Learning-Augmented Streaming Algorithms for Correlation Clustering
Keywords: Correlation Clustering, graph streaming algorithms, learning-augmented algorithms
TL;DR: We introduce the first learning-augmented streaming algorithms for Correlation Clustering, achieving the first better-than-$3$-approximation in dynamic streams.
Abstract: We study streaming algorithms for Correlation Clustering. Given a complete graph as an arbitrary-order stream of edges, with each edge labelled as positive or negative, the goal is to partition the vertices into disjoint clusters, such that the number of disagreements is minimized. In this paper, we introduce the first learning-augmented streaming algorithms for the problem, achieving the first better-than-$3$-approximation in dynamic streams. Our algorithms draw inspiration from recent works of Cambus et al. (SODA'24), and Chakrabarty and Makarychev (NeurIPS'23). Our algorithms use the predictions of pairwise dissimilarities between vertices provided by a predictor and achieve an approximation ratio that is close to $2.06$ under good prediction quality. Even if the prediction quality is poor, our algorithms cannot perform worse than the well known Pivot algorithm, which achieves a $3$-approximation. Our algorithms are much simpler than the recent $1.847$-approximation streaming algorithm by Cohen-Addad et al. (STOC'24) which appears to be challenging to implement and is restricted to insertion-only streams. Experimental results on synthetic and real-world datasets demonstrate the superiority of our proposed algorithms over their non-learning counterparts.
Supplementary Material: zip
Primary Area: learning theory
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: 13914
Loading