Fast Approximate Natural Gradient Descent in a Kronecker-factored Eigenbasis
Citations Over Time
Abstract
Optimization algorithms that leverage gradient covariance information, such as variants of natural gradient descent (Amari, 1998), offer the prospect of yielding more effective descent directions. For models with many parameters, the covariance matrix they are based on becomes gigantic, making them inapplicable in their original form. This has motivated research into both simple diagonal approximations and more sophisticated factored approximations such as KFAC (Heskes, 2000; Martens & Grosse, 2015; Grosse & Martens, 2016). In the present work we draw inspiration from both to propose a novel approximation that is provably better than KFAC and amendable to cheap partial updates. It consists in tracking a diagonal variance, not in parameter coordinates, but in a Kronecker-factored eigenbasis, in which the diagonal approximation is likely to be more effective. Experiments show improvements over KFAC in optimization speed for several deep network architectures.
Related Papers
- → Subrepresentations of Kronecker representations(2005)18 cited
- How Far from Kronecker can a MIMO Channel be? Does it Matter?(2011)
- → Combinatorics on a family of reduced Kronecker coefficients(2015)2 cited