JEPAMatch: Geometric Representation Shaping for Semi-Supervised Learning

arXiv cs.LG / 4/24/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

JEPAMatch is a new semi-supervised learning approach aimed at improving FixMatch-style methods that can be biased by majority classes and degraded by incorrect early pseudo-labels.
The method shifts from relying primarily on confidence thresholding to explicitly shaping the geometry of latent representations using a latent-space regularization inspired by LeJEPA’s isotropic Gaussian structure assumption.
It combines a FlexMatch-like semi-supervised loss (an adaptive extension of FixMatch) with the proposed LeJEPA-derived regularization term to promote clearer decision boundaries.
Experiments on CIFAR-100, STL-10, and Tiny-ImageNet show consistent gains over existing baselines and faster convergence with substantially reduced overall compute versus standard FixMatch-style pipelines.
Overall, the paper argues that imposing geometric structure in latent space can preserve the benefits of pseudo-labeling while mitigating bias and training inefficiency.

Abstract

Semi-supervised learning has emerged as a powerful paradigm for leveraging large amounts of unlabeled data to improve the performance of machine learning models when labeled data are scarce. Among existing approaches, methods derived from FixMatch have achieved state-of-the-art results in image classification by combining weak and strong data augmentations with confidence-based pseudo-labeling. Despite their strong empirical performance, these methods typically struggle with two critical bottlenecks: majority classes tend to dominate the learning process, which is amplified by incorrect pseudo-labels, leading to biased models. Furthermore, noisy early pseudo-labels prevent the model from forming clear decision boundaries, requiring prolonged training to learn informative representation. In this paper, we introduce a paradigm shift from conventional logical output threshold base, toward an explicit shaping of geometric representations. Our approach is inspired by the recently proposed Latent-Euclidean Joint-Embedding Predictive Architectures (LeJEPA), a theoretically grounded framework asserting that meaningful representations should exhibit an isotropic Gaussian structure in latent space. Building on this principle, we propose a new training objective that combines the classical semi-supervised loss used in FlexMatch, an adaptive extension of FixMatch, with a latent-space regularization term derived from LeJEPA. Our proposed approach, encourages well-structured representations while preserving the advantages of pseudo-labeling strategies. Through extensive experiments on CIFAR-100, STL-10 and Tiny-ImageNet, we demonstrate that the proposed method consistently outperforms existing baselines. In addition, our method significantly accelerates the convergence, drastically reducing the overall computational cost compared to standard FixMatch-based pipelines.