ConMem Framework Improves Inspection Log Reasoning via Shapley-style Estimation

ConMem uses a contribution-aware memory framework to improve LLM accuracy in manufacturing inspections. It segments logs into functional units and applies Shapley-style estimation to retain high-value evidence within constrained memory budgets, achieving 76.0% QA accuracy.