arxivst stuff from arxiv that you should probably bookmark

Differentiable Scheduled Sampling for Credit Assignment

Abstract · Apr 23, 2017 20:05 ·

scheduled training greedy decoding bengio differentiable sampling decisions cs-cl cs-lg cs-ne

Arxiv Abstract

  • Kartik Goyal
  • Chris Dyer
  • Taylor Berg-Kirkpatrick

We demonstrate that a continuous relaxation of the argmax operation can be used to create a differentiable approximation to greedy decoding for sequence-to-sequence (seq2seq) models. By incorporating this approximation into the scheduled sampling training procedure (Bengio et al., 2015)–a well-known technique for correcting exposure bias–we introduce a new training objective that is continuous and differentiable everywhere and that can provide informative gradients near points where previous decoding decisions change their value. In addition, by using a related approximation, we demonstrate a similar approach to sampled-based training. Finally, we show that our approach outperforms cross-entropy training and scheduled sampling procedures in two sequence prediction tasks: named entity recognition and machine translation.

Read the paper (pdf) »