# Attention-based multiple-instance learning (ABMIL, CLAM)

Source: https://onco.cc/terms/abmil/  
OnCo record `abmil` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Attention-based multiple-instance learning gives each tile of a slide a learned weight and sums the weighted tile vectors into one slide vector, so a slide-level label can train the model and the weights show which regions mattered.

## Summary

In multiple-instance learning the learner receives labelled bags of instances rather than labelled instances (Wikipedia). Ilse, Tomczak and Welling proposed pooling the instances with a small attention network, and Lu and colleagues' CLAM applied it to whole-slide images with clustering constraints for data-efficient, weakly supervised pathology. It is the default slide encoder head over frozen tile embeddings, the interface most pathology pipelines expose, and its attention maps are read as heat maps of evidence.

## Fields

- Kind: Term
- Last checked: 2026-09-24
- Also known as: ABMIL; attention-based MIL; attention-based multiple instance learning; multiple-instance learning; multiple instance learning; MIL; CLAM; attention pooling; slide-level aggregation; weakly supervised slide classification
- Tags: cansim-terms

## Notes

- Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/abmil.

## Sources

- Wikipedia: https://en.wikipedia.org/wiki/Multiple_instance_learning
- Ilse, Tomczak and Welling, Attention-based deep multiple instance learning (arXiv 2018): https://arxiv.org/abs/1802.04712
- Lu et al., CLAM (Nature Biomedical Engineering 2021): https://doi.org/10.1038/s41551-020-00682-w
- Wikipedia: https://en.wikipedia.org/wiki/Multiple_instance_learning

## Connected records

- terms: [Cancer AI vocabulary (CanSim terms map)](https://onco.cc/terms/cancer-ai-vocabulary/), [Pathology foundation models: UNI, UNI2, Virchow2, CTransPath, CONCH, TITAN](https://onco.cc/terms/pathology-foundation-models/), [Tile and patch encoding of slides](https://onco.cc/terms/tile-patch-encoding/), [Transformer and attention](https://onco.cc/terms/transformer-architecture/)

---
JSON: https://onco.cc/api/v1/entities/abmil.json