On the Variance of Unbiased Online Recurrent Optimization

Abstract

The recently proposed Unbiased Online Recurrent Optimization algorithm (UORO, arXiv:1702.05043) uses an unbiased approximation of RTRL to achieve fully online gradient-based learning in RNNs. In this work we analyze the variance of the gradient estimate computed by UORO, and propose several possible changes to the method which reduce this variance both in theory and practice. We also contribute significantly to the theoretical and intuitive understanding of UORO (and its existing variance reduction technique), and demonstrate a fundamental connection between its gradient estimate and the one that would be computed by REINFORCE if small amounts of noise were added to the RNN's hidden units.

References in corpus (2)

• A Simple Way to Initialize Recurrent Networks of Rectified Linear Units
• A review of matrix scaling and Sinkhorn's normal form for matrices and positive maps

Cited by in corpus (4)

• A Unified Framework of Online Learning Algorithms for Training Recurrent Neural Networks
• Online hyperparameter optimization by real-time recurrent learning
• Decoupling Hierarchical Recurrent Neural Networks With Locally Computable Losses
• Rapid training of deep neural networks without skip connections or normalization layers using Deep Kernel Shaping