Reimagining service delivery in the agentic era with Google Public Sector
cloud.google.com·21h ago
AI Research
arXiv CS.AI·Jun 10·5 min read
STORM: Stepwise Token Optimization with Reward-Guided Beam Search
TL;DR
Existing retrieval methods struggle with vocabulary mismatch and inefficiencies when adapting to new models. STORM (Stepwise Token Optimization with Reward-Guided Beam Search) is introduced as a self-supervised framework that optimizes lexical query expansion by scoring candidate terms against a BM25 index.
✦ Why It Matters
Engineers can leverage STORM for efficient and effective query expansion in retrieval systems without heavy infrastructure.
Key Takeaways
- Utilize STORM for efficient lexical query expansion.
- Leverage reward-guided beam search for better token selection.
- Consider STORM as a lightweight alternative to dense retrieval models.
- Explore STORM's multilingual capabilities for diverse applications.
How It Works
STORM operates by generating candidate query expansions and scoring them against a BM25 index. It uses a reward-guided beam search to prune low-reward continuations, effectively turning retrieval performance into a token-level optimization signal.
This allows the model to focus on vocabulary that enhances retrieval effectiveness, rather than generating well-formed but ineffective terms.