%0 Journal Article %1 du2018adapting %A Du, Yunshu %A Czarnecki, Wojciech M. %A Jayakumar, Siddhant M. %A Pascanu, Razvan %A Lakshminarayanan, Balaji %D 2018 %K objectives optimization %T Adapting Auxiliary Losses Using Gradient Similarity %U http://arxiv.org/abs/1812.02224 %X One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations. However, it is not always trivial to know if an auxiliary task will be helpful for the main task and when it could start hurting. We propose to use the cosine similarity between gradients of tasks as an adaptive weight to detect when an auxiliary loss is helpful to the main loss. We show that our approach is guaranteed to converge to critical points of the main task and demonstrate the practical usefulness of the proposed algorithm in a few domains: multi-task supervised learning on subsets of ImageNet, reinforcement learning on gridworld, and reinforcement learning on Atari games.

@article{du2018adapting, abstract = {One approach to deal with the statistical inefficiency of neural networks is to rely on auxiliary losses that help to build useful representations. However, it is not always trivial to know if an auxiliary task will be helpful for the main task and when it could start hurting. We propose to use the cosine similarity between gradients of tasks as an adaptive weight to detect when an auxiliary loss is helpful to the main loss. We show that our approach is guaranteed to converge to critical points of the main task and demonstrate the practical usefulness of the proposed algorithm in a few domains: multi-task supervised learning on subsets of ImageNet, reinforcement learning on gridworld, and reinforcement learning on Atari games.}, added-at = {2019-12-11T05:53:36.000+0100}, author = {Du, Yunshu and Czarnecki, Wojciech M. and Jayakumar, Siddhant M. and Pascanu, Razvan and Lakshminarayanan, Balaji}, biburl = {https://www.bibsonomy.org/bibtex/2f4711c500c51d563d804eb976dc0e75b/kirk86}, description = {[1812.02224] Adapting Auxiliary Losses Using Gradient Similarity}, interhash = {aec2488e688eba5dd8077e2b47544046}, intrahash = {f4711c500c51d563d804eb976dc0e75b}, keywords = {objectives optimization}, note = {cite arxiv:1812.02224}, timestamp = {2019-12-11T05:53:59.000+0100}, title = {Adapting Auxiliary Losses Using Gradient Similarity}, url = {http://arxiv.org/abs/1812.02224}, year = 2018 }