|
|
Efektivnost datových strukur v implementaci automatů
Koval, Milan ; Smrčka, Aleš (oponent) ; Holík, Lukáš (vedoucí práce)
This Thesis focuses on optimazing the library Mata, that is curently the leader in performance it the fieal operations on automatas. Sygnificant performance increase is achived with the use of a modified linear allocator, using the append-only represetattion of an automata allows for blazing fast construction and blazing fast memery acces owning to resulting spatial locality. The findings of this thesis is not only faster already fastest library Mata but also the viability of linear allocaotors as a more common practice in append only structures, espessialy in performance critical uses.
|
|
|
Transducers in Automata Library Mata
Chocholatý, David ; Lengál, Ondřej (oponent) ; Holík, Lukáš (vedoucí práce)
We implement finite transducers in a new fast and simple automata library Mata. Finite transducers are finite state machines modelling rational relations. Our primary use case for finite transducers is encoding replace operations (replacing a word or a regular pattern with a string literal). A recent automata-based SMT string solver Z3-Noodler uses Mata as a backbone of its decision procedure. Z3-Noodler needs finite transducers to analyse string manipulating programs with replace operations. The analysis of said programs used in web applications prevents software attacks such as cross-site scripting (XSS) or code injection. The distinctive features of Mata include simplicity (simple to use, modify and extend) and efficiency (fast to run). We design the representation and algorithms for finite transducers to fit the simplicity and efficiency requirements. We inherit and extend the existing data structures and algorithms for finite automata in Mata to represent the finite transducers and their operations. The representation for finite transducers serves as a common data structure and interface for the finite transducers and future representation of automata using multi-terminal binary decision diagrams to handle large alphabets. We further extend the design with algorithms to construct finite transducers modelling replace operations defined in SMT-LIB. Finally, we run an experimental evaluation of performance of finite transducers in Mata on a new benchmark with replace operations from runs of Z3-Noodler and from solving problems in pattern matching.
|
|
|
String Constraint Solving Through Parikh Images
Bartoš, Petr ; Havlena, Vojtěch (oponent) ; Holík, Lukáš (vedoucí práce)
This bachelor thesis aims to implement an alternative way of solving string constraints using the so-called flattening algorithm. The algorithm makes use of Parikh images and parametric flat automata to effectively convert string constraints to linear arithmetic, which allows for leveraging powerful SMT solvers. Solving constraints as an algebraic problem is supposed to be more efficient than standardly used automata-based techniques, as it avoids common pitfalls, such as state-space exposion. The thesis covers the theoretical knowledge required to understand the flattening algorithm and introduces alternative modern solution strategies. The implementation results are then compared to other solvers using conventional competition benchmarks. The conducted experiments show that while the speed of the implementation compared to other state-of-the-art solvers is worse, the effectiveness of the underapproximation itself is fairly promising, thus yielding mixed results.
|
|
|
Deciding Logic with Automata
Hečko, Michal ; Holík, Lukáš (oponent) ; Lengál, Ondřej (vedoucí práce)
The work presented in this thesis focuses on deciding quantified linear integer arithmetic using finite automata. We present a novel implementation of the classical automata-based decision procedure supporting the SMT-LIB input format. Our comprehensive presentation of the developed tool focuses on various aspects and design decisions that play a prominent role in the performance of the implementation. We identify the lack of theory-based reasoning as the primary reason for the overall poor performance of the decision procedure and give a range of cheap heuristics that significantly improve its speed. We also give a~novel top-down reformulation of the procedure that allows to perform theory-based reasoning during the construction of automata. We also compare our tool to the state-of-the-art SMT solvers, showing that our prototype implementation is comparative and even superior to the state of the art.
|
|
|
Optimizing Inductive Controller Synthesis Methods for POMDPs with Discounted Rewards Properties
Kříž, Ondřej ; Holík, Lukáš (oponent) ; Češka, Milan (vedoucí práce)
Probabilistic model checking is essential for verifying systems in diverse domains. A key limitation of the PAYNT tool, which synthesises probabilistic programs satisfying given specifications, lies in its handling of discounted properties. This thesis extends the STORM framework upon which PAYNT is built, incorporating the discounted value iteration method within inductive synthesis process to address this issue. The discounted value iteration function was developed within STORM, involving solver environment identification, decision-making segments in the code, and Gauss-Seidel multiplication for enhanced computational capabilities. The necessity for a PRISM formula in PAYNT's model checking process presented challenges for bypassing the discount factor transformation step. To overcome this, a discount factor transformation with a factor close to one was employed, comparing potential optima vectors between discounted and undiscounted iterations. This study improves discounted property handling in PAYNT and the STORM framework, providing a foundation for further advancements in the development of PAYNT.
|
|
|
Repetitive Substructures for Efficient Representation of Automata
Šedý, Michal ; Češka, Milan (oponent) ; Holík, Lukáš (vedoucí práce)
Nondeterministic finite automata (NFAs) are widely used across almost every field of computer science, such as for the representation of regular expressions, monitoring high-speed networks, in abstract regular model checking, program verification, in decision procedures of WS1S and WS2S logics, linear integer arithmetic, temporal logics, or even in bioinformatics for searching sequences of nucleotides in DNA. Automata with a large number of states can lead to an exponential increase in the state space in many algorithms. To address this issue, minimization techniques, such as state merging and transition pruning, are used. Despite the strong minimization potential of these methods, the resulting automata can still contain duplicate substructures with equivalent transition sequences. There are even types of automata that cannot be minimized by these standard methods at all. This work presents a novel automata minimization approach based on a transformation of an NFA into a nondeterministic pushdown automaton (NPDA). The transformation identifies multiple similar substructures and replaces them with one common structure (called a procedure). By doing so, we were able to further reduce automata by up to 67.3%. The principle of transforming NFA into NPDA can be understood as a transformation of a purely sequential program into a program with functions and a call stack.
|
|
|
Automata in Verification
Šmahlíková, Barbora ; Holík, Lukáš (oponent) ; Lengál, Ondřej (vedoucí práce)
Regular model checking is an automata-based technique used for verification of infinite-state systems. The configurations of a system are encoded as a finite automaton and transitions between these configurations as a finite transducer. A technique for verifying arbitrary properties of parameterized systems specified in a temporal logic LTL(MSO) has already been introduced. We present an extension of this algorithm allowing verification of hyperproperties of parameterized systems where an explicit quantification over multiple execution traces is allowed. We specify conditions that need to hold for a pair of advice bits (a finite automaton and a finite transducer) that serves as a witness of the fact that the property holds in the system. The technique presented in this work is implemented in our tool ParaHyper - the only existing tool for the verification of hyperproperties of parameterized systems. The tool uses a SAT solver to generate automata and transducers. If a pair satisfying the conditions for advice bits is found, the property holds in the system. We performed an experimental evaluation of our approach and found that ParaHyper is able to generate advice bits for formulae with an alphabet up to 4 symbols if both the automaton and the transducer have at most 2 states. When a candidate pair is given by the user, ParaHyper can, however, efficiently check if it satisfies the conditions for advice bits even for larger alphabets and greater number of states.
|
|
|
Using Reinforcement learning and inductive synthesis for designing robust controllers in POMDPs
Hudák, David ; Holík, Lukáš (oponent) ; Češka, Milan (vedoucí práce)
A significant challenge in sequential decision-making involves dealing with uncertainty, which arises from inaccurate sensors or only a partial knowledge of the agent's environment. This uncertainty is formally described through the framework of partially observable Markov decision processes (POMDPs). Unlike Markov decision processes (MDP), POMDPs only provide limited information about the exact state through imprecise observations. Decision-making in such settings requires estimating the current state, and generally, achieving optimal decisions is not tractable. There are two primary strategies to address this issue. The first strategy involves formal methods that concentrate on computing belief MDPs or synthesizing finite state controllers, known for their robustness and verifiability. However, these methods often struggle with scalability and require to know the underlying model. Conversely, informal methods like reinforcement learning offer scalability but lack verifiability. This thesis aims to merge these approaches by developing and implementing various techniques for interpreting and integrating the results and communication strategies between both methods. In this thesis, our experiments show that this symbiosis can improve both approaches, and we also show that our implementation overcomes other RL implementations for similar tasks.
|
|
|
Učení se automatů pro rychlou detekci anomálií v síťovém provozu
Hošták, Viliam Samuel ; Matoušek, Petr (oponent) ; Holík, Lukáš (vedoucí práce)
Táto práca sa zaoberá rýchlou detekciou sieťových anomálií na základe učenia automatov. Popisuje a porovnáva niekoľko vybraných algoritmov učenia automatov, vrátane ich aplikácie na učenie sieťových charakteristík. Pre takto naučené automaty je navrhnutých niekoľko metód detekcie sieťových anomálií, ktoré umožňujú odhaliť tak sekvenčné, ako aj štatistické anomálie v rámci komunikácie. Za týmto účelom využívajú mechanizmy samotných automatov, ich transformáciu, či štatistickú analýzu. Navrhované metódy detekcie boli implementované a vyhodnotené na prevádzke protokolu IEC 60870-5-104 používaného v industriálnych kontrolných systémoch.
|
|
|
Podnikatelský záměr
Holík, Lukáš ; Smolíková,, Lenka (oponent) ; Heralecký, Tomáš (vedoucí práce)
Cílem diplomové práce je vytvořit podnikatelský plán pro pana Davida Mokrého, majitele kavárny s názvem Blue Queen v Boskovicích, který v blízké době chce rozšířit svoji podnikatelskou činnost o další podnik - Black King. Pomocí četných analýz trhu a analýz vnějšího i vnitřního okolí získám data potřebná k vytvoření projektu na vybudování úspěšné nové kavárny, která by se měla stát nejvyhledávanějším podnikem v okolí.
|
host ::
RSS.