Reproduce “Correlation Clustering with Active Learning of Pairwise Similarities” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Linus Aronsson, Morteza Haghir Chehreghani
A generic active learning framework for correlation clustering is introduced, allowing for cost-efficient acquisition of real-valued, noisy pairwise similarities. The framework, featuring novel inconsistency-based query strategies, achieves near-perfect Adjusted Rand Index scores even with high noise levels while outperforming existing baselines.
Key insights
- Decoupling the query selection process from the underlying correlation clustering algorithm significantly enhances the framework's generality and adaptability.
- Averaging multiple queries for the same pairwise similarity effectively mitigates the impact of non-persistent noise from the oracle, enhancing robustness.
- Leveraging 'bad triangles' (inconsistent triplets of pairwise similarities) provides highly informative query targets, leading to more efficient learning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.