Of Notes

JEPA (Joint Embedding Predictive Architecture)

1 min read · revised 25 September 2026 · referred to by 4 notes · ·

Table of Contents

… my beloved.

It begins with two observations and where the first is the information we are always allowed to see and the latter is something related that we want to predict.

JEPA learns while learning a mapping while trying to minimize the between .

These mappings into the denoted latent space are usually done with some embedding method.

Finally the objective becomes some distance metric based minimization between and which might look like:

Actions

If we have a pair of two steps and some action (this is very similar to the Markov process)

We can embed the states and condition also on some such that .

Formally .

AutoEncoder Distinctions

The main objective of an auto encoder is to minimize which forces a lot of pressure on the embedding space to preserve information to reconstruct the observation data … how can we formalize this information mathematically?

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.

/ or Ctrl K
search; empty, it shows notes to revise and lately read
m
mark this note to revise, or unmark it
c
cover maths, proofs and solutions to test yourself
n
uncover the next covered item
y
copy this note’s org-roam link
e
copy this note as markdown, LaTeX kept as TeX
p
the next problem to work; there h hints, s shows the solution, x passes it for today
→ ← j k space
turn the page, when reading in pages
l
on the map, trace the greatest distance, then the longest path
double-click
a formula, to copy its TeX
rest on maths
bind a part of a formula and open the wheel of notes it comes from; scroll or ↑ ↓ widens it, 1–6 goes
?
this card