BaRe-Mem: Bayesian Reliability Memory Makes Multi-Agent Consultation Robust to Misleading Advisors
NanyangTechnologicalUniversity · hf · 2026-09-29
In multi-agent systems, advisor quality varies by task, and misleading advice can make consultation worse than reasoning alone. NTU researchers propose BaRe-Mem, an online Bayesian reliability memory that estimates each advisor's reliability from the central model's internal belief representations, updates estimates from historical interactions, and uses them to modulate advisor influence and decide between consultation and autonomous reasoning.
- Across 9 benchmarks and 6 central models, it is more robust to misleading advisors than debate and majority voting
- On harder tasks it stays above autonomous reasoning at all tested misleading levels
- Extended to worker allocation in agent teams, it beats routing by historical success counts on MuSiQue and identifies capable workers earlier
More from coding & agent
- PHP WASM kernel hits major milestone with PR #936 merge, author vows to keep backing open source — MickeySteamboat · 2026-09-29
- Google to replace Gemini Gems with Skills starting November 17 — mark_k · 2026-09-29
- Carla v0.1.0: a local llama.cpp loom TUI for growing AI characters — max_paperclips · 2026-09-29
- Running Jev at high frame rate with full-state snap inferences makes it a true System 1 — mathemagic1an · 2026-09-29
- Meeting agent memory design: retain facts instead of patching tags — Extension-Answer4821 · 2026-09-29
- MegaMemory: a persistent knowledge graph MCP server that lets coding agents remember across sessions — tom_doerr · 2026-09-29