Distributed Bilevel Optimization with Dual Pruning for Resource-limited Clients
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918266651279360 |
|---|---|
| author | Li, Mingyi Zhang, Xiao Zheng, Ruisheng Shi, Hongjian Yuan, Yuan Cheng, Xiuzhen Yu, Dongxiao |
| author_facet | Li, Mingyi Zhang, Xiao Zheng, Ruisheng Shi, Hongjian Yuan, Yuan Cheng, Xiuzhen Yu, Dongxiao |
| contents | With the development of large-scale models, traditional distributed bilevel optimization algorithms cannot be applied directly in low-resource clients. The key reason lies in the excessive computation involved in optimizing both the lower- and upper-level functions. Thus, we present the first resource-adaptive distributed bilevel optimization framework with a second-order free hypergradient estimator, which allows each client to optimize the submodels adapted to the available resources. Due to the coupled influence of partial outer parameters x and inner parameters y, it's challenging to theoretically analyze the upper bound regarding the globally averaged hypergradient for full model parameters. The error bound of inner parameter also needs to be reformulated since the local partial training. The provable theorems show that both RABO and RAFBO can achieve an asymptotically optimal convergence rate of $O(1/\sqrt{C_x^{\ast}Q})$, which is dominated by the minimum coverage of the outer parameter $C_x^{\ast}$. Extensive experiments on two different tasks demonstrate the effectiveness and computation efficiency of our proposed methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_24667 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Distributed Bilevel Optimization with Dual Pruning for Resource-limited Clients Li, Mingyi Zhang, Xiao Zheng, Ruisheng Shi, Hongjian Yuan, Yuan Cheng, Xiuzhen Yu, Dongxiao Distributed, Parallel, and Cluster Computing With the development of large-scale models, traditional distributed bilevel optimization algorithms cannot be applied directly in low-resource clients. The key reason lies in the excessive computation involved in optimizing both the lower- and upper-level functions. Thus, we present the first resource-adaptive distributed bilevel optimization framework with a second-order free hypergradient estimator, which allows each client to optimize the submodels adapted to the available resources. Due to the coupled influence of partial outer parameters x and inner parameters y, it's challenging to theoretically analyze the upper bound regarding the globally averaged hypergradient for full model parameters. The error bound of inner parameter also needs to be reformulated since the local partial training. The provable theorems show that both RABO and RAFBO can achieve an asymptotically optimal convergence rate of $O(1/\sqrt{C_x^{\ast}Q})$, which is dominated by the minimum coverage of the outer parameter $C_x^{\ast}$. Extensive experiments on two different tasks demonstrate the effectiveness and computation efficiency of our proposed methods. |
| title | Distributed Bilevel Optimization with Dual Pruning for Resource-limited Clients |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2512.24667 |