Probing Biological and Artificial Neural Networks with Task-dependent Neural Manifolds

Michael Kuoch, Chi-Ning Chou, Nikhil Parthasarathy, Joel Dapello, James J. DiCarlo, Haim Sompolinsky, SueYeon Chung
Conference on Parsimony and Learning, PMLR 234:395-418, 2024.

Abstract

In recent years, growth in our understanding of the computations performed in both biological and artificial neural networks has largely been driven by either low-level mechanistic studies or global normative approaches. However, concrete methodologies for bridging the gap between these levels of abstraction remain elusive. In this work, we investigate the internal mechanisms of neural networks through the lens of neural population geometry, aiming to provide understanding at an intermediate level of abstraction, as a way to bridge that gap. Utilizing manifold capacity theory (MCT) from statistical physics and manifold alignment analysis (MAA) from high-dimensional statistics, we probe the underlying organization of task-dependent manifolds in deep neural networks and neural recordings from the macaque visual cortex. Specifically, we quantitatively characterize how different learning objectives lead to differences in the organizational strategies of these models and demonstrate how these geometric analyses are connected to the decodability of task-relevant information. Furthermore, these metrics show that macaque visual cortex data are more similar to unsupervised DNNs in terms of geometrical properties such as manifold position and manifold alignment. These analyses present a strong direction for bridging mechanistic and normative theories in neural networks through neural population geometry, potentially opening up many future research avenues in both machine learning and neuroscience.

Cite this Paper

BibTeX
@InProceedings{pmlr-v234-kuoch24a, title = {Probing Biological and Artificial Neural Networks with Task-dependent Neural Manifolds}, author = {Kuoch, Michael and Chou, Chi-Ning and Parthasarathy, Nikhil and Dapello, Joel and DiCarlo, James J. and Sompolinsky, Haim and Chung, SueYeon}, booktitle = {Conference on Parsimony and Learning}, pages = {395--418}, year = {2024}, editor = {Chi, Yuejie and Dziugaite, Gintare Karolina and Qu, Qing and Wang, Atlas Wang and Zhu, Zhihui}, volume = {234}, series = {Proceedings of Machine Learning Research}, month = {03--06 Jan}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v234/kuoch24a/kuoch24a.pdf}, url = {https://proceedings.mlr.press/v234/kuoch24a.html}, abstract = {In recent years, growth in our understanding of the computations performed in both biological and artificial neural networks has largely been driven by either low-level mechanistic studies or global normative approaches. However, concrete methodologies for bridging the gap between these levels of abstraction remain elusive. In this work, we investigate the internal mechanisms of neural networks through the lens of neural population geometry, aiming to provide understanding at an intermediate level of abstraction, as a way to bridge that gap. Utilizing manifold capacity theory (MCT) from statistical physics and manifold alignment analysis (MAA) from high-dimensional statistics, we probe the underlying organization of task-dependent manifolds in deep neural networks and neural recordings from the macaque visual cortex. Specifically, we quantitatively characterize how different learning objectives lead to differences in the organizational strategies of these models and demonstrate how these geometric analyses are connected to the decodability of task-relevant information. Furthermore, these metrics show that macaque visual cortex data are more similar to unsupervised DNNs in terms of geometrical properties such as manifold position and manifold alignment. These analyses present a strong direction for bridging mechanistic and normative theories in neural networks through neural population geometry, potentially opening up many future research avenues in both machine learning and neuroscience.} }
Endnote
%0 Conference Paper %T Probing Biological and Artificial Neural Networks with Task-dependent Neural Manifolds %A Michael Kuoch %A Chi-Ning Chou %A Nikhil Parthasarathy %A Joel Dapello %A James J. DiCarlo %A Haim Sompolinsky %A SueYeon Chung %B Conference on Parsimony and Learning %C Proceedings of Machine Learning Research %D 2024 %E Yuejie Chi %E Gintare Karolina Dziugaite %E Qing Qu %E Atlas Wang Wang %E Zhihui Zhu %F pmlr-v234-kuoch24a %I PMLR %P 395--418 %U https://proceedings.mlr.press/v234/kuoch24a.html %V 234 %X In recent years, growth in our understanding of the computations performed in both biological and artificial neural networks has largely been driven by either low-level mechanistic studies or global normative approaches. However, concrete methodologies for bridging the gap between these levels of abstraction remain elusive. In this work, we investigate the internal mechanisms of neural networks through the lens of neural population geometry, aiming to provide understanding at an intermediate level of abstraction, as a way to bridge that gap. Utilizing manifold capacity theory (MCT) from statistical physics and manifold alignment analysis (MAA) from high-dimensional statistics, we probe the underlying organization of task-dependent manifolds in deep neural networks and neural recordings from the macaque visual cortex. Specifically, we quantitatively characterize how different learning objectives lead to differences in the organizational strategies of these models and demonstrate how these geometric analyses are connected to the decodability of task-relevant information. Furthermore, these metrics show that macaque visual cortex data are more similar to unsupervised DNNs in terms of geometrical properties such as manifold position and manifold alignment. These analyses present a strong direction for bridging mechanistic and normative theories in neural networks through neural population geometry, potentially opening up many future research avenues in both machine learning and neuroscience.
APA
Kuoch, M., Chou, C., Parthasarathy, N., Dapello, J., DiCarlo, J.J., Sompolinsky, H. & Chung, S.. (2024). Probing Biological and Artificial Neural Networks with Task-dependent Neural Manifolds. Conference on Parsimony and Learning, in Proceedings of Machine Learning Research 234:395-418 Available from https://proceedings.mlr.press/v234/kuoch24a.html.

Related Material