Tag: Sequence Modeling
All the articles with the tag "Sequence Modeling".
-
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.
-
Test-time regression: a unifying framework for designing sequence models with associative memory
本文提出一个基于测试时回归的统一框架,通过将关联回忆形式化为回归问题,推导出多种序列模型(如线性注意力、状态空间模型、softmax注意力),并通过合成实验验证其回归能力,同时提出高阶注意力泛化。