Reproduce “Graph Contrastive Clustering” 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
Huasong Zhong, Jianlong Wu, Chong Chen, Jianqiang Huang, Minghua Deng, Liqiang Nie, Zhouchen Lin, Xian-Sheng Hua
Graph Contrastive Clustering (GCC) advances deep clustering by explicitly modeling cluster-level relationships through graph-based contrastive learning. The method learns representations inherently more suitable for clustering and produces more compact cluster assignments, outperforming prior state-of-the-art methods by achieving an accuracy of 0.856 on CIFAR-10, a 12.9% improvement over the previous best.
Key insights
- Effective clustering requires understanding relationships between multiple instances within the same latent category, extending beyond instance-level consistency.
- Graph-based contrastive learning can explicitly model cluster-level relationships, leading to representations inherently more suitable for clustering tasks.
- Applying graph contrastive principles to both feature representation and cluster assignment spaces simultaneously yields improved and more compact cluster assignments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.