A Training-Free Regeneration Paradigm: Contrastive Reflection Memory Guided Self-Verification and Self-Improvement
无训练再生范式:对比反思记忆引导的自验证与自改进
机构 * Department of Electrical and Computer Engineering, McGill University(麦吉尔大学电气与计算机工程系) ; Department of Mechanical, Industrial and Aerospace Engineering, Concordia University(Concordia大学机械、工业与航空航天工程系)
专题命中 长上下文与记忆 :LLM(abstract);large language model(abstract);language model(abstract);分类 cs.CL
AI总结 本文提出一种无需训练的再生范式,利用离线整理的对比反思记忆提供纠正指导,通过从头再生突破错误推理,提升大语言模型输出准确性。
Comments 18 pages, 5 figures