Multi-Layer Softmaxing during Training Neural Machine Translation for Flexible Decoding with Fewer Layers

TL;DR

This paper introduces a training method for neural machine translation models that allows flexible decoding with fewer layers, reducing computational costs while maintaining translation quality.

Contribution

It proposes a novel multi-layer softmax training procedure that enables dynamic layer selection during decoding, saving space and computation.

Findings

Faster decoding with minimal quality loss

Reduced need for training multiple models

Cost-benefit analysis supports practical advantages

Abstract

This paper proposes a novel procedure for training an encoder-decoder based deep neural network which compresses NxM models into a single model enabling us to dynamically choose the number of encoder and decoder layers for decoding. Usually, the output of the last layer of the N-layer encoder is fed to the M-layer decoder, and the output of the last decoder layer is used to compute softmax loss. Instead, our method computes a single loss consisting of NxM losses: the softmax loss for the output of each of the M decoder layers derived using the output of each of the N encoder layers. A single model trained by our method can be used for decoding with an arbitrary fewer number of encoder and decoder layers. In practical scenarios, this (a) enables faster decoding with insignificant losses in translation quality and (b) alleviates the need to train NxM models, thereby saving space. We take a case study of neural machine translation and show the advantage and give a cost-benefit analysis of our approach.