MORES: Mobile Reasoning-as-a-Service via Distributed LLM Inference-Time Scaling

This collection of recent papers covers systems and ML advances for edge/cloud inference, wireless networks, autonomous-vehicle federated RL, data-center memory architecture, and reproducible network research. Key contributions: MORES — an edge/server cooperative implicit-reasoning framework that partitions recurrent LLM hidden-state updates and uses a semantic MoE + DRL scheduler for wireless heterogeneity; ADORN — a Q-learning retraining policy for handling concept drift in O‑RAN forecasting; StemGNN — a spatio-temporal GNN to predict UE scheduling and mitigate backhaul-delay harms to beam‑f