Research grants

Collaboration on Fermilab experiments 2

CZ.02.1.01/0.0/0.0/18_046/0015954 2020 - 2022

The proposed project follows the OPVVV project CZ.02.1.01 / 0.0 / 0.0 / 16_013 / 0001787 of similar name and abbreviation. Its objective is the continuation of reconditioning and increasing the computing and storage capacity for processing data from experiments in Fermilab. The aim is to improve the infrastructure to gain new scientific knowledge in collaborative experiments in Fermilab.

Understanding, Handling, and Exploiting Biases in Knowledge Graphs (BiasKG)

20-02080J 2020 - 2022

In computer science, especially in data science and artificial intelligence, bias is usually used to refer to an unintential skew in a dataset or behavior of a program. This project primarily focuses on biases demonstrating as algorithmic biases, which affect many artificial intelligence applications, resulting, for example, in discrimination and underperforming models. We aim to study biases in knowledge graphs, which provide machine readable semantics for the emerging generation of AI applications. We want to take a dual look at the removal of bias from knowledge graphs by identifying the cases and scenarios in which removing biases results in an improvement, and in which cases removing the bias deteriorates the knowledge graph or machine learning model, since the bias should actually be considered a useful signal. The project will build on prior psychological research on cognitive biases to identify the pathways through which biases enter data and explore how they manifest. Methods for bias removal based on active learning and crowdsourcing will be proposed.

Random discrete structures

20-27757Y 2020 - 2022

The project focuses on problems in the overlap of discrete mathematics and probability theory. We consider basic discrete structures: graphs, digraphs, trees, uniform hypergraphs, which in applied areas are used as abstract models for networks, population dynamics, etc. We will study randomly generated discrete structures from a theoretical perspective. Focusing on random variables which count the number of certain substructures (e.g., copies of a given graph), we seek answers to the following questions: what are their asymptotic distributions, how likely are events that these random variables deviate significantly from their expected values. Among the objectives of the project is to study the similarity between the random regular graph and the binomial graph by resolving the Sandwiching Conjecture of Kim and Vu; estimating the upper tail probability for small subgraph counts in sparse random graphs; determining the limit distribution of functionals (e.g., number of large matchings, maximum independent set) in random graphs and Galton-Watson trees.

Modelling the sleeping brain: towards a neural mass model of sleep rhythms and their interactions

CZ.02.2.69/0.0/0.0/19_074/0016209 2019 - 2021

Memory consolidation, a prominent example of higher cognitive processes, relies on two important neural phenomena: slow- wave sleep with a wealth of distinct rhythms, and cross-frequency coupling between these rhythms. Neither of these processes is, to this day, fully understood. However, according to the two-stage model of memory consolidation, the interplay between slow oscillations and sleep spindles by means of phase-amplitude coupling, as well as the interplay between sleep spindles and hippocampal sharp-wave ripples, seem to promote neural plasticity and initiate a cortical- hippocampal dialogue that leads to experience replay and, ultimately, migration of newly encoded memories to longer-lasting storage. The overall aim of this project is to shed light on the slow oscillation-spindle interplay using a biologically realistic neural mass model and, additionally, reproduce the cross-frequency phenomena.

Structural properties of visibility in terrains and farthest color Voronoi diagrams

GJ19-06792Y [Registered results] 2019 - 2021

This research project deals with two popular topics in Combinatorial and computational geometry: visibility and Voronoi diagrams. The first concrete topic is visibility in terrains in the presence of multiple observers. This variant has received much less attention than the case of a single guard and presents a great number of applications. Given a terrain and a set of observers, the most fundamental question is being able to describe which parts of the terrain are visible by at least one of the observers; we will try to improve on the current fastest algorithms to solve this problem. We also plan to study approximate versions of the visibility maps, and realistic settings where the observers or the terrain satisfy some natural assumptions. The second topic concerns the farthest color Voronoi diagram, which has not been as studied as other types of Voronoi diagrams. We intend to get new insights on the structure of this diagram, and explore their algorithmic consequences.

National Competence Center - Cybernetics and Artificial Intelligence

TN01000024 [Registered results] 2019 - 2020

The NCK KUI project aims to create a national platform for cybernetics and artificial intelligence which interlinks research and application oriented centers of robotics and cybernetics for Industry 4.0, Smart Cities, intelligent transport systems and cybersecurity. The connection of innovation leaders will raise effectivity of applied research in key areas, as advanced technology for globally competitive industry, ICT and transportation for the 21st century. NCK KUI is closely related to application sector and enables cross-domain collaboration, innovation development and technology transfer.

Embedding, Packing, and Limits in Graphs

GA19-08740S [Registered results] 2019 - 2021

Graphs are among the simplest mathematical structures. They forms the foundation of much of Computer Science and their importance grew enormously with the development of computer networks. In this project, we focus on central problems from extremal graph theory, as well as the recent related area of limits of graphs. We shall exploit classical methods from extremal graph theory, as well as probabilistic and analytical methods. Our main topics are embedding problems, packing problems, and the study of graph limits via a weak* topology approach.

FoNeCo: Analytical Foundations of Neurocomputing

GA19-05704S [Registered results] 2019 - 2021

The computational paradigm has recently been shifted towards intelligent information processing which can be demonstrated by tremendous success of (deep) neural networks (NNs) producing state-of-the-art results in artificial intelligence. The majority of techniques used in NNs are of heuristic nature and therefore only partially theoretically justified. FoNeCo is a basic research project whose ambition is to contribute to the development of analytical foundations of selected practical neurocomputing models. Its aim is to characterize the computational power of subrecursive NNs between integer and rational weights using quasiperiodicity, and bio-inspired NNs based on synfire rings, to study approximation properties (e.g. model complexity) and robust fitting of various regression NNs, and to analyze the complexity of loading deep NNs. The analysis of respective NN models will be the basis of new architectures and more efficient, reliable, and theoretically-founded learning algorithms which will be implemented as open-source software and experimentally tested on benchmark problems.

Boolean Representation Languages Complete for Unit Propagation

GA19-19463S [Registered results] 2019 - 2021

This is a basic research project in which we plan to work on problems in the area of knowledge compilation. We consider a particular case of knowledge compilation where the knowledge base is represented by a boolean function, usually in a conjunctive normal form (CNF) formula. We plan to study a compilation of such a formula into another CNF formula which is complete for unit propagation. More precisely, we will consider compilation into the target language of unit refutation complete formulas (URC), the language of propagation complete formulas (PC), and the corresponding variants of encodings with auxiliary variables, namely URC and PC encodings. The goal of this project is to solve theoretical questions related to compilation into URC or PC representation or encoding and develop a compiler for this task. To this end we shall develop and test algorithms and heuristic for automatic compilation. The outputs of this project will be published in journal and conference papers.

Timing of the spatial scene processing in the dorsal and ventral visual stream of the human brain

GA19-11753S [Registered results] 2019 - 2021

The project’s focus is the dynamics of brain processes during spatial scene processing in the dorsal and ventral visual streams. While the dorsal stream, going from visual cortex to parietal lobe, processes mainly the spatial information, the ventral stream, going to the temporal lobe, is focused on object recognition. The level of separation of the two streams and their interconnection is not fully established. Both streams converge in the temporal lobe, enabling the emergence of the spatial representation containing orientation marks, but the precise localization and dynamics of this connection is unclear. The project will examine several models of cooperation between these visual streams and the information flow between occipital, parietal and temporal lobe and retrosplenial cortex. Using intracranial recordings in epileptic patients, we will determine the time windows of activation and synchronization of these brain areas with high time resolution. The project will bring results concerning the dynamics of brain processes fundamental for visual scene processing theories.

Nonlinear interactions and information transfer in complex systems with extreme events

GA19-16066S [Registered results] 2019 - 2021

Theory of synchronization of nonlinear dynamical systems and information theory meet in effort to understand cooperative behaviour in complex systems. Yet features such as multiscale dynamics and fat-tailed probability distributions have not been adequately addressed in development of tools for uncovering interactions and causal information flow from experimental time series. In this project we will employ and further develop methods of information theory and superstatistics in order to detect and quantify internal dynamics and information flow in real-world complex systems in which extreme events occur. In particular, we will study Shannon and Renyi information transfer in multivariate and multiscale time series and advance the application of the superstatistics paradigm in complex dynamics with non-Gaussian fluctuations. The primary application ground will be the multivariate meteorological data reflecting the changing Earth climate, evolving on multiple time and space scales.

Introduction of targeted protection of cereal crops against insect pests in precision farming

QK1910281 [Registered results] 2019 - 2022

This project will develop statistical tools for prediction of crop pests in the age of precision agriculture. The developed methodologz will be based on modern semiparametric and dznamical modeling in the GAM framework. The models will be developed in several variants and the most suitable model will be selected by formalized statistical procedures. Based on the validated model, we will construct both routine predictions and derive recommendations for crop management timing.

Enhancing human resources for research in theoretical computer science

EF17_050/0008361 [Registered results] 2018 - 2020

Theoretical computer science uses mathematical methods in order to clarify some of the most important notions used in computer science and, therefore, it is a key research area within computer science. Logic is an important part of theoretical computer science, focusing on modelling bodies of information and ways how agents use them in reasoning. So called non-classical logics, alternatives to classical logic, aim at a better representation of various aspects of information and reasoning than the representations based on classical logic (these aspects include vagueness, incompleteness and inconsistency of information or limitations of reasoning agents). Within this project, the Institute of Computer Science hosts three researchers working on non-classical logics and their applications to specific problems in theoretical computer science. The overall goal of the project is to study (i) applications of non-classical logics in program verification, (ii) applications of non-classical logics in the study of weighted structures (with focus on the Valued Constraint Satisfaction Problem) and (iii) non-classical logics with so-called generalized quantifiers (with focus on their computational properties).This project is funded by the European structural and investment funds within the Operational Programme Research, Development, Education.

Urban Adaptation Challenges: Promoting Sustainable Planning Using Integrated Vulnerability Analysis

TL01000238 [Registered results] 2018 - 2022

The main objective of the proposed multidisciplinary project is to conduct vulnerability analysis and to create a methodology of integrated vulnerability assessment of cities and their inhabitants to temperature extremes, using the classification of urban surfaces (delivered 02/2022), aiming on supporting the adaptation planning and adaptation measures in cities. Detailed spatial vulnerability assessment will be performed for three pilot cities (Prague, Brno, Ostrava), using climatological analysis and participatory approaches, considering projections of future developments (climatological scenarios, land cover changes, socio-demographic trends), with an output of specialized vulnerability maps. The integrated assessment aims to contribute to the sustainable urban planning in Czechia.

Capabilities and limitations of shallow and deep networks

GA18-23827S [Registered results] 2018 - 2020

The goal of the project is to contribute to development of neurocomputing by combining theoretical and experimental research investigating capabilities and limitations of shallow and deep artificial neural networks from the point of view of their model complexity and robustness of learning.

URBI PRAGENSI Urbanization of weather and air quality forecast and climatic scenarios for Prague

UH0383 [Registered results] 2018 - 2020

The project Urbi Pragensi addresses improvements and implementation of the weather prediction and air quality prediction for region of Prague together with more detailed assessment of impacts of climate change in the city. The prediction improvements are achieved by utilization of the modern methods based on incorporation of parameterization of urban level processes into the high resolution atmospheric models. Complementary approach are microscale simulations which can expose in detail situation in given parts of the town with substantial burden of heat island and air pollution. This can contribute to estimation of the health risks as well as to efficiency assessment of proposed mitigation measures. ICS is involved in works on concept KK1 (meteorological prediction), KK2 (air quality prediction) and it is the coordinator of the concept KK4 (microscale simulations). The main partner of the project is Charles University in Prague, another partner is Czech Hydrometeorological Institute in Prague. In the scope of the project, we tightly collaborate with other Czech and foreign academic institutions (from Germany and Finland). We also tightly cooperate with principal intended users of the project outputs (e.g. Prague municipality and Prague Institute of Planning and Development).

Reasoning with graded properties

GA18-00113S [Registered results] 2018 - 2020

The goal of the grant is to create logical framework to model reasoning with graded predicates. We distinguish several types of such predicates, discuss their ubiquity in rational interaction and the logical challenges they may pose. We plan to utilize mathematical fuzzy logic as a set of logical tools that allow to model reasoning with graded predicates and concentrate on a philosophical account of the vagueness problem that makes use of these tools. Other aim is to generalize this approach to other kinds of graded predicates a lay foundation for a general research program towards a logic-based account of reasoning with graded predicates.

Fusion-Based Knowledge Discovery in Human Activity Data

GA18-18080S [Registered results] 2018 - 2020

The proposed project of basic research in the area of multimedia data mining attempts to contribute to further development of methods for knowledge discovery in video data recording human activities. It concentrates on the one hand on combining knowledge about the detected human bodies, their parts and their motion with other knowledge obtainable from the available data, such as knowledge about scene, context, or person-object interactions, on the other hand on research into different ways of fusion of multiple classification or regression models during such knowledge discovery, including fusion based on deep artificial neural networks. These objectives are supported by experimental equipment and by the composition of the project team. At the Faculty of Information Technology of the Czech Technical University, a modern Image Processing Laboratory has been recently inaugurated, and the team includes, apart from the head of the laboratory and several PhD students performing in it their research, also two internationally recognized researchers in the area of data mining.

Non-classical logical models of information dynamics

GJ18-19162Y [Registered results] 2018 - 2020

Reasoning about modifications of information available to cognitive agents (about information dynamics) is an essential part of many everyday activities. A better understanding of mechanisms of such reasoning is important both from the practical viewpoint and in the context of epistemology. The project studies logical models of information dynamics based on non-classical logics. These models are mode adequate that the standard model based on classical logic. Their advantages include a better representations of information modifications resulting from reasoning of logically imperfect agents, a better representation of graded acceptance of information and the fact that they are able to model dynamics of questions in the context of logically imperfect agents. The project will provide a complete picture of some of these models. We will provide new results concerning their properties and study their applications in epistemology. The project will therefore lay a technically sound foundation of future research in the intersection of formal epistemology and philosophical logic.

Collaboration on Fermilab experiments

EF16_013/0001787 [Registered results] 2017 - 2020

The subject of the project is to improve the methods for data processing of Fermilab experiments and design of new analyzing procedures of these data based on machine learning algorithms and artificial intelligence. The goal is to continually improve the potential of acquiring new scientific knowledge on collaborative experiments in Fermilab, the development of new innovative data processing methods in high energy physics and the improvements of computing infrastructures of participating experimental institutions in Fermilab. The target group of the project results are the researchers of universities and R&D experts in general. From project outputs benefit experts from all research institutions cooperating experiments in Fermilab.

Functional and structural reorganization of brain networks after stroke: implications for diagnosis and therapy of associated comorbidities

NV17-28427A [Registered results] 2017 - 2021

Cognitive impairment and epilepsy are the most common and serious post-stroke comorbidities. They have far-reaching impacts on patient health status and decrease the efficacy of rehabilitation and delay the recovery. Severe attention deficits, post-stroke dementia or epilepsy are comorbidities which have failed to be explained in terms of being a direct conseguence of localized damage to specific brain structures. Converging evidence suggests that understanding the emergence of these comorbidities requires radically new research paradigms that consider the impact of stroke on whole-brain connectivity. In this multidisciplinary project we seek to brain areas with high connectivity and information transfer - network hubs. Application of such a fundamental and innovative framework has potential to advance our understanding of stroke consequences and to develop early, network-targeting. therapeutic interventions to prevent or ameliorate these detrimental comorbiditiens.

Effectiveness analysis of prenatal diagnosis of congenital malformations and survival of children born with a birth defect in 1994-2015

NV17-29622A [Registered results] 2017 - 2021

Retrospective analysis of the consecutive data on prenatally and postnatally diagnosed cases of birth defects (BD) diagnosed in the Czech Republic between 1994 and 2015. We will perform 1) an analysis of incidences of prenatally diagnosed cases of BDs in the Czech Republic; 2) an analysis of incidences of postnatally diagnosed cases of BDs in the Czech Republic (incidence in births); 3) an analysis of effectiveness of prenatal diagnosis according to the individual diagnoses of BDs; 4) an analysis of stillbirth rate, perinatal neonatal and infant mortality in children with BDs.

Metalearning for Extraction of Rules with Numerical Consequents

GA17-01251S [Registered results] 2017 - 2019

One of the key directions of data mining, particularly important if human-comprehensibility plays a role, is the extraction of rules from data. The project aims at rules with consequents corresponding to numerical variables. In spite of the ubiquity of such variables, rules extraction is not yet as mature for them as for classification and association rules. The main objective of the project is to develop a framework to enable assessing different algorithms for the extraction of rules with numerical consequents from a given dataset. Traditional algorithms view consequent variables as responses and antecedent variables as regressors of regression models. They are complemented by emerging algorithms of computational topology. The framework will be based on metalearning, i.e., learning from metadata about the past performance of the algorithms on datasets with similar values of metafeatures. Metalearning has been for several decades successfully used in classification and some other areas of data mining, but its application to the extraction of this kind of rules is novel. Targets: development of a metalearning framework for the extraction of rules with numerice consequents, search for metafeatures in two selected application domains, research into robust metalearning, validation of the metalearning framework on metadata from both application domains.

Predicate graded logics and their applications to computer science

GA17-04630S [Registered results] 2017 - 2019

Classical mathematical logic, built on the conceptually simple core of propositional Boolean calculus, plays a crucial role in modern computer science. A critical limit to its applicability is the underlying bivalent principle that forces all propositions to be either true or false. Propositional logics of graded notions (such as tall, rich, etc.) have been deeply studied for over two decades but their predicate extensions (accommodating, among others, modalities and quantifiers) are still only very partially developed and scarcely applied to particular komputer science problems. The overall goal of the proposed project is to develop predicate graded logics in two complementary directions: (1) studying logical systems in full generality in order to provide a solid mathematical framework and (2) applying achieved results to three particular problems in computer science which heavily involve graded notions: representation of vague and uncertain knowledge, valued constraint satisfaction problems, and modelling of coalition games. We plan to develop predicate graded logics by giving them solid mathematical foundations and applying the achieved results to three particular computer science problems involving graded notions: management of uncertainty, valued constraint satisfaction problems, and modelling of coalition games.

Strength of materials and mechanical components based on iron: Multi-scale approach

GA17-12925S [Registered results] 2017 - 2019

Ductile or brittle behavior of cracks is one of the key phenomena which may have a crucial influence on static and dynamic strength of mechanical structures utilizing bcc iron based materials, e.g. ferritic steels. Continuum predictions on ductile/brittle behavior of a central crack under biaxial tension show that the change of called T-stress can change ductile crack behavior to brittle crack extension. We utilize 3D atomistic molecular dynamic (MD) simulations in bcc iron at various temperatures to verify predictions on ductile-brittle transition caused by T-stress. It will be done for central cracked specimens under biaxial tension and as well for edge cracked samples under uniaxial tension, available for experiments. The topic is important for reactor pressure vessels and interpretation of fracture experiments. Another important aim is interconnecting with first-principles calculations for model clusters of restricted size, pointed at cohesive energy, tension and shear strength, atomic configurations and forces at defects, determining interatomistic potential parameters for MD. The aim of the project is to find out influence of T-stress under different temperatures on the ductile-brittle behavior of crack in 3D bcc iron crystals by means of atomistic and multi-scale (with ab-initio interconnected) calculations and comparison with experiments.

Research infrastructure for Fermilab experiments

LM2015068 [Registered results] 2016 - 2019

RI serves for Czech contribution to particle physics research on experiments at Fermilab. It consists of experiments on which Czech physicists collaborate in Fermilab and of infrastructures of the Czech collaborating institutions. Members of RI work on the Fermilab's experiments NOvA, D0 and plan to join a new experiment in two years to contribute to its design and construction. In the Czech Republic it is a RCCPP computing farm and physics laboratory in FZU, cluster for artificial intelligence and neural networks algorithms in ICS and numerical and statistical computing servers at CTU. The whole infrastructure serves for particle physics experiments and for researchers for many years. The RI as top world research environment serves also for education of undergraduate and postgraduate students.