Skip to content

Latest commit

 

History

History
13 lines (8 loc) · 848 Bytes

File metadata and controls

13 lines (8 loc) · 848 Bytes

gradient-estimators-in-stochastic-computation-graphs

Gradient Estimation in Stochastic Computation Graphs using TensorFlow.

In this IPython notebook, we aim at exploring different ways to estimate gradients in stochastic computation graphs in order to optimize an objective function defined by an expectation over a set of random variables.

In terms of implementation, we'll use the TensorFlow library, mainly because of its automatic reverse-mode differentiation capabilities.

The notation and theory are based on the following NIPS paper:

Schulman, J., Heess, N., Weber, T. and Abbeel, P., 2015.
Gradient estimation using stochastic computation graphs. In Advances in Neural Information Processing Systems (pp. 3528-3536).