Classes for Fast Maximum Entropy Training

TL;DR

This paper introduces a class-based modeling approach that significantly accelerates maximum entropy training for language models, reducing training time by up to 35 times while maintaining comparable perplexity.

Contribution

The paper proposes a novel class-based factorization technique for maximum entropy models, enabling much faster training and broader applicability to large output spaces.

Findings

Achieved up to 35x speedup in training time.

Typically slightly lower perplexities compared to previous methods.

Applicable to other machine learning models with large output spaces.

Abstract

Maximum entropy models are considered by many to be one of the most promising avenues of language modeling research. Unfortunately, long training times make maximum entropy research difficult. We present a novel speedup technique: we change the form of the model to use classes. Our speedup works by creating two maximum entropy models, the first of which predicts the class of each word, and the second of which predicts the word itself. This factoring of the model leads to fewer non-zero indicator functions, and faster normalization, achieving speedups of up to a factor of 35 over one of the best previous techniques. It also results in typically slightly lower perplexities. The same trick can be used to speed training of other machine learning techniques, e.g. neural networks, applied to any problem with a large number of outputs, such as language modeling.