|
|
GPU-Accelerated Synthesis of Probabilistic Programs
Marcin, Vladimír ; Matyáš, Jiří (oponent) ; Češka, Milan (vedoucí práce)
This paper examines the problem of automatic synthesis of probabilistic programs: having a finite family of candidate programs, how can one efficiently identify a program that satisfies a given specification. Even the most straightforward synthesis problems prove to be NP-hard. An improvement to this state of practice is brought by the PAYNT tool, which tackles this problem with a novel integrated technique for synthesising probabilistic programs. Even though it efficiently deals with the exponential growth of the family size, there is still a problem with the underlying state-space explosion. To solve this problem, we have implemented GPU-oriented model-checking algorithms that takes advantage of the GPU architecture and parallelise the task at a state level of a probabilistic model. The overall acceleration that we were able to achieve with this approach was, under certain conditions, close to the theoretically possible limit of the acceleration of the whole synthesis process.
|
|
|
Novel Methods for Semi-Quantitative Analysis of Biochemical Systems
Bíl, Jan ; Andriushchenko, Roman (oponent) ; Češka, Milan (vedoucí práce)
This thesis aims on providing novel methods for analysis of stochastic bio-chemical systems. It introduces a new population abstraction based on dirac semi-Markov processes. It turned out, that this abstraction is more precise than Continuous time Markov chain abstraction. On this abstraction, analysis methods are provided. Algorithm for transient analysis over this abstraction is described and also novel timed temporal logic formulas, that allow to express interesting biological properties are presented. Further, model-checking algorithm for these formulas is proposed and implemented. Preliminary experiments showing potential of this approach are also described.
|
|
|
GPU-Accelerated Synthesis of Probabilistic Programs
Marcin, Vladimír ; Matyáš, Jiří (oponent) ; Češka, Milan (vedoucí práce)
This paper examines the problem of automatic synthesis of probabilistic programs: having a finite family of candidate programs, how can one efficiently identify a program that satisfies a given specification. Even the most straightforward synthesis problems prove to be NP-hard. An improvement to this state of practice is brought by the PAYNT tool, which tackles this problem with a novel integrated technique for synthesising probabilistic programs. Even though it efficiently deals with the exponential growth of the family size, there is still a problem with the underlying state-space explosion. To solve this problem, we have implemented GPU-oriented model-checking algorithms that takes advantage of the GPU architecture and parallelise the task at a state level of a probabilistic model. The overall acceleration that we were able to achieve with this approach was, under certain conditions, close to the theoretically possible limit of the acceleration of the whole synthesis process.
|
|
|
Novel Methods for Semi-Quantitative Analysis of Biochemical Systems
Bíl, Jan ; Andriushchenko, Roman (oponent) ; Češka, Milan (vedoucí práce)
This thesis aims on providing novel methods for analysis of stochastic bio-chemical systems. It introduces a new population abstraction based on dirac semi-Markov processes. It turned out, that this abstraction is more precise than Continuous time Markov chain abstraction. On this abstraction, analysis methods are provided. Algorithm for transient analysis over this abstraction is described and also novel timed temporal logic formulas, that allow to express interesting biological properties are presented. Further, model-checking algorithm for these formulas is proposed and implemented. Preliminary experiments showing potential of this approach are also described.
|
host ::
RSS.