Representation Learning via Consistent Assignment of Views over Random Partitions

Thalles Santos Silva, Adín Ramírez Rivera

37th Conference on Neural Information Processing Systems (NeurIPS 2023).

...

Abstract

We present Consistent Assignment of Views over Random Partitions (CARP), a self-supervised clustering method for representation learning of visual features. CARP learns prototypes in an end-to-end online fashion using gradient descent without additional non-differentiable modules to solve the cluster assignment problem. CARP optimizes a new pretext task based on random partitions of prototypes that regularizes the model and enforces consistency between views' assignments. Additionally, our method improves training stability and prevents collapsed solutions in joint-embedding training. Through an extensive evaluation, we demonstrate that CARP's representations are suitable for learning downstream tasks. We evaluate CARP's representations capabilities in 17 datasets across many standard protocols, including linear evaluation, few-shot classification, k-NN, k-means, image retrieval, and copy detection. We compare CARP performance to 11 existing self-supervised methods. We extensively ablate our method and demonstrate that our proposed random partition pretext task improves the quality of the learned representations by devising multiple random classification tasks. In transfer learning tasks, CARP achieves the best performance on average against many SSL methods trained for a longer time.

Short video presentation

TODO

Pre-trained models

Models trained with ResNet50 encoders.

Epochs Multicrop Linear K-NN URL
CARP 100 2x224 + 6x96 72.5 63.5 checkpoints
CARP 200 2x224 + 6x96 74.2 66.5 checkpoints
CARP 400 2x224 73.0 67.6 checkpoints
CARP 400 2x224 + 6x96 75.3 67.7 checkpoints

Important links

Code arXiv NeurIPS 2023 Proceedings OpenReview

Reference


@inproceedings{
    Silva2023,
    title={Representation Learning via Consistent Assignment of Views over Random Partitions},
    author={Silva, Thalles and Ram\'irez Rivera, Ad\'in},
    booktitle={Thirty-seventh Conference on Neural Information Processing Systems ({NeurIPS})},
    year={2023},
    url={https://openreview.net/forum?id=fem6BIJkdv}
}

Authors

Thalles Santos Silva

Adín Ramírez Rivera

Acknowledgements

The computations were performed in part on resources provided by Sigma2---the National Infrastructure for High Performance Computing and Data Storage in Norway---through Project NN8104K. This work was funded in part by the Research Council of Norway, through its Centre for Research-based Innovation funding scheme (grant no. 309439), and Consortium Partners. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior---Brasil (CAPES)---Finance Code 001