Backpropagation through the Void: Optimizing control variates for black-box gradient estimation

Will Grathwohl
Will Grathwohl
Yuhuai Wu
Yuhuai Wu

international conference on learning representations, 2018.

Cited by: 123|Views10
EI

Abstract:

Gradient-based optimization is the foundation of deep learning and reinforcement learning. Even when the mechanism being optimized is unknown or not differentiable, optimization using high-variance or biased gradient estimates is still often the best strategy. We introduce a general framework for learning low-variance, unbiased gradient e...More

Code:

Data:

Your rating :
0

 

Tags
Comments