Unbiasing Truncated Backpropagation Through Time
This addresses a bias issue in training recurrent neural networks, offering a more reliable method for researchers and practitioners, though it is incremental as it builds on existing truncated BPTT.
The paper tackled the bias in truncated Backpropagation Through Time (BPTT), which favors short-term dependencies and lacks convergence guarantees, by introducing Anticipated Reweighted Truncated Backpropagation (ARTBP) to provide unbiasedness while maintaining computational efficiency. The result showed ARTBP converges reliably on a synthetic task where truncated BPTT diverged, and it slightly outperforms truncated BPTT on Penn Treebank character-level language modeling.
Truncated Backpropagation Through Time (truncated BPTT) is a widespread method for learning recurrent computational graphs. Truncated BPTT keeps the computational benefits of Backpropagation Through Time (BPTT) while relieving the need for a complete backtrack through the whole data sequence at every step. However, truncation favors short-term dependencies: the gradient estimate of truncated BPTT is biased, so that it does not benefit from the convergence guarantees from stochastic gradient theory. We introduce Anticipated Reweighted Truncated Backpropagation (ARTBP), an algorithm that keeps the computational benefits of truncated BPTT, while providing unbiasedness. ARTBP works by using variable truncation lengths together with carefully chosen compensation factors in the backpropagation equation. We check the viability of ARTBP on two tasks. First, a simple synthetic task where careful balancing of temporal dependencies at different scales is needed: truncated BPTT displays unreliable performance, and in worst case scenarios, divergence, while ARTBP converges reliably. Second, on Penn Treebank character-level language modelling, ARTBP slightly outperforms truncated BPTT.