Reproduce “Learning to Affiliate: Mutual Centralized Learning for Few-shot Classification” 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
Yang Liu, Weifeng Zhang, Chao Xiang, Tu Zheng, Deng Cai, Xiaofei He
The Mutual Centralized Learning (MCL) framework enables few-shot classification by modeling mutual affiliations between dense feature sets via bidirectional random walks and graph centrality. It achieved competitive performance on standard benchmarks like miniImageNet and tieredImageNet, and improved existing global feature-based methods by up to 4.5% on miniImageNet.
Key insights
- Bidirectional modeling of mutual affiliations between query and support dense feature sets provides a more comprehensive understanding of their relationships than unidirectional approaches.
- Few-shot classification can be formally linked to learning single-mode eigenvector centrality on a bipartite graph (affiliation network) formed by query and support features.
- Traditional transductive methods like label propagation can be adapted for inductive FSL by treating dense features of a single query image as an unlabeled dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.