Understanding Mode Connectivity via Parameter Space Symmetry

Bo Zhao, Nima Dehmamy, Robin Walters, Rose Yu
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:77441-77460, 2025.

Abstract

Neural network minima are often connected by curves along which train and test loss remain nearly constant, a phenomenon known as mode connectivity. While this property has enabled applications such as model merging and fine-tuning, its theoretical explanation remains unclear. We propose a new approach to exploring the connectedness of minima using parameter space symmetry. By linking the topology of symmetry groups to that of the minima, we derive the number of connected components of the minima of linear networks and show that skip connections reduce this number. We then examine when mode connectivity and linear mode connectivity hold or fail, using parameter symmetries which account for a significant part of the minimum. Finally, we provide explicit expressions for connecting curves in the minima induced by symmetry. Using the curvature of these curves, we derive conditions under which linear mode connectivity approximately holds. Our findings highlight the role of continuous symmetries in understanding the neural network loss landscape.

Cite this Paper


BibTeX
@InProceedings{pmlr-v267-zhao25i, title = {Understanding Mode Connectivity via Parameter Space Symmetry}, author = {Zhao, Bo and Dehmamy, Nima and Walters, Robin and Yu, Rose}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {77441--77460}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhao25i/zhao25i.pdf}, url = {https://proceedings.mlr.press/v267/zhao25i.html}, abstract = {Neural network minima are often connected by curves along which train and test loss remain nearly constant, a phenomenon known as mode connectivity. While this property has enabled applications such as model merging and fine-tuning, its theoretical explanation remains unclear. We propose a new approach to exploring the connectedness of minima using parameter space symmetry. By linking the topology of symmetry groups to that of the minima, we derive the number of connected components of the minima of linear networks and show that skip connections reduce this number. We then examine when mode connectivity and linear mode connectivity hold or fail, using parameter symmetries which account for a significant part of the minimum. Finally, we provide explicit expressions for connecting curves in the minima induced by symmetry. Using the curvature of these curves, we derive conditions under which linear mode connectivity approximately holds. Our findings highlight the role of continuous symmetries in understanding the neural network loss landscape.} }
Endnote
%0 Conference Paper %T Understanding Mode Connectivity via Parameter Space Symmetry %A Bo Zhao %A Nima Dehmamy %A Robin Walters %A Rose Yu %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhao25i %I PMLR %P 77441--77460 %U https://proceedings.mlr.press/v267/zhao25i.html %V 267 %X Neural network minima are often connected by curves along which train and test loss remain nearly constant, a phenomenon known as mode connectivity. While this property has enabled applications such as model merging and fine-tuning, its theoretical explanation remains unclear. We propose a new approach to exploring the connectedness of minima using parameter space symmetry. By linking the topology of symmetry groups to that of the minima, we derive the number of connected components of the minima of linear networks and show that skip connections reduce this number. We then examine when mode connectivity and linear mode connectivity hold or fail, using parameter symmetries which account for a significant part of the minimum. Finally, we provide explicit expressions for connecting curves in the minima induced by symmetry. Using the curvature of these curves, we derive conditions under which linear mode connectivity approximately holds. Our findings highlight the role of continuous symmetries in understanding the neural network loss landscape.
APA
Zhao, B., Dehmamy, N., Walters, R. & Yu, R.. (2025). Understanding Mode Connectivity via Parameter Space Symmetry. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:77441-77460 Available from https://proceedings.mlr.press/v267/zhao25i.html.

Related Material