Supercm: Revisiting Clustering for Semi-Supervised Learning

TL;DR

Supercm introduces a simple, end-to-end deep semi-supervised learning method that explicitly uses clustering to improve performance, complementing existing SSL techniques.

Contribution

It extends a differentiable clustering module to incorporate labeled data, creating a novel, effective SSL approach with straightforward training.

Findings

Improves over supervised-only baseline

Can enhance other SSL methods

Demonstrates effectiveness on benchmark datasets

Abstract

The development of semi-supervised learning (SSL) has in recent years largely focused on the development of new consistency regularization or entropy minimization approaches, often resulting in models with complex training strategies to obtain the desired results. In this work, we instead propose a novel approach that explicitly incorporates the underlying clustering assumption in SSL through extending a recently proposed differentiable clustering module. Leveraging annotated data to guide the cluster centroids results in a simple end-to-end trainable deep SSL approach. We demonstrate that the proposed model improves the performance over the supervised-only baseline and show that our framework can be used in conjunction with other SSL methods to further boost their performance.