learn.ifs.name

Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

Asai et al., 2024, Self-RAG (The Twelfth International Conference on Learning Representations (ICLR 2024))

Subscriber deck — free during early access

1 cards in “3.1 Problem Formalization and Overview” · whole deck

3.1 Problem Formalization and Overview

  1. How does Self-RAG generate a response at inference time?
    Segment by segment (a sentence in the experiments): it first predicts Retrieve; if no, it continues like an ordinary LM. If yes, it retrieves several passages, processes each in parallel (predicting IsRel, a candidate segment and IsSup), then ranks the candidates by their critique tokens and finally predicts IsUse.
    [Retrieve]?noyesnextsegmentd₁RelSup ✓d₂Irrel.–d₃RelSup ½keep d₁’s segmentcritique tokens rank the candidates
    One Self-RAG step: retrieve on demand, generate per passage, critique, choose (invented example)
    • Asai et al., 2024, Self-RAG (The Twelfth International Conference on Learning Representations (ICLR 2024)), §3.1

Source

  • Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil and Hannaneh Hajishirzi. 2024. Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection. In The Twelfth International Conference on Learning Representations (ICLR 2024). https://arxiv.org/abs/2310.11511