Mitigating representation bottlenecks in multiple instance learning
File(s)
Published version
Author(s)
Papadopoulos, Marios
Giakoumoglou, Nikolaos
Floros, Andreas
Dragotti, Pier Luigi
Stathaki, Tania
Type
Conference Paper
Abstract
Multiple Instance Learning (MIL) is widely used for Whole Slide Image classification in computational pathology, yet existing approaches suffer from a representation bottleneck where diverse patch-level features are compressed into a single slide-level embedding. We propose Divide-and-Distill (D&D), which clusters the feature space into coherent regions, trains expert models on each cluster, and distills their knowledge into a unified model. Experiments demonstrate that D&D consistently improves six state-of-the-art MIL methods in both accuracy and AUC while maintaining single-model inference efficiency.
Date Issued
2025-12-06
Date Acceptance
2025-10-31
Citation
Medical Imaging meets EurIPS: MedEurIPS 2025, 2025, pp.1-5
Publisher
OpenReview
Start Page
1
End Page
5
Journal / Book Title
Medical Imaging meets EurIPS: MedEurIPS 2025
Copyright Statement
© 2025 The Author(s). This paper is licenced under a CC-BY Attribution licence 4.0 (https://creativecommons.org/licenses/by/4.0/)
License URL
Source
EurIPS 2025 (MedEurIPS workshop)
Publication Status
Published
Start Date
2025-12-06
Finish Date
2025-12-07
Coverage Spatial
Copenhagen, Denmark