Tag: Few-Shot Learning

All the articles with the tag "Few-Shot Learning".

• Less is More: Enhancing Structured Multi-Agent Reasoning via Quality-Guided Distillation

  Published: 4 May, 2025 |  at  04:28 PM

  

  77.44 🤔

  本文提出了一种质量导向的多代理框架，通过提示诱导、检索增强合成和奖励过滤从少量标注数据中提炼高质量监督信号，提升LLMs在低资源结构化推理任务中的性能。

• The dynamic interplay between in-context and in-weight learning in humans and neural networks

  Published: 6 May, 2025 |  at  11:20 PM

  

  70.07 🤔

  本文通过神经网络中上下文学习（ICL）与权重学习（IWL）的动态交互，统一解释了人类学习中的组合性泛化、课程效应及灵活性与保留性权衡，为认知科学双过程理论提供了新视角。

• HyPerAlign: Hypotheses-driven Personalized Alignment

  Published: 4 May, 2025 |  at  04:32 PM

  

  65.01 🤔

  本文提出HyPerAlign方法，通过假设驱动的少样本学习实现LLM的个性化对齐，提高了模型对个体用户的适应性和安全性，同时减少了对微调的依赖。

• How do Humans and Language Models Reason About Creativity? A Comparative Analysis

  Published: 10 May, 2025 |  at  10:59 AM

  

  60.58 🤔

  This paper conducts a comparative analysis of creativity evaluation in STEM, revealing that human experts and LLMs prioritize different facets of originality (cleverness vs. remoteness/uncommonness) and are differentially influenced by contextual examples, with LLMs showing higher predictive accuracy but poorer construct validity due to homogenized facet correlations.

• Which Attention Heads Matter for In-Context Learning?

  Published: 5 May, 2025 |  at  11:15 PM

  

  90.67 👍

  本文通过对12个大型语言模型进行消融研究和训练动态分析，发现函数向量头是驱动少样本上下文学习的主要机制，尤其在大型模型中，并且许多函数向量头在训练过程中从归纳头演变而来，纠正了先前认为归纳头是主要驱动力的观点。