JEPAMatch paper introduces geometric shaping for semi-supervised learning

Researchers have introduced JEPAMatch, a novel approach to semi-supervised learning that aims to improve model performance when labeled data is scarce. This method moves beyond traditional confidence-based pseudo-labeling by explicitly shaping geometric representations in the latent space, drawing inspiration from the Latent-Euclidean Joint-Embedding Predictive Architectures (LeJEPA) framework. JEPAMatch combines standard semi-supervised loss with a latent-space regularization term, encouraging better-structured representations and faster convergence. Experiments on CIFAR-100, STL-10, and Tiny-ImageNet datasets show that JEPAMatch outperforms existing baselines and significantly reduces computational costs. AI

影响 Introduces a new method to improve model training efficiency and performance in low-data scenarios.

排序理由 This is a research paper introducing a new method for semi-supervised learning.

在 arXiv cs.LG 阅读 →

• CIFAR-100
• FixMatch
• FlexMatch
• JEPAMatch
• LeJEPA
• STL-10
• Tiny-ImageNet

• 论文
• 其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →