Professor Forcing: A New Algorithm for Training Recurrent Networks
This addresses a key challenge in sequence generation for tasks like language modeling and synthesis, offering a regularization method that improves performance, though it is incremental relative to existing techniques like Teacher Forcing.
The paper tackles the problem of training recurrent networks where the dynamics differ between training and sampling phases, introducing the Professor Forcing algorithm that uses adversarial domain adaptation to align these dynamics. The result shows improved test likelihood on character-level Penn Treebank and sequential MNIST, with qualitative enhancements in sample quality supported by human evaluation.
The Teacher Forcing algorithm trains recurrent networks by supplying observed sequence values as inputs during training and using the network's own one-step-ahead predictions to do multi-step sampling. We introduce the Professor Forcing algorithm, which uses adversarial domain adaptation to encourage the dynamics of the recurrent network to be the same when training the network and when sampling from the network over multiple time steps. We apply Professor Forcing to language modeling, vocal synthesis on raw waveforms, handwriting generation, and image generation. Empirically we find that Professor Forcing acts as a regularizer, improving test likelihood on character level Penn Treebank and sequential MNIST. We also find that the model qualitatively improves samples, especially when sampling for a large number of time steps. This is supported by human evaluation of sample quality. Trade-offs between Professor Forcing and Scheduled Sampling are discussed. We produce T-SNEs showing that Professor Forcing successfully makes the dynamics of the network during training and sampling more similar.