Addressing the Loss-Metric Mismatch with Adaptive Loss Alignment

Chen Huang, Shuangfei Zhai, Walter Talbott, Miguel Bautista Martin, Shih-Yu Sun, Carlos Guestrin, Josh Susskind
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:2891-2900, 2019.

Abstract

In most machine learning training paradigms a fixed, often handcrafted, loss function is assumed to be a good proxy for an underlying evaluation metric. In this work we assess this assumption by meta-learning an adaptive loss function to directly optimize the evaluation metric. We propose a sample efficient reinforcement learning approach for adapting the loss dynamically during training. We empirically show how this formulation improves performance by simultaneously optimizing the evaluation metric and smoothing the loss landscape. We verify our method in metric learning and classification scenarios, showing considerable improvements over the state-of-the-art on a diverse set of tasks. Importantly, our method is applicable to a wide range of loss functions and evaluation metrics. Furthermore, the learned policies are transferable across tasks and data, demonstrating the versatility of the method.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-huang19f, title = {Addressing the Loss-Metric Mismatch with Adaptive Loss Alignment}, author = {Huang, Chen and Zhai, Shuangfei and Talbott, Walter and Martin, Miguel Bautista and Sun, Shih-Yu and Guestrin, Carlos and Susskind, Josh}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {2891--2900}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/huang19f/huang19f.pdf}, url = {https://proceedings.mlr.press/v97/huang19f.html}, abstract = {In most machine learning training paradigms a fixed, often handcrafted, loss function is assumed to be a good proxy for an underlying evaluation metric. In this work we assess this assumption by meta-learning an adaptive loss function to directly optimize the evaluation metric. We propose a sample efficient reinforcement learning approach for adapting the loss dynamically during training. We empirically show how this formulation improves performance by simultaneously optimizing the evaluation metric and smoothing the loss landscape. We verify our method in metric learning and classification scenarios, showing considerable improvements over the state-of-the-art on a diverse set of tasks. Importantly, our method is applicable to a wide range of loss functions and evaluation metrics. Furthermore, the learned policies are transferable across tasks and data, demonstrating the versatility of the method.} }
Endnote
%0 Conference Paper %T Addressing the Loss-Metric Mismatch with Adaptive Loss Alignment %A Chen Huang %A Shuangfei Zhai %A Walter Talbott %A Miguel Bautista Martin %A Shih-Yu Sun %A Carlos Guestrin %A Josh Susskind %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-huang19f %I PMLR %P 2891--2900 %U https://proceedings.mlr.press/v97/huang19f.html %V 97 %X In most machine learning training paradigms a fixed, often handcrafted, loss function is assumed to be a good proxy for an underlying evaluation metric. In this work we assess this assumption by meta-learning an adaptive loss function to directly optimize the evaluation metric. We propose a sample efficient reinforcement learning approach for adapting the loss dynamically during training. We empirically show how this formulation improves performance by simultaneously optimizing the evaluation metric and smoothing the loss landscape. We verify our method in metric learning and classification scenarios, showing considerable improvements over the state-of-the-art on a diverse set of tasks. Importantly, our method is applicable to a wide range of loss functions and evaluation metrics. Furthermore, the learned policies are transferable across tasks and data, demonstrating the versatility of the method.
APA
Huang, C., Zhai, S., Talbott, W., Martin, M.B., Sun, S., Guestrin, C. & Susskind, J.. (2019). Addressing the Loss-Metric Mismatch with Adaptive Loss Alignment. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:2891-2900 Available from https://proceedings.mlr.press/v97/huang19f.html.

Related Material