Tag: In-Context Learning
All the articles with the tag "In-Context Learning".
-
Competition Dynamics Shape Algorithmic Phases of In-Context Learning
Published: at 11:08 AM
85.25 🤔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.
-
From Distributional to Overton Pluralism: Investigating Large Language Model Alignment
Published: at 11:16 AM 85.12 🤔本文通过分析对齐前后LLM输出分布的变化，揭示了对齐虽减少分布性多元化但通过更长响应实现奥弗顿多元化，且基础模型通过上下文学习可有效模仿对齐模型行为，支持表面对齐假说。
-
Data Whisperer: Efficient Data Selection for Task-Specific LLM Fine-Tuning via Few-Shot In-Context Learning
Published: at 11:22 AM 85.04 🤔Data Whisperer 提出了一种高效、无需训练的基于注意力机制的数据选择方法，通过少样本上下文学习为任务特定的大型语言模型微调选择最优数据子集，在小数据场景下显著提升性能并大幅降低计算成本。
-
Emergence and Effectiveness of Task Vectors in In-Context Learning: An Encoder Decoder Perspective
Published: at 11:25 AM 85.01 🤔本文通过编码-解码框架研究任务向量在上下文学习中的浮现与有效性，提出任务可解码性（TD）指标预测ICL性能，并发现微调早期层比后期层更能提升任务编码和性能。
-
ICLR: In-Context Learning of Representations
Published: at 08:41 AM 84.18 🤔本文通过上下文图追踪任务揭示了大型语言模型能随上下文规模增加而突现地重组概念表示以适应新语义，并提出能量最小化假设解释这一过程。