Microsoft's MemSteer turns unlabeled user histories into agent memory supervision
MicrosoftResearch · hf · 2026-11-02
- Motivation: Agent memory systems need supervision to adapt to user histories, but labeled QA pairs are rarely available for new histories and are costly to build manually.
- Method: MemSteer converts unlabeled user histories into evidence-grounded QA supervision via an answer plan specifying how records jointly support a reference answer, with source-grounded auditing and blind answer recovery to validate annotations.
- Application: MemSteer-Opt combines QA-based design evaluation with evidence-guided failure diagnosis to revise memory store, retrieval, and context construction — no model updates needed. Competitive with official-QA-guided optimization on ATM-Bench and RHELM, with significant gains over the Base-BM25 baseline. From Peking University and Microsoft Research Asia.
More from coding & agent
- Meme: Wake Up, Use Claude Code, Repeat All Day — tekbog · 2026-11-02
- Non-technical execs drive AI coding course boom around Doubao, prizing output over SOTA — huangyun_122 · 2026-11-02
- FlowHMR gets 5x faster, demoing an LLM agent pipeline from video to photoreal mocap — camenduru · 2026-11-02
- One agent goal has been running for 4 days, burning 30K tokens — CtrlAltDwayne · 2026-11-02
- ESP32-powered 4-inch touchscreen builds a TUI dashboard for monitoring a DGX inference cluster — CATLLM · 2026-11-02
- Daily AI Brief: Gemini 4 Carbon spotted on Arena, DeepSeek open-sources coding agent — testingcatalog · 2026-11-02