Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2305.19844 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915684977475584 |
|---|---|
| author | Sun, Yi Xu, Xin Li, Jian Hu, Xiaochang Shi, Yifei Zeng, Ling-Li |
| author_facet | Sun, Yi Xu, Xin Li, Jian Hu, Xiaochang Shi, Yifei Zeng, Ling-Li |
| contents | Multi-output deep neural networks(MONs) contain multiple task branches, and these tasks usually share partial network filters that lead to the entanglement of different task inference routes. Due to the inconsistent optimization objectives, the task gradients used for training MONs will interfere with each other on the shared routes, which will decrease the overall model performance. To address this issue, we propose a novel gradient de-conflict algorithm named DR-MGF(Dynamic Routes and Meta-weighted Gradient Fusion) in this work. Different from existing de-conflict methods, DR-MGF achieves gradient de-conflict in MONs by learning task-preferred inference routes. The proposed method is motivated by our experimental findings: the shared filters are not equally important to different tasks. By designing the learnable task-specific importance variables, DR-MGF evaluates the importance of filters for different tasks. Through making the dominances of tasks over filters be proportional to the task-specific importance of filters, DR-MGF can effectively reduce the inter-task interference. The task-specific importance variables ultimately determine task-preferred inference routes at the end of training iterations. Extensive experimental results on CIFAR, ImageNet, and NYUv2 illustrate that DR-MGF outperforms the existing de-conflict methods both in prediction accuracy and convergence speed of MONs. Furthermore, DR-MGF can be extended to general MONs without modifying the overall network structures. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2305_19844 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Learning Task-preferred Inference Routes for Gradient De-conflict in Multi-output DNNs Sun, Yi Xu, Xin Li, Jian Hu, Xiaochang Shi, Yifei Zeng, Ling-Li Computer Vision and Pattern Recognition Multi-output deep neural networks(MONs) contain multiple task branches, and these tasks usually share partial network filters that lead to the entanglement of different task inference routes. Due to the inconsistent optimization objectives, the task gradients used for training MONs will interfere with each other on the shared routes, which will decrease the overall model performance. To address this issue, we propose a novel gradient de-conflict algorithm named DR-MGF(Dynamic Routes and Meta-weighted Gradient Fusion) in this work. Different from existing de-conflict methods, DR-MGF achieves gradient de-conflict in MONs by learning task-preferred inference routes. The proposed method is motivated by our experimental findings: the shared filters are not equally important to different tasks. By designing the learnable task-specific importance variables, DR-MGF evaluates the importance of filters for different tasks. Through making the dominances of tasks over filters be proportional to the task-specific importance of filters, DR-MGF can effectively reduce the inter-task interference. The task-specific importance variables ultimately determine task-preferred inference routes at the end of training iterations. Extensive experimental results on CIFAR, ImageNet, and NYUv2 illustrate that DR-MGF outperforms the existing de-conflict methods both in prediction accuracy and convergence speed of MONs. Furthermore, DR-MGF can be extended to general MONs without modifying the overall network structures. |
| title | Learning Task-preferred Inference Routes for Gradient De-conflict in Multi-output DNNs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2305.19844 |