On Characterizing the Capacity of Neural Networks using Algebraic Topology

  • William Guss | Carnegie Mellon University

The learnability of different neural architectures can be characterized directly by computable measures of data complexity. In this talk, we reframe the problem of architecture selection as understanding how data determines the most expressive and generalizable architectures suited to that data, beyond inductive bias. After suggesting algebraic topology as a measure for data complexity, we show that the power of a network to express the topological complexity of a dataset in its decision region is a strictly limiting factor in its ability to generalize. We then provide the first empirical characterization of the topological capacity of neural networks. Our empirical analysis shows that at every level of dataset complexity, neural networks exhibit topological phase transitions. This observation allows us to connect existing theory to empirically driven conjectures on the choice of architectures for fully-connected neural networks. Finally, we provide some first steps in building a general theory of neural homology.

Series: Microsoft Research Talks