Few-Example Clustering via Contrastive Learning

TL;DR

This paper introduces Few-Example Clustering (FEC), a contrastive learning-based algorithm that effectively clusters few examples by generating and selecting the best candidate cluster assignments, demonstrating superior performance on image datasets.

Contribution

The paper presents a novel FEC algorithm that combines candidate generation, contrastive learning, and early-stage loss-based selection for few-example clustering.

Findings

FEC outperforms baseline methods by about 3.2% on mini-ImageNet and CUB-200-2011.

FEC shows a characteristic learning curve with gradual improvement followed by a sharp drop.

Contrastive learning with early loss minimization effectively identifies true cluster assignments.

Abstract

We propose Few-Example Clustering (FEC), a novel algorithm that performs contrastive learning to cluster few examples. Our method is composed of the following three steps: (1) generation of candidate cluster assignments, (2) contrastive learning for each cluster assignment, and (3) selection of the best candidate. Based on the hypothesis that the contrastive learner with the ground-truth cluster assignment is trained faster than the others, we choose the candidate with the smallest training loss in the early stage of learning in step (3). Extensive experiments on the \textit{mini}-ImageNet and CUB-200-2011 datasets show that FEC outperforms other baselines by about 3.2% on average under various scenarios. FEC also exhibits an interesting learning curve where clustering performance gradually increases and then sharply drops.