Národní úložiště šedé literatury Nalezeno 1 záznamů.  Hledání trvalo 0.01 vteřin. 
Using inheritance dependencies to accelerate abstraction-based synthesis of finite-state controllers for POMDPs.
Shevchenko, Aleksandr ; Macák, Filip (oponent) ; Češka, Milan (vedoucí práce)
Partially observable Markov decision process is an important model for autonomous planning used in many areas, such as robotics and biology. This work focuses on the Abstraction-Refinement framework for the inductive synthesis of finite-state controllers (FSCs) for POMDPs. The classic version of AR requires model checking of a quotient MDP for an entire set of compatible choices of the subfamily in each iteration. We propose an algorithm that uses inheritance dependencies to reduce the size of the quotient MDP’s mask and accelerate model checking for subfamilies of FSCs. We also introduce a smart version of this algorithm, which preserves all its advantages and reduces its weaknesses. During the experiments, it turned out that our approach also affects the operation of other parts of the synthesis, e.g. model building. Depending on the POMDP model, we observe both speedups and slowdowns in comparison to AR. On average, our approach speeds up the overall synthesis time by 1.2 times, and in some cases up to the factor 10.

Viz též: podobná jména autorů
1 Shevchenko, Andrii
1 Shevchenko, Anna
1 Shevchenko, Artur
Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.