(대학원생 마일리지 적용) Seminar Notice (4/23(Tue) 4:00pm, Alexandre Proutiere at KTH)
A unified approach towards lower bounds for estimation and online sequential decision problems
Alexandre Proutiere, KTH
We review existing techniques towards the derivation of fundamental limits (or lower bounds) for estimation and sequential decision problems. We distinguish minimax and problem-specific lower bounds, and provide a unified method towards the latter. The method is applicable to a wide array of problems including linear system identification, bandit and reinforcement learning problems, clustering, etc. It also allows to investigate problems with different performance metrics in mind, e.g., expected error, sample complexity or regret.