Compositional Generalization with Tree Stack Memory Units

TL;DR

This paper introduces Tree Stack Memory Units (Tree-SMU), a recursive neural network with differentiable stack memory, designed to improve compositional generalization in neural networks, especially for mathematical reasoning tasks.

Contribution

The paper proposes Tree-SMU, a novel recursive neural network with stack memory units, enhancing zero-shot compositional generalization over existing models like Transformers and Tree-LSTMs.

Findings

Tree-SMU outperforms baselines on mathematical reasoning benchmarks.

It demonstrates strong generalization on compositionality tests.

Stack memory captures long-range dependencies effectively.

Abstract

We study compositional generalization, viz., the problem of zero-shot generalization to novel compositions of concepts in a domain. Standard neural networks fail to a large extent on compositional learning. We propose Tree Stack Memory Units (Tree-SMU) to enable strong compositional generalization. Tree-SMU is a recursive neural network with Stack Memory Units (\SMU s), a novel memory augmented neural network whose memory has a differentiable stack structure. Each SMU in the tree architecture learns to read from its stack and to write to it by combining the stacks and states of its children through gating. The stack helps capture long-range dependencies in the problem domain, thereby enabling compositional generalization. Additionally, the stack also preserves the ordering of each node's descendants, thereby retaining locality on the tree. We demonstrate strong empirical results on two mathematical reasoning benchmarks. We use four compositionality tests to assess the generalization performance of Tree-SMU and show that it enables accurate compositional generalization compared to strong baselines such as Transformers and Tree-LSTMs.