Sign in
A phase transition between positional and semantic learning in a solvable model of dot-product attention.
Hugo Cui
Freya Behrens
Florent Krzakala
Lenka Zdeborová
Published in:
CoRR (2024)
Keyphrases
</>
phase transition
learning algorithm
reinforcement learning
np complete
machine learning
objective function
learning problems
similarity measure
multiscale
probabilistic model
supervised learning
constraint satisfaction problems
constraint satisfaction
learning models
linear model
domain adaptation