SynDaCaTE: A Synthetic Dataset For Evaluating Part-Whole Hierarchical Inference

TL;DR

This paper introduces SynDaCaTE, a synthetic dataset designed to evaluate part-whole hierarchy inference in models, revealing limitations in existing capsule networks and highlighting the effectiveness of self-attention mechanisms.

Contribution

The paper presents SynDaCaTE, a synthetic dataset for evaluating part-whole inference, and demonstrates its utility in analyzing capsule models and self-attention methods.

Findings

Capsule models have a specific bottleneck in part-whole inference.

Permutation-equivariant self-attention effectively infers parts-to-wholes.

SynDaCaTE enables precise evaluation of hierarchical inference capabilities.

Abstract

Learning to infer object representations, and in particular part-whole hierarchies, has been the focus of extensive research in computer vision, in pursuit of improving data efficiency, systematic generalisation, and robustness. Models which are \emph{designed} to infer part-whole hierarchies, often referred to as capsule networks, are typically trained end-to-end on supervised tasks such as object classification, in which case it is difficult to evaluate whether such a model \emph{actually} learns to infer part-whole hierarchies, as claimed. To address this difficulty, we present a SYNthetic DAtaset for CApsule Testing and Evaluation, abbreviated as SynDaCaTE, and establish its utility by (1) demonstrating the precise bottleneck in a prominent existing capsule model, and (2) demonstrating that permutation-equivariant self-attention is highly effective for parts-to-wholes inference, which motivates future directions for designing effective inductive biases for computer vision.