Learn how to permute a set, then encode permuted set with RNN to obtain a set representation.

[arxiv] [code] [poster]

Representations of sets are challenging to learn because operations on sets should be permutation-invariant. To this end, we propose a Permutation-Optimisation module that learns how to permute a set end-to-end. The permuted set can be further processed to learn a permutation-invariant representation of that set, avoiding a bottleneck in traditional set models. We demonstrate our model's ability to learn permutations and set representations with either explicit or implicit supervision on four datasets, on which we achieve state-of-the-art results: number sorting, image mosaics, classification from image mosaics, and visual question answering.

    title         = {Learning Representations of Sets through Optimized Permutations},
    author        = {Yan Zhang and Jonathon Hare and Adam Pr\"ugel-Bennett},
    booktitle     = {International Conference on Learning Representations},
    year          = {2019},
    url           = {https://openreview.net/forum?id=HJMCcjAcYX},