Publications [#382588] of Sayan Mukherjee

Papers Published

1. Busatto-Gaston, D; Chakraborty, D; Majumdar, A; Mukherjee, S; Pérez, GA; Raskin, JF, Bi-objective Lexicographic Optimization in Markov Decision Processes with Related Objectives, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14215 LNCS (January, 2023), pp. 203-223 [doi]
   (last updated on 2025/04/11)

   Abstract:
   We consider lexicographic bi-objective problems on Markov Decision Processes (MDPs), where we optimize one objective while guaranteeing optimality of another. We propose a two-stage technique for solving such problems when the objectives are related (in a way that we formalize). We instantiate our technique for two natural pairs of objectives: minimizing the (conditional) expected number of steps to a target while guaranteeing the optimal probability of reaching it; and maximizing the (conditional) expected average reward while guaranteeing an optimal probability of staying safe (w.r.t. some safe set of states). For the first combination of objectives, which covers the classical frozen lake environment from reinforcement learning, we also report on experiments performed using a prototype implementation of our algorithm and compare it with what can be obtained from state-of-the-art probabilistic model checkers solving optimal reachability.