AI・機械学習
上級

Self-Consistency (2024年)(セルフコンシステンシー)

2024年成熟Self-Consistency・Industry-leading majority voting CoT sampling + Industry-leading 17.9% MATH improvement + Industry-leading Wang+Wei+Schuurmans Google Self-Consistency CoT enhancement 2024。

0 回閲覧
0 いいね
2026/5/23 更新
関連タグ
self-consistency-2024
self-consistency-cot-enhancement-technique
majority-voting-cot-sampling
17-9-percent-math-improvement
wang-wei-schuurmans-google
google-self-consistency-cot-enhancement-technique

概要

Self-Consistency は2024年成熟・Industry-leading majority voting CoT sampling 2024年Google Self-Consistency CoT enhancement technique position確立。Industry-leading 17.9% MATH improvement vs greedy decode + multiple reasoning paths sampling + temperature sampling + majority answer voting + Wang+Wei+Schuurmans Google 2022 ICLR paper → 2024 LLM standard。

主な特徴・仕組み

  • Industry-leading Xuezhi Wang + Jason Wei + Dale Schuurmans (Google Research) 2022年3月 paper → 2024 mature standard
  • Industry-leading Self-Consistency CoT enhancement technique signature
  • Industry-leading majority voting CoT sampling signature
  • Industry-leading 17.9% MATH improvement vs greedy decode signature
  • Industry-leading multiple reasoning paths sampling signature (40 paths typical)
  • Industry-leading temperature sampling (T=0.7+) signature
  • Industry-leading majority answer voting aggregation signature
  • Industry-leading GSM8K 78% accuracy improvement signature
  • Industry-leading 2022年3月 ICLR 2023 paper "Self-Consistency Improves Chain of Thought Reasoning in Language Models"
  • Industry-leading 2024 LLM ecosystem standard for reasoning enhancement
  • Industry-leading o1 + DeepSeek-R1 inspiration signature (rumored)

まとめ

Self-Consistency = 2024年成熟Google Research Xuezhi Wang + Jason Wei + Dale Schuurmans 2022→2024発表のpremium CoT enhancement technique。Industry-leading majority voting CoT sampling + 17.9% MATH improvement vs greedy decode + multiple reasoning paths sampling 40 paths typical + temperature sampling T=0.7+ + majority answer voting aggregation + GSM8K 78% accuracy improvement + 2022年3月 ICLR 2023 paper "Self-Consistency Improves Chain of Thought Reasoning in Language Models" + 2024 LLM ecosystem standard for reasoning enhancement + o1 + DeepSeek-R1 inspiration rumored position確立。

この記事について
カテゴリーAI・機械学習
難易度上級
作成日2026/5/23