SAEs-BrainMap: Unveiling the Emergence of Specialized Concepts in Deep Models via Brain Alignment

Ziming Mao, Jia Xu, Wenxuan Pan, Mufan Xue, Yaochu Jin, Guoyuan Yang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:86540-86565, 2026.

Abstract

Understanding the internal mechanisms of deep neural networks remains a significant challenge, particularly in elucidating how generic visual concepts emerge within latent spaces. In this work, we propose SAEs-BrainMap, a novel framework that utilizes human brain activation patterns from the ventral visual pathway as objective probes to guide the identification of features decomposed by Sparse Autoencoders (SAEs). Our quantitative and qualitative empirical results demonstrate a robust representational alignment between sparse model features and biological Regions of Interests (ROIs), confirming the feasibility of utilizing brain signals to characterize model functionality. By leveraging this alignment, we trace the hierarchical trajectory of generic concepts cross layers and utilize the brain’s hierarchical structure to visualize the model’s global processing flow, providing novel insights into model interpretability. Our code is available at this repository.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-mao26h, title = {{SAE}s-{B}rain{M}ap: Unveiling the Emergence of Specialized Concepts in Deep Models via Brain Alignment}, author = {Mao, Ziming and Xu, Jia and Pan, Wenxuan and Xue, Mufan and Jin, Yaochu and Yang, Guoyuan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {86540--86565}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/mao26h/mao26h.pdf}, url = {https://proceedings.mlr.press/v306/mao26h.html}, abstract = {Understanding the internal mechanisms of deep neural networks remains a significant challenge, particularly in elucidating how generic visual concepts emerge within latent spaces. In this work, we propose SAEs-BrainMap, a novel framework that utilizes human brain activation patterns from the ventral visual pathway as objective probes to guide the identification of features decomposed by Sparse Autoencoders (SAEs). Our quantitative and qualitative empirical results demonstrate a robust representational alignment between sparse model features and biological Regions of Interests (ROIs), confirming the feasibility of utilizing brain signals to characterize model functionality. By leveraging this alignment, we trace the hierarchical trajectory of generic concepts cross layers and utilize the brain’s hierarchical structure to visualize the model’s global processing flow, providing novel insights into model interpretability. Our code is available at this repository.} }
Endnote
%0 Conference Paper %T SAEs-BrainMap: Unveiling the Emergence of Specialized Concepts in Deep Models via Brain Alignment %A Ziming Mao %A Jia Xu %A Wenxuan Pan %A Mufan Xue %A Yaochu Jin %A Guoyuan Yang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-mao26h %I PMLR %P 86540--86565 %U https://proceedings.mlr.press/v306/mao26h.html %V 306 %X Understanding the internal mechanisms of deep neural networks remains a significant challenge, particularly in elucidating how generic visual concepts emerge within latent spaces. In this work, we propose SAEs-BrainMap, a novel framework that utilizes human brain activation patterns from the ventral visual pathway as objective probes to guide the identification of features decomposed by Sparse Autoencoders (SAEs). Our quantitative and qualitative empirical results demonstrate a robust representational alignment between sparse model features and biological Regions of Interests (ROIs), confirming the feasibility of utilizing brain signals to characterize model functionality. By leveraging this alignment, we trace the hierarchical trajectory of generic concepts cross layers and utilize the brain’s hierarchical structure to visualize the model’s global processing flow, providing novel insights into model interpretability. Our code is available at this repository.
APA
Mao, Z., Xu, J., Pan, W., Xue, M., Jin, Y. & Yang, G.. (2026). SAEs-BrainMap: Unveiling the Emergence of Specialized Concepts in Deep Models via Brain Alignment. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:86540-86565 Available from https://proceedings.mlr.press/v306/mao26h.html.

Related Material