Traditional sequential multi-object attention models rely on a recurrent mechanism to infer object relations. We propose a relational extension (R-SQAIR) of one such attention model (SQAIR) by endowing it with a module with strong relational inductive bias that computes in parallel pairwise interactions between inferred objects. Two recently proposed relational modules are studied on tasks of unsupervised learning from videos. We demonstrate gains over sequential relational mechanisms, also in terms of combinatorial generalization.
Cite
@article{arxiv.1910.05231,
title = {R-SQAIR: Relational Sequential Attend, Infer, Repeat},
author = {Aleksandar Stanić and Jürgen Schmidhuber},
journal= {arXiv preprint arXiv:1910.05231},
year = {2019}
}
Comments
4 page workshop paper accepted at the NeurIPS 2019 Workshop on Perception as Generative Reasoning: Structure, Causality, Probability