This morning I asked a question about
Tomasz came back with this preprint:: Gradient-based Hyperparameter Optimization through Reversible Learning by Dougal Maclaurin, David Duvenaud, Ryan P. Adams
Tuning hyperparameters of learning algorithms is hard because gradients are usually unavailable. We compute exact gradients of cross-validation performance with respect to all hyperparameters by chaining derivatives backwards through the entire training procedure. These gradients allow us to optimize thousands of hyperparameters, including step-size and momentum schedules, weight initialization distributions, richly parameterized regularization schemes, and neural network architectures. We compute hyperparameter gradients by exactly reversing the dynamics of stochastic gradient descent with momentum.The reddit comments are here . The Autograd implementation mentioned in the paper is at:
with examples using RNN and LTSMs at https://github.com/HIPS/autograd/tree/master/examples
No comments:
Post a Comment