JEPA for RL: Investigating Joint-Embedding Predictive Architectures for Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kenneweg, Tristan, Kenneweg, Philip, Hammer, Barbara
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912343175200768
author Kenneweg, Tristan
Kenneweg, Philip
Hammer, Barbara
author_facet Kenneweg, Tristan
Kenneweg, Philip
Hammer, Barbara
contents Joint-Embedding Predictive Architectures (JEPA) have recently become popular as promising architectures for self-supervised learning. Vision transformers have been trained using JEPA to produce embeddings from images and videos, which have been shown to be highly suitable for downstream tasks like classification and segmentation. In this paper, we show how to adapt the JEPA architecture to reinforcement learning from images. We discuss model collapse, show how to prevent it, and provide exemplary data on the classical Cart Pole task.
format Preprint
id arxiv_https___arxiv_org_abs_2504_16591
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle JEPA for RL: Investigating Joint-Embedding Predictive Architectures for Reinforcement Learning
Kenneweg, Tristan
Kenneweg, Philip
Hammer, Barbara
Computer Vision and Pattern Recognition
Joint-Embedding Predictive Architectures (JEPA) have recently become popular as promising architectures for self-supervised learning. Vision transformers have been trained using JEPA to produce embeddings from images and videos, which have been shown to be highly suitable for downstream tasks like classification and segmentation. In this paper, we show how to adapt the JEPA architecture to reinforcement learning from images. We discuss model collapse, show how to prevent it, and provide exemplary data on the classical Cart Pole task.
title JEPA for RL: Investigating Joint-Embedding Predictive Architectures for Reinforcement Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.16591