Not All Prefills Are Equal: PPD Disaggregation for Multi-turn LLM Serving

Collection of recent arXiv papers (networking, wireless, and ML-for-telecom) covering: PPD (Prefill-capable Decode) — a dynamic routing scheme to reduce multi-turn LLM serving latency by routing append-prefills to decode nodes and reusing KV caches; EEI-BMA — an AI-assisted, probability-aware MAC for energy-efficient IoT scheduling; SAIL — an unsupervised GAN for controllable RF-map synthesis; Linnaeus — hierarchical AS classification combining network features and LLMs; an LLM-based Net Analyzer rApp for the O-RAN Non-RT RIC emphasizing separation of reasoning and actuation; CATS — a Conducor