Better set representations for relational reasoning

Set-structured latent spaces improve generalisation and robustness.

[arxiv] [code] [video]

@inproceedings{
    huang2020srn,
    author        = {Qian Huang and Horace He and Abhay Singh and Yan Zhang and Ser-Nam Lim and Austin Benson},
    title         = {Better Set Representations For Relational Reasoning},
    booktitle     = {Advances in Neural Information Processing Systems},
    year          = {2020},
    eprint        = {2003.04448},
    url           = {https://arxiv.org/abs/2003.04448},
}

Deep set prediction networks

To predict a set from a vector, use gradient descent to find a set the encodes to that vector.

[arxiv] [code] [poster 1] [poster 2]

@inproceedings{
    zhang2019dspn,
    author        = {Yan Zhang and Jonathon Hare and Adam Pr\"ugel-Bennett},
    title         = {{Deep Set Prediction Networks}},
    booktitle     = {Advances in Neural Information Processing Systems},
    year          = {2019},
    eprint        = {1906.06565},
    url           = {https://arxiv.org/abs/1906.06565},
}

FSPool: Learning set representations with featurewise sort pooling

Sort in encoder and undo sorting in decoder to avoid responsibility problem in set auto-encoders.

[arxiv] [code] [video] [poster]

@inproceedings{
    zhang2019fspool,
    author        = {Yan Zhang and Jonathon Hare and Adam Pr\"ugel-Bennett},
    title         = {{FSPool}: Learning Set Representations with Featurewise Sort Pooling},
    booktitle     = {International Conference on Learning Representations},
    year          = {2020},
    eprint        = {1906.02795},
    url           = {https://openreview.net/forum?id=HJgBA2VYwH}
}

Learning representations of sets through optimized permutations

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

[arxiv] [code] [poster]

@inproceedings{
    zhang2018permoptim,
    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},
}

Learning to count objects in natural images

Enabling visual question answering models to count by handling overlapping object proposals.

[arxiv] [code] [poster]

@inproceedings{
    zhang2018vqacounting,
    title         = {Learning to Count Objects in Natural Images for Visual Question Answering},
    author        = {Yan Zhang and Jonathon Hare and Adam Pr\"ugel-Bennett},
    booktitle     = {International Conference on Learning Representations},
    year          = {2018},
    eprint        = {1802.05766},
    url           = {https://openreview.net/forum?id=B12Js_yRb},
}