Online LLM Selection via Constrained Bandits with Time-Varying Demand

In three linesOnline learning algorithm for dynamic LLM selection in edge-cloud systems under budget constraints (cost, latency). Formulated as constrained stochastic bandit with time-varying demand. Theoretical guarantees: sublinear regret and sublinear constraint violations.

Summary generated by Claude — human-verified