Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization

Christopher Liao, Theodoros Tsiligkaridis, Brian Kulis
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:20906-20938, 2023.

Abstract

There is extensive interest in metric learning methods for image retrieval. Many metric learning loss functions focus on learning a correct ranking of training samples, but strongly overfit semantically inconsistent labels and require a large amount of data. To address these shortcomings, we propose a new metric learning method, called contextual loss, which optimizes contextual similarity in addition to cosine similarity. Our contextual loss implicitly enforces semantic consistency among neighbors while converging to the correct ranking. We empirically show that the proposed loss is more robust to label noise, and is less prone to overfitting even when a large portion of train data is withheld. Extensive experiments demonstrate that our method achieves a new state-of-the-art across four image retrieval benchmarks and multiple different evaluation settings. Code is available at: https://github.com/Chris210634/metric-learning-using-contextual-similarity

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-liao23b, title = {Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization}, author = {Liao, Christopher and Tsiligkaridis, Theodoros and Kulis, Brian}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {20906--20938}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/liao23b/liao23b.pdf}, url = {https://proceedings.mlr.press/v202/liao23b.html}, abstract = {There is extensive interest in metric learning methods for image retrieval. Many metric learning loss functions focus on learning a correct ranking of training samples, but strongly overfit semantically inconsistent labels and require a large amount of data. To address these shortcomings, we propose a new metric learning method, called contextual loss, which optimizes contextual similarity in addition to cosine similarity. Our contextual loss implicitly enforces semantic consistency among neighbors while converging to the correct ranking. We empirically show that the proposed loss is more robust to label noise, and is less prone to overfitting even when a large portion of train data is withheld. Extensive experiments demonstrate that our method achieves a new state-of-the-art across four image retrieval benchmarks and multiple different evaluation settings. Code is available at: https://github.com/Chris210634/metric-learning-using-contextual-similarity} }
Endnote
%0 Conference Paper %T Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization %A Christopher Liao %A Theodoros Tsiligkaridis %A Brian Kulis %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-liao23b %I PMLR %P 20906--20938 %U https://proceedings.mlr.press/v202/liao23b.html %V 202 %X There is extensive interest in metric learning methods for image retrieval. Many metric learning loss functions focus on learning a correct ranking of training samples, but strongly overfit semantically inconsistent labels and require a large amount of data. To address these shortcomings, we propose a new metric learning method, called contextual loss, which optimizes contextual similarity in addition to cosine similarity. Our contextual loss implicitly enforces semantic consistency among neighbors while converging to the correct ranking. We empirically show that the proposed loss is more robust to label noise, and is less prone to overfitting even when a large portion of train data is withheld. Extensive experiments demonstrate that our method achieves a new state-of-the-art across four image retrieval benchmarks and multiple different evaluation settings. Code is available at: https://github.com/Chris210634/metric-learning-using-contextual-similarity
APA
Liao, C., Tsiligkaridis, T. & Kulis, B.. (2023). Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:20906-20938 Available from https://proceedings.mlr.press/v202/liao23b.html.

Related Material