Of Notes
JEPA (Joint Embedding Predictive Architecture)
Table of Contents
… my beloved.
It begins with two observations
JEPA learns
These mappings into the
Finally the objective becomes some
Actions
If we have a pair of two steps
We can embed the states
Formally
AutoEncoder Distinctions
The main objective of an auto encoder is to minimize
But JEPA compares the embedding which offloads some of the complexity onto the encoders rather than the comparison. This allows JEPA to ignore details that aren't useful for prediction.