Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Symmetry and Geometry in Neural Representations (NeurReps)

Connectedness of loss landscapes via the lens of Morse theory

Danil Akhtiamov · Matt Thomson

Keywords: [ Morse theory ] [ Non-Convex Optimization ] [ Loss Landscapes ] [ Saddle Points ] [ Mode connectivity ]


Abstract:

Mode connectivity is a recently discovered property of neural networks saying that two weights of small loss can usually be connected by a path of small loss. This property is interesting practically as it has applications to design of optimizers with better generalization properties and various other applied topics as well as theoretically as it suggests that loss landscapes of deep networks have very nice properties even though they are known to be highly non-convex. The goal of this work is to study connectedness of loss landscapes via the lens of Morse theory. A brief introduction to Morse theory is provided.

Chat is not available.