Data Distributional Properties As Inductive Bias for Systematic Generalization

TL;DR

This paper investigates how specific data distributional properties like diversity, burstiness, and latent intervention serve as inductive biases to significantly improve systematic generalization in multi-modal language models, revealing the role of NMI and representation geometry.

Contribution

It introduces and empirically evaluates three data properties as inductive biases for systematic generalization, highlighting the importance of NMI and representation geometry in out-of-distribution performance.

Findings

All three data properties significantly enhance systematic generalization.

Diversity increases accuracy by 89% in affected properties.

Lower NMI correlates with better out-of-distribution generalization.

Abstract

Deep neural networks (DNNs) struggle at systematic generalization (SG). Several studies have evaluated the possibility to promote SG through the proposal of novel architectures, loss functions or training methodologies. Few studies, however, have focused on the role of training data properties in promoting SG. In this work, we investigate the impact of certain data distributional properties, as inductive biases for the SG ability of a multi-modal language model. To this end, we study three different properties. First, data diversity, instantiated as an increase in the possible values a latent property in the training distribution may take. Second, burstiness, where we probabilistically restrict the number of possible values of latent factors on particular inputs during training. Third, latent intervention, where a particular latent factor is altered randomly during training. We find that all three factors significantly enhance SG, with diversity contributing an 89% absolute increase in accuracy in the most affected property. Through a series of experiments, we test various hypotheses to understand why these properties promote SG. Finally, we find that Normalized Mutual Information (NMI) between latent attributes in the training distribution is strongly predictive of out-of-distribution generalization. We find that a mechanism by which lower NMI induces SG is in the geometry of representations. In particular, we find that NMI induces more parallelism in neural representations (i.e., input features coded in parallel neural vectors) of the model, a property related to the capacity of reasoning by analogy.