@inproceedings{fan2020adaptive,
author = {Fan, Xinjie and Zhang, Yizhe and Wang, Zhendong and Zhou, Mingyuan},
title = {Adaptive Correlated Monte Carlo for Contextual Categorical Sequence Generation},
booktitle = {Eighth International Conference on Learning Representations (ICLR)},
year = {2020},
month = {April},
abstract = {Sequence generation models are commonly refined with reinforcement learning over user-defined metrics. However, high gradient variance hinders the practical use of this method. To stabilize this method, we adapt to contextual generation of categorical sequences a policy gradient estimator, which evaluates a set of correlated Monte Carlo (MC) rollouts for variance control. Due to the correlation, the number of unique rollouts is random and adaptive to model uncertainty; those rollouts naturally become baselines for each other, and hence are combined to effectively reduce gradient variance. We also demonstrate the use of correlated MC rollouts for binary-tree softmax models, which reduce the high generation cost in large vocabulary scenarios by decomposing each categorical action into a sequence of binary actions. We evaluate our methods on both neural program synthesis and image captioning. The proposed methods yield lower gradient variance and consistent improvement over related baselines.},
url = {http://approjects.co.za/?big=en-us/research/publication/adaptive-correlated-monte-carlo-for-contextual-categorical-sequence-generation/},
}