Low-variance Black-box Gradient Estimates for the Plackett-Luce Distribution

Research paper by Artyom Gadetsky, Kirill Struminsky, Christopher Robinson, Novi Quadrianto, Dmitry Vetrov

Indexed on: 25 Nov '19Published on: 22 Nov '19Published in: arXiv - Computer Science - Learning


Learning models with discrete latent variables using stochastic gradient descent remains a challenge due to the high variance of gradient estimates. Modern variance reduction techniques mostly consider categorical distributions and have limited applicability when the number of possible outcomes becomes large. In this work, we consider models with latent permutations and propose control variates for the Plackett-Luce distribution. In particular, the control variates allow us to optimize black-box functions over permutations using stochastic gradient descent. To illustrate the approach, we consider a variety of causal structure learning tasks for continuous and discrete data. We show that our method outperforms competitive relaxation-based optimization methods and is also applicable to non-differentiable score functions.