Boosting Explainability through Selective Rationalization in Pre-trained Language Modelscited by 2
  18. 2026ReaMIL: Reasoning- and Evidence-Aware Multiple Instance Learning for Whole-Slide Histopathologycited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026CHARTER: Auditing Reference Substitution in Hierarchical Compact-Evidence Evaluation for Computational Pathologycited by 0