Ústav inteligentních systémů

Browse

Recent Submissions

Now showing 1 - 5 of 21
  • Item
    Reconstruction and enhancement techniques for overcoming occlusion in facial recognition
    (Springer Nature, 2025-05-20) Pleško, Filip; Goldmann, Tomáš; Malinka, Kamil
    Face occlusions on CCTV cameras obscure important key facial features, preventing face recognition (FR) systems from recognizing people. This work mainly focuses on reconstructing these missing facial parts using Generative Adversarial Neural Networks (GANs) to improve FR accuracy while maintaining a low False Acceptance Rate (FAR). In addition, we are trying to improve the generated images further by using different image enhancement methods to test whether they can be used to improve the FR accuracy. To test the results, we perform experiments using state-of-the-art FR methods such as QMagFace and ArcFace to see whether image reconstruction and image enhancement help to improve FR accuracy.
  • Item
    A Uniform Framework for Handling Position Constraints in String Solving
    (ACM, 2025-06-16) Chen, Yu-Fang; Havlena, Vojtěch; Hečko, Michal; Holík, Lukáš; Lengál, Ondřej
    We introduce a novel decision procedure for solving the class of position string constraints, which includes string disequalities, prefixof, suffixof, str.at, and str.at. These constraints are generated frequently in almost any application of string constraint solving. Our procedure avoids expensive encoding of the constraints to word equations and, instead, reduces the problem to checking conflicts on positions satisfying an integerconstraint obtained from the Parikh image of a polynomial-sized finite automaton with a special structure. By the reduction to counting, solving position constraints becomes NP-complete and for some cases even falls into PTime. This is much cheaper than the previously used techniques, which either used reductions generating word equations and length constraints (for which modern string solvers use exponential-space algorithms) or incomplete techniques. Our method is relevant especially for automata-based string solvers, which have recently achieved the best results in terms of practical efficiency, generality, and completeness guarantees. This work allows them to excel also on position constraints, which used to be their weakness. Besides the efficiency gains, we show that our framework may be extended to solve a large fragment of contains (in NExpTime), for which decidability has been long open, and gives a hope to solve the general problem. Our implementation of the technique within the Z3-Noodler solver significantly improves its performance on position constraints.
  • Item
    Stokes problem with the Coulomb stick-slip boundary conditions in 3D: formulations, approximation, algorithms, and experiments
    (Elsevier, 2024-02-01) Haslinger, Jaroslav; Kučera, Radek; Šátek, Václav; Motyčková, Kristina
    The paper deals with the approximation and numerical realization of the Stokes system in 3D with Coulomb's slip boundary conditions. The weak velocity-pressure formulation leads to an implicit in- equality type problem which is discretized by the P1+bubble/P1 elements. To regularize the discrete non-smooth slip term and to release the discrete impermeability condition the duality approach is used. For numerical realization of the resulting saddle-point problem two strategies are proposed, namely i) its fixed-point formulation solved by the method of successive approximations ii) the direct numerical solu- tion of the saddle-point problem. The semi-smooth Newton method is used to solve non-smooth equations appearing in both these approaches.
  • Item
    Comparing Variable Handling Strategies in BDI Agents: Experimental Study
    (SciTePress - Science and Technology Publications, 2024-02-04) Vídeňský, František; Zbořil, František; Beran, Jan; Kočí, Radek; Zbořil, František
    BDI (Belief-Desire-Intention) agents represent a paradigm in artificial intelligence, demonstrating proficiency in reasoning, planning, and decision-making. They offer a versatile framework to construct intelligent agents capable of reasoning about their beliefs, desires, and intentions. Our research focuses on AgentSpeak(L), a popular BDI language, and its interpreter using late variable bindings. Unlike traditional interpreters, it defers substitution selection until execution, enhancing rationality by preventing premature, erroneous selections. To validate our approach, we conducted experiments in a virtual collectable card marketplace. We implemented a system that can use both late and early variable binding strategies, comparing their performance. In shared and independent experiments, the late bindings strategy outperformed the early bindings strategy, although overhead costs were observed. We also conduct a brief discussion of the situations in which it is appropriate to use late bindings given the structure of the declared plans.
  • Item
    Z3-Noodler: An Automata-based String Solver
    (Springer Verlag, 2024-04-04) Chen, Yu-Fang; Chocholatý, David; Havlena, Vojtěch; Holík, Lukáš; Lengál, Ondřej; Síč, Juraj
    Z3-Noodler is a fork of Z3 that replaces its string theory solver with a custom solver implementing the recently introduced stabilization-based algorithm for solving word equations with regular constraints. An extensive experimental evaluation shows that Z3-Noodler is a fully-fledged solver that can compete with state-of-the-art solvers, surpassing them by far on many benchmarks. Moreover, it is often complementary to other solvers, making it a suitable choice as a candidate to a solver portfolio.