Conveners
Poster session + káva: prezentácie vedeckých výsledkov FMFI UK Zamestnanci Informatika
- Martin Homola
- Kristina Malinovska (Department of Applied Informatics, FMPI CU)
- Robert Lukotka
Metamodelling in ontologies enables the structured representation of complex domains by defining relationships between concepts across multiple levels of abstraction. Subsumption, a core relation in hierarchical reasoning, provides a strong foundation for organizing ontological knowledge. In this work, we build on an extended form of higher-order description logic, denoted $\mathscr{H}...
Attack Trees (ATs) are a widely adopted formalism for modeling security threats. However, their conventional use relies on an unrealistic assumption of perfect knowledge, where the system's entire state and all adversarial actions are fully known. Real-world security interactions are characterized by limited visibility and finite resource constraints for both the attacker and the...
In most eukaryotes, chromosomal DNA terminates with tandem repeats of a short G-rich motif, such as the canonical TTAGGG sequence. Here we report that nuclear chromosomes of several basidiomycetous yeasts classified into the order Microstromatales carry unusual telomeres. We demonstrate that instead of TTAGGG-like repeats these telomeres are composed of unique tandem arrays which are in most...
Malware analysis increasingly relies on machine learning classification. In this mission-critical domain, analysts require deeper insights that justify the classification results and help them understand how the results were reached. In the project EMA (NextGenerationEU/Recovery and Resilience Plan project No. 09I05-03-V02-00064), we focus on applications of eXplainable (XAI) methods that...
Pangenomes are becoming increasingly popular data structures for genomics analyses due to their ability to compactly represent the genetic diversity within populations. Constructing a pangenome graph, however, is still a time-consuming and expensive process. A promising approach for pangenome construction consists in progressively augmenting a pangenome graph with additional high-quality...
ACM and IEEE Curricula for Computing science defines seven disciplines of computing. One of the disciplines is Software engineering which deals with development of complex software systems. Software engineering is the important discipline as, according to U.S. Bureau of Labor Statistics, 42% of ‘computer and Information technology jobs’ are software developers and testers. The issue, however,...
The growing demand for large, diverse datasets in AI and machine learning is often limited by the cost, privacy, and complexity of real-world data collection. Synthetic Data Generation offers a powerful alternative-creating realistic, controllable datasets through computer graphics, generative models, and physics-based simulations.
This approach enables precise control over environments,...
Circular colourings are a relaxation of proper graph colourings where we allow real numbers as colours. They serve as a model for scheduling problems in which we have arbitrary starting times instead of aligned slots. We provide an overview of computational methods we successfully used to determine circular chromatic index of small graphs and discuss recent results related to the Upper Gap...
Jedným zo spôsobov, akým dokáže systém umelej inteligencie pomenovať predmet na obraze je spriahnutie extraktora príznakov z obrazu s extraktorom príznakov textu. V takomto spoločnom priestore obrazových a textových príznakov je potom pre príznaky obrazu hľadať akým príznakom textu sú podobné. Robí to takto napríklad model CLIP, pričom používa kosínusovú podobnosť medzi príznakovým vektorom...
In this paper, we propose a novel approach for recovering focal lengths from three-view homographies. By examining the consistency of normal vectors between two homographies, we derive new explicit constraints between the focal lengths and homographies using an elimination technique. We demonstrate that three-view homographies provide two additional constraints, enabling the recovery of one or...
Artificial neural networks are currently at their prime, mainly due to the bloom of deep learning. Despite their inspiration from brain processes, learning in such systems, based on error backpropagation (BP), is only loosely inspired by the actual neural mechanisms. Learning in the brain is local and makes use of the bidirectional flow of information. Our Universal Bidirectional...
Graph Neural Networks (GNNs) are increasingly applied to cybersecurity tasks such as malware detection, intrusion detection, and program analysis, as they can model structured program representations and capture relational dependencies beyond flat feature vectors. However, their black-box nature poses challenges in security-critical do- mains, where analysts and stakeholders require...
Binary quantization approaches, which replace weight matrices with binary matrices and substitute costly multiplications with cheaper additions, offer a computationally efficient approach to address the increasing computational and storage requirements of Large Language Models (LLMs). However, the severe quantization constraint ($\pm1$) can lead to significant accuracy degradation.
In this...
Sparse Matrix-Vector Multiplication (SpMV) is a fundamental operation in the inference of sparse Large Language Models (LLMs). Because existing SpMV methods perform poorly under the low and unstructured sparsity ($30–90\%$) commonly observed in pruned LLMs, unstructured pruning struggled to deliver real memory reduction or speedup. We propose MACKO-SpMV, a GPU-optimized format and kernel...
