Paper Detail
Luca Zedda, Davide Antonio Mura, Cecilia Di Ruberto, Maurizio Atzori, Muhammed Furkan Dasdelen, Carsten Marr, Andrea Loddo
Attention-based Multiple Instance Learning aggregators in medical imaging are prone to attention concentration, producing overconfident and unstable predictions. We introduce QG-MIL, a gated transformer aggregator that addresses this through four synergistic architectural components: RMSNorm-based pre-normalization, per-head QK normalization, fine-grained attention output gating, and SwiGLU-style feed-forward modules. Together, these design choices stabilize training and distribute attention more uniformly across instances without auxiliary losses, masking, or multi-stage regularization. We evaluate QG-MIL across six benchmarks spanning whole-slide pathology and cell-level hematology, covering two fundamentally different MIL scales. The best-performing QG-MIL variants outperform leading baselines on all six benchmarks, with an average improvement of +6.1 mean macro F1 points. Attention overlays and attention mass analysis confirm more distributed instance weighting. Ablation studies show that while individual components can match the full model on specific datasets, the QG-MIL design provides the most consistent cross-domain performance and tightest variance when compared to selected baselines. We release a configurable implementation to support reproducibility at: https://github.com/unica-visual-intelligence-lab/QG-MIL
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{zedda2026qg,
title = {QG-MIL: A Gated Transformer Aggregator for Domain-Agnostic Multiple Instance Learning in Medical Imaging},
author = {Luca Zedda and Davide Antonio Mura and Cecilia Di Ruberto and Maurizio Atzori and Muhammed Furkan Dasdelen and Carsten Marr and Andrea Loddo},
year = {2026},
abstract = {Attention-based Multiple Instance Learning aggregators in medical imaging are prone to attention concentration, producing overconfident and unstable predictions. We introduce QG-MIL, a gated transformer aggregator that addresses this through four synergistic architectural components: RMSNorm-based pre-normalization, per-head QK normalization, fine-grained attention output gating, and SwiGLU-style feed-forward modules. Together, these design choices stabilize training and distribute attention mor},
url = {https://huggingface.co/papers/2606.20027},
keywords = {Attention-based Multiple Instance Learning, gated transformer aggregator, RMSNorm-based pre-normalization, per-head QK normalization, fine-grained attention output gating, SwiGLU-style feed-forward modules, attention concentration, overconfident predictions, unstable predictions, medical imaging, whole-slide pathology, cell-level hematology, MIL scales, mean macro F1, attention overlays, attention mass analysis, ablation studies, cross-domain performance, code available, huggingface daily},
eprint = {2606.20027},
archiveprefix = {arXiv},
}
{}