Abstract | ||
---|---|---|
We learn recurrent neural network optimizers trained on simple synthetic functions by gradient descent. We show that these learned optimizers exhibit a remarkable degree of transfer in that they can be used to efficiently optimize a broad range of derivative-free black-box functions, including Gaussian process bandits, simple control objectives, global optimization benchmarks and hyper-parameter tuning tasks. Up to the training horizon, the learned optimizers learn to tradeoff exploration and exploitation, and compare favourably with heavily engineered Bayesian optimization packages for hyper-parameter tuning. |
Year | Venue | Field |
---|---|---|
2017 | ICML | Gradient descent,Global optimization,Computer science,Bayesian optimization,Recurrent neural network,Descent direction,Gaussian process,Artificial intelligence,Machine learning,Learning to learn |
DocType | Citations | PageRank |
Conference | 26 | 0.95 |
References | Authors | |
25 | 7 |
Name | Order | Citations | PageRank |
---|---|---|---|
Yutian Chen | 1 | 680 | 36.28 |
Matt Hoffman | 2 | 227 | 14.27 |
Sergio Gomez Colmenarejo | 3 | 42 | 2.43 |
Misha Denil | 4 | 397 | 26.18 |
Timothy P. Lillicrap | 5 | 4377 | 170.65 |
Matthew M Botvinick | 6 | 494 | 25.34 |
Nando De Freitas | 7 | 3284 | 273.68 |