تقرير
ARSM: Augment-REINFORCE-Swap-Merge Estimator for Gradient Backpropagation Through Categorical Variables
| العنوان: | ARSM: Augment-REINFORCE-Swap-Merge Estimator for Gradient Backpropagation Through Categorical Variables |
|---|---|
| المؤلفون: | Yin, Mingzhang, Yue, Yuguang, Zhou, Mingyuan |
| سنة النشر: | 2019 |
| المجموعة: | Computer Science Statistics |
| مصطلحات موضوعية: | Statistics - Machine Learning, Computer Science - Machine Learning, Statistics - Computation, Statistics - Methodology |
| الوصف: | To address the challenge of backpropagating the gradient through categorical variables, we propose the augment-REINFORCE-swap-merge (ARSM) gradient estimator that is unbiased and has low variance. ARSM first uses variable augmentation, REINFORCE, and Rao-Blackwellization to re-express the gradient as an expectation under the Dirichlet distribution, then uses variable swapping to construct differently expressed but equivalent expectations, and finally shares common random numbers between these expectations to achieve significant variance reduction. Experimental results show ARSM closely resembles the performance of the true gradient for optimization in univariate settings; outperforms existing estimators by a large margin when applied to categorical variational auto-encoders; and provides a "try-and-see self-critic" variance reduction method for discrete-action policy gradient, which removes the need of estimating baselines by generating a random number of pseudo actions and estimating their action-value functions. Comment: Published in ICML 2019. We have updated Section 4.2 and the Appendix to reflect the improvements brought by fixing some bugs hidden in our original code. Please find the Errata in the authors' websites and check the updated code in Github |
| نوع الوثيقة: | Working Paper |
| URL الوصول: | http://arxiv.org/abs/1905.01413 |
| رقم الأكسشن: | edsarx.1905.01413 |
| قاعدة البيانات: | arXiv |
| الوصف غير متاح. |