SEArch: A Self-Evolving Framework for Network Architecture Optimization

Yongqing Liang, Dawei Xiang, Xin Li

Primary Area: optimization

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: network architecture optimization, network pruning, knowledge distillation

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: This paper studies a fundamental network optimization problem that finds a network architecture with optimal performance (low losses) under given resource budgets (small parameter size and/or fast inference). Different from existing network optimization approaches such as network pruning, knowledge distillation (KD), and network architecture search (NAS), in this work we introduce a novel self-evolving pipeline to perform network optimization. In this framework, a simple network iteratively and adaptively modifies its structures by using the guidance from the teacher network, until it reaches the resource budget. An attention module is introduced to transfer the knowledge from teacher network to student network. The splitting edge scheme helps the student model find an optimal macro architecture. The proposed framework combines the advantages of pruning, KD, and NAS, and hence, can efficiently generate networks with flexible structure and desirable performance. Extensive experiments on CIFAR-10, CIFAR-100 and ImageNet demonstrated that our framework achieves state-of-the-art performance in this network architecture optimization task.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8353