Improving the Neural GPU Architecture for Algorithm Learning

TL;DR

This paper enhances the Neural GPU architecture to learn algorithms more efficiently, introducing new techniques that reduce training time and enable end-to-end learning of decimal multiplication.

Contribution

The paper proposes novel improvements to Neural GPU, including hard nonlinearities with saturation costs and diagonal gates, enabling better generalization and decimal multiplication learning.

Findings

Reduced training time for Neural GPU models

Achieved end-to-end decimal multiplication learning

Introduced general techniques applicable to active-memory models

Abstract

Algorithm learning is a core problem in artificial intelligence with significant implications on automation level that can be achieved by machines. Recently deep learning methods are emerging for synthesizing an algorithm from its input-output examples, the most successful being the Neural GPU, capable of learning multiplication. We present several improvements to the Neural GPU that substantially reduces training time and improves generalization. We introduce a new technique - hard nonlinearities with saturation costs- that has general applicability. We also introduce a technique of diagonal gates that can be applied to active-memory models. The proposed architecture is the first capable of learning decimal multiplication end-to-end.