GraphDx: A Cost-Aware Knowledge-Enhanced Multi-Agent Framework for Sequential Diagnosis

Collection of recent arXiv ML/AI papers (20 Jul 2026) covering medical diagnosis LLM systems, causal/graph-based reasoning, healthcare-specialized LLM training, local voice assistants, multi-agent reasoning behavior, multimodal engineering benchmarks, agentic game-solving components, meme harm detection, and explainable RL via Prolog. Highlights include GraphDx (knowledge-enhanced MDKG + cost-aware multi-agent pipeline improving diagnostic success and lowering test costs on MedQA/MIMIC‑IV), Causal-Audit (target-aware causal-graph construction and auditable path aggregation), Cura 1T (human-g-g