Timezone: »

Wide Neural Networks of Any Depth Evolve as Linear Models Under Gradient Descent
Jaehoon Lee · Lechao Xiao · Samuel Schoenholz · Yasaman Bahri · Roman Novak · Jascha Sohl-Dickstein · Jeffrey Pennington

Thu Dec 12 10:45 AM -- 12:45 PM (PST) @ East Exhibition Hall B + C #175

A longstanding goal in deep learning research has been to precisely characterize training and generalization. However, the often complex loss landscapes of neural networks have made a theory of learning dynamics elusive. In this work, we show that for wide neural networks the learning dynamics simplify considerably and that, in the infinite width limit, they are governed by a linear model obtained from the first-order Taylor expansion of the network around its initial parameters. Furthermore, mirroring the correspondence between wide Bayesian neural networks and Gaussian processes, gradient-based training of wide neural networks with a squared loss produces test set predictions drawn from a Gaussian process with a particular compositional kernel. While these theoretical results are only exact in the infinite width limit, we nevertheless find excellent empirical agreement between the predictions of the original network and those of the linearized version even for finite practically-sized networks. This agreement is robust across different architectures, optimization methods, and loss functions.

Author Information

Jaehoon Lee (Google Brain)
Lechao Xiao (Google Brain)
Samuel Schoenholz (Google Brain)
Yasaman Bahri (Google Brain)
Roman Novak (Google Brain)
Jascha Sohl-Dickstein (Google Brain)
Jeffrey Pennington (Google Brain)

More from the Same Authors