JEPA (Joint Embedding Predictive Architecture)

Definition

A neural network architecture that learns representations by predicting abstract representations of future states given past observations and actions, without attempting to reconstruct pixel-level details. Prevents collapse through information maximization or distillation methods. Core innovation: learn representations that eliminate unpredictable noise rather than attempting to predict all details.

Cast within

No topic-region cast yet — this appears once JEPA (Joint Embedding Predictive Architecture)'s compiled claims are aligned into a topic region's argument tree.

Claims mentioning JEPA (Joint Embedding Predictive Architecture) (10)

Your notes

Loading notes...