Enabling Fine-Tuning of Direct Feedback Alignment via Feedback-Weight Matching
Keywords: direct feedback alignment, deep learning, fine tuning
TL;DR: First study on fine-tuning with DFA (Direct Feedback Alignment)
Abstract: In this paper, we introduce feedback-weight matching, a new method that facilitates reliable fine-tuning of fully connected neural networks using Direct Feedback Alignment (DFA). Although DFA has demonstrated potential by enabling efficient and parallel updates of weight parameters through direct propagation of the network's output error, its usage has been primarily restricted to training networks from scratch. We provide the first analysis showing that existing standard DFA struggles to fine-tune networks pre-trained via back-propagation. Through an analysis of weight alignment (WA) and gradient alignment (GA), we show that the proposed feedback-weight matching enhances DFA's ability and stability in fine-tuning pre-trained networks, providing insights into DFA's behavior and characteristics when applied to fine-tuning. In addition, we find that feedback-weight matching, when combined with weight decay, not only mitigates over-fitting but also further reduces the network output error, leading to improved learning performance during DFA-based fine-tuning. Our experimental results show that, for the first time, feedback-weight matching enables reliable and superior fine-tuning across various fine-tuning tasks compared to existing standard DFA, e.g., achieving 7.97\% accuracy improvement on image classification tasks (i.e., 82.67\% vs. 74.70\%) and 0.66 higher correlation score on NLP tasks (i.e., 0.76 vs. 0.10). The code implementation is available at an anonymous GitHub repository.
Primary Area: transfer learning, meta learning, and lifelong learning
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.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
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: 6057
Loading