Sparse Relational Reasoning with Object-Centric Representations

TL;DR

This paper explores how increasing sparsity in relational neural networks with object-centric representations affects their performance and interpretability, revealing trade-offs and failure modes.

Contribution

It systematically studies the impact of sparsity constraints on object-centric relational models, highlighting benefits and limitations.

Findings

Sparsity improves model performance and simplifies relations.

Object-centric representations can be detrimental if objects are not fully captured.

Trade-offs exist between interpretability and performance.

Abstract

We investigate the composability of soft-rules learned by relational neural architectures when operating over object-centric (slot-based) representations, under a variety of sparsity-inducing constraints. We find that increasing sparsity, especially on features, improves the performance of some models and leads to simpler relations. Additionally, we observe that object-centric representations can be detrimental when not all objects are fully captured; a failure mode to which CNNs are less prone. These findings demonstrate the trade-offs between interpretability and performance, even for models designed to tackle relational tasks.