Hugging Face Papers ↗ · 2026-10-26
UMM-JEPA: Generative Pretraining via Masked Next-Scale Representation Prediction for Unified Multimodal Models
Media reports Hugging Face Papers Source language: Chinese
论文摘要(作者报告,未经本站复测):Unified multimodal models can support both visual understanding and generation, yet generative training does not necessarily improve understanding.
Full text
Reformulating generation as progressive prediction of continuous perceptual representations improves visual understanding in unified multimodal models more than discrete-code or latent-flow objectives.