Improving the Neural GPU Architecture for Algorithm Learning

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.

Minor edits

References in corpus (1)

• Extensions and Limitations of the Neural GPU

Cited by in corpus (12)

• Compositional Generalization in Semantic Parsing: Pre-training vs. Specialized Architectures
• On the Turing Completeness of Modern Neural Network Architectures
• Attending to Mathematical Language with Transformers
• Neural Arithmetic Units
• Neural Shuffle-Exchange Networks -- Sequence Processing in O(n log n) Time
• Measuring Arithmetic Extrapolation Performance
• Image segmentation via Cellular Automata
• Neural Arithmetic Expression Calculator
• Neural Status Registers
• Is Attention All What You Need? -- An Empirical Investigation on Convolution-Based Active Memory and Self-Attention
• A Primer for Neural Arithmetic Logic Modules
• Progress Extrapolating Algorithmic Learning to Arbitrary Sequence Lengths