Online Structured Laplace Approximations For Overcoming Catastrophic Forgetting
This addresses the problem of catastrophic forgetting for neural networks in sequential learning scenarios, representing a strong incremental improvement with specific gains.
The authors tackled catastrophic forgetting in neural networks by developing a Kronecker factored online Laplace approximation method, which achieved over 90% test accuracy across 50 permuted MNIST tasks, substantially outperforming related methods.
We introduce the Kronecker factored online Laplace approximation for overcoming catastrophic forgetting in neural networks. The method is grounded in a Bayesian online learning framework, where we recursively approximate the posterior after every task with a Gaussian, leading to a quadratic penalty on changes to the weights. The Laplace approximation requires calculating the Hessian around a mode, which is typically intractable for modern architectures. In order to make our method scalable, we leverage recent block-diagonal Kronecker factored approximations to the curvature. Our algorithm achieves over 90% test accuracy across a sequence of 50 instantiations of the permuted MNIST dataset, substantially outperforming related methods for overcoming catastrophic forgetting.