Collaborative Data Optimization

Xinyi Shang, Peng Sun, Fengyuan Liu, Tao Lin

Keywords: Unlabeled Data, Data Optimization, Efficiency

TL;DR: We propose a highly efficient, parallelized framework designed for collaborative data optimization, demonstrating the effectiveness and efficiency across various datasets and architectures.

Abstract:

Training efficiency plays a pivotal role in deep learning.
This paper begins by analyzing current methods for enhancing efficiency, highlighting the necessity of optimizing targets, a process we define as data optimization.
Subsequently, we reveal that current data optimization methods incur significant additional costs, e.g., human resources or computational overhead, due to their inherently sequential optimization process.
To address these issues, we propose CoOpt, a highly efficient, parallelized framework designed for collaborative data optimization.
CoOpt enables participants to independently optimize data subsets, ensuring that the overall performance, once these subsets are collected, remains comparable to the sequential optimization of the entire dataset, thus significantly reducing optimization costs for individual participants.
Extensive experiments have been conducted on various real-world scenarios to demonstrate the effectiveness and efficiency of CoOpt across various datasets and architectures.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation 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.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 10390