Incorporating probabilistic domain knowledge into deep multiple instance learning

Ghadi S. Al Hajj, Aliaksandr Hubin, Chakravarthi Kanduri, Milena Pavlovic, Knut Dagestad Rand, Michael Widrich, Anne Schistad Solberg, Victor Greiff, Johan Pensar, Günter Klambauer, Geir Kjetil Sandve
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:17279-17297, 2024.

Abstract

Deep learning methods, including deep multiple instance learning methods, have been criticized for their limited ability to incorporate domain knowledge. A reason that knowledge incorporation is challenging in deep learning is that the models usually lack a mapping between their model components and the entities of the domain, making it a non-trivial task to incorporate probabilistic prior information. In this work, we show that such a mapping between domain entities and model components can be defined for a multiple instance learning setting and propose a framework DeeMILIP that encompasses multiple strategies to exploit this mapping for prior knowledge incorporation. We motivate and formalize these strategies from a probabilistic perspective. Experiments on an immune-based diagnostics case show that our proposed strategies allow to learn generalizable models even in settings with weak signals, limited dataset size, and limited compute.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-hajj24a, title = {Incorporating probabilistic domain knowledge into deep multiple instance learning}, author = {Hajj, Ghadi S. Al and Hubin, Aliaksandr and Kanduri, Chakravarthi and Pavlovic, Milena and Rand, Knut Dagestad and Widrich, Michael and Solberg, Anne Schistad and Greiff, Victor and Pensar, Johan and Klambauer, G\"{u}nter and Sandve, Geir Kjetil}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {17279--17297}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/hajj24a/hajj24a.pdf}, url = {https://proceedings.mlr.press/v235/hajj24a.html}, abstract = {Deep learning methods, including deep multiple instance learning methods, have been criticized for their limited ability to incorporate domain knowledge. A reason that knowledge incorporation is challenging in deep learning is that the models usually lack a mapping between their model components and the entities of the domain, making it a non-trivial task to incorporate probabilistic prior information. In this work, we show that such a mapping between domain entities and model components can be defined for a multiple instance learning setting and propose a framework DeeMILIP that encompasses multiple strategies to exploit this mapping for prior knowledge incorporation. We motivate and formalize these strategies from a probabilistic perspective. Experiments on an immune-based diagnostics case show that our proposed strategies allow to learn generalizable models even in settings with weak signals, limited dataset size, and limited compute.} }
Endnote
%0 Conference Paper %T Incorporating probabilistic domain knowledge into deep multiple instance learning %A Ghadi S. Al Hajj %A Aliaksandr Hubin %A Chakravarthi Kanduri %A Milena Pavlovic %A Knut Dagestad Rand %A Michael Widrich %A Anne Schistad Solberg %A Victor Greiff %A Johan Pensar %A Günter Klambauer %A Geir Kjetil Sandve %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-hajj24a %I PMLR %P 17279--17297 %U https://proceedings.mlr.press/v235/hajj24a.html %V 235 %X Deep learning methods, including deep multiple instance learning methods, have been criticized for their limited ability to incorporate domain knowledge. A reason that knowledge incorporation is challenging in deep learning is that the models usually lack a mapping between their model components and the entities of the domain, making it a non-trivial task to incorporate probabilistic prior information. In this work, we show that such a mapping between domain entities and model components can be defined for a multiple instance learning setting and propose a framework DeeMILIP that encompasses multiple strategies to exploit this mapping for prior knowledge incorporation. We motivate and formalize these strategies from a probabilistic perspective. Experiments on an immune-based diagnostics case show that our proposed strategies allow to learn generalizable models even in settings with weak signals, limited dataset size, and limited compute.
APA
Hajj, G.S.A., Hubin, A., Kanduri, C., Pavlovic, M., Rand, K.D., Widrich, M., Solberg, A.S., Greiff, V., Pensar, J., Klambauer, G. & Sandve, G.K.. (2024). Incorporating probabilistic domain knowledge into deep multiple instance learning. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:17279-17297 Available from https://proceedings.mlr.press/v235/hajj24a.html.

Related Material