Proactive Pseudo-Intervention: Pre-informed Contrastive Learning For Interpretable Vision Models

Dong Wang, Yuewei Yang, Liqun Chen, Zhe Gan, Ricardo Henao, Lawrence Carin
Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 281:20-34, 2025.

Abstract

Deep neural networks excel at comprehending complex visual signals, delivering on par or even superior performance to that of human experts. However, ad-hoc visual explanations of model decisions often reveal an alarming level of reliance on exploiting non-causal visual cues that strongly correlate with the target label in training data. As such, deep neural nets suffer compromised generalization to novel inputs collected from different sources, and the reverse engineering of their decision rules offers limited interpretability. To overcome these limitations, we present a novel contrastive learning strategy called Proactive Pseudo-Intervention (PPI) that leverages proactive interventions to guard against image features with no causal relevance. We also devise a novel pre-informed salience mapping module to identify key image pixels to intervene and show it greatly facilitates model interpretability. To demonstrate the utility of our proposals, we benchmark it on both standard natural images and challenging medical image datasets. PPI-enhanced models consistently deliver superior performance relative to competing solutions, especially on out-of-domain predictions and data integration from heterogeneous sources. Further, saliency maps of models that are trained in our PPI framework are more succinct and meaningful.

Cite this Paper

BibTeX
@InProceedings{pmlr-v281-wang25a, title = {Proactive Pseudo-Intervention: Pre-informed Contrastive Learning For Interpretable Vision Models}, author = {Wang, Dong and Yang, Yuewei and Chen, Liqun and Gan, Zhe and Henao, Ricardo and Carin, Lawrence}, booktitle = {Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {20--34}, year = {2025}, editor = {Wu, Junde and Zhu, Jiayuan and Xu, Min and Jin, Yueming}, volume = {281}, series = {Proceedings of Machine Learning Research}, month = {25 Feb}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v281/main/assets/wang25a/wang25a.pdf}, url = {https://proceedings.mlr.press/v281/wang25a.html}, abstract = {Deep neural networks excel at comprehending complex visual signals, delivering on par or even superior performance to that of human experts. However, ad-hoc visual explanations of model decisions often reveal an alarming level of reliance on exploiting non-causal visual cues that strongly correlate with the target label in training data. As such, deep neural nets suffer compromised generalization to novel inputs collected from different sources, and the reverse engineering of their decision rules offers limited interpretability. To overcome these limitations, we present a novel contrastive learning strategy called Proactive Pseudo-Intervention (PPI) that leverages proactive interventions to guard against image features with no causal relevance. We also devise a novel pre-informed salience mapping module to identify key image pixels to intervene and show it greatly facilitates model interpretability. To demonstrate the utility of our proposals, we benchmark it on both standard natural images and challenging medical image datasets. PPI-enhanced models consistently deliver superior performance relative to competing solutions, especially on out-of-domain predictions and data integration from heterogeneous sources. Further, saliency maps of models that are trained in our PPI framework are more succinct and meaningful.} }
Endnote
%0 Conference Paper %T Proactive Pseudo-Intervention: Pre-informed Contrastive Learning For Interpretable Vision Models %A Dong Wang %A Yuewei Yang %A Liqun Chen %A Zhe Gan %A Ricardo Henao %A Lawrence Carin %B Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2025 %E Junde Wu %E Jiayuan Zhu %E Min Xu %E Yueming Jin %F pmlr-v281-wang25a %I PMLR %P 20--34 %U https://proceedings.mlr.press/v281/wang25a.html %V 281 %X Deep neural networks excel at comprehending complex visual signals, delivering on par or even superior performance to that of human experts. However, ad-hoc visual explanations of model decisions often reveal an alarming level of reliance on exploiting non-causal visual cues that strongly correlate with the target label in training data. As such, deep neural nets suffer compromised generalization to novel inputs collected from different sources, and the reverse engineering of their decision rules offers limited interpretability. To overcome these limitations, we present a novel contrastive learning strategy called Proactive Pseudo-Intervention (PPI) that leverages proactive interventions to guard against image features with no causal relevance. We also devise a novel pre-informed salience mapping module to identify key image pixels to intervene and show it greatly facilitates model interpretability. To demonstrate the utility of our proposals, we benchmark it on both standard natural images and challenging medical image datasets. PPI-enhanced models consistently deliver superior performance relative to competing solutions, especially on out-of-domain predictions and data integration from heterogeneous sources. Further, saliency maps of models that are trained in our PPI framework are more succinct and meaningful.
APA
Wang, D., Yang, Y., Chen, L., Gan, Z., Henao, R. & Carin, L.. (2025). Proactive Pseudo-Intervention: Pre-informed Contrastive Learning For Interpretable Vision Models. Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 281:20-34 Available from https://proceedings.mlr.press/v281/wang25a.html.

Related Material