Tag: Algorithmic Competition
All the articles with the tag "Algorithmic Competition".
-
Competition Dynamics Shape Algorithmic Phases of In-Context Learning
This paper introduces a synthetic sequence modeling task using finite Markov mixtures to unify the study of in-context learning (ICL), identifying four competing algorithms that explain model behavior and phase transitions, thus offering insights into ICL's transient nature and phenomenology.