Network Edge Inference for Large Language Models: Principles, Techniques, and Opportunities

This document is an arXiv feed (multiple CS systems/ML papers) covering LLM edge inference, distributed GNN/LLM training optimizations, blockchain transaction sequencing and MEV vs. fair ordering, federated learning client-exclusion strategies, formal verification of local-first access control (CRDTs/Matrix/Keyhive), performant GPU fusion for transformer decoding, distributed sparse solvers, Kubernetes spot-instance provisioning, and atomistic simulation at scale. Security-relevant highlights: (1) the formal-verification paper identifies that existing local-first systems (Matrix, Keyhive) pair