Sparse Relational Reasoning with Object-Centric Representations

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.

Recorded Talk on SlidesLive

Download paper here

Recommended citation: Spies, A.F., Russo, A., Shanahan, M. (2022). Sparse Relational Reasoning with Object-Centric Representations. ICML 2022 DyNN Workshop.

@article{spies2022SparseRelational,
  title = {Sparse Relational Reasoning with Object-Centric Representations},
  author = {Spies, Alex F. and Russo, Alessandra and Shanahan, Murray},
  doi = {10.48550/ARXIV.2207.07512},
  url = {https://arxiv.org/abs/2207.07512},
  publisher = {ICML DyNN Workshop},
  year = {2022},
}