Transformers Learn Nonlinear Features In Context: Nonconvex Mean-field Dynamics on the Attention Landscape.

Juno Kim Taiji Suzuki

Published in: CoRR (2024)

Keyphrases

feature space
low level
classification accuracy
markov random field
image features
dynamical systems
extracted features
image segmentation
feature extraction
prior knowledge
co occurrence
contextual information
convex optimization