Machine learning models are often tuned by nesting optimization of model
weights inside the optimization of hyperparameters. We give a method to
collapse this nested optimization into joint stochastic optimization of weights
and hyperparameters. Our process trains a neural network to output
approximately optimal weights as a function of hyperparameters. We show that
our technique converges to locally optimal weights and hyperparameters for
sufficiently large hypernetworks. We compare this method to standard
hyperparameter optimization strategies and demonstrate its effectiveness for
tuning thousands of hyperparameters.
Описание
Stochastic Hyperparameter Optimization through Hypernetworks
%0 Generic
%1 lorraine2018stochastic
%A Lorraine, Jonathan
%A Duvenaud, David
%D 2018
%K SGD optimization
%T Stochastic Hyperparameter Optimization through Hypernetworks
%U http://arxiv.org/abs/1802.09419
%X Machine learning models are often tuned by nesting optimization of model
weights inside the optimization of hyperparameters. We give a method to
collapse this nested optimization into joint stochastic optimization of weights
and hyperparameters. Our process trains a neural network to output
approximately optimal weights as a function of hyperparameters. We show that
our technique converges to locally optimal weights and hyperparameters for
sufficiently large hypernetworks. We compare this method to standard
hyperparameter optimization strategies and demonstrate its effectiveness for
tuning thousands of hyperparameters.
@misc{lorraine2018stochastic,
abstract = {Machine learning models are often tuned by nesting optimization of model
weights inside the optimization of hyperparameters. We give a method to
collapse this nested optimization into joint stochastic optimization of weights
and hyperparameters. Our process trains a neural network to output
approximately optimal weights as a function of hyperparameters. We show that
our technique converges to locally optimal weights and hyperparameters for
sufficiently large hypernetworks. We compare this method to standard
hyperparameter optimization strategies and demonstrate its effectiveness for
tuning thousands of hyperparameters.},
added-at = {2018-02-27T08:08:31.000+0100},
author = {Lorraine, Jonathan and Duvenaud, David},
biburl = {https://www.bibsonomy.org/bibtex/23e6f00829df831920a356981d988337c/jk_itwm},
description = {Stochastic Hyperparameter Optimization through Hypernetworks},
interhash = {7be79d20d99ba5af5c670ff33956a151},
intrahash = {3e6f00829df831920a356981d988337c},
keywords = {SGD optimization},
note = {cite arxiv:1802.09419Comment: 9 pages, 6 figures},
timestamp = {2018-02-27T08:08:31.000+0100},
title = {Stochastic Hyperparameter Optimization through Hypernetworks},
url = {http://arxiv.org/abs/1802.09419},
year = 2018
}