Műegyetemi Digitális Archívum
 

BME MIT PhD Minisymposium, 2022, 29th

Permanent URI for this collectionhttp://hdl.handle.net/10890/16854

Browse

Recent Submissions

Now showing 1 - 18 of 18
  • Item
    ZKP-Based Audit for Blockchain Systems Managing Central Bank Digital Currency
    (2022) Péter, Bertalan Zoltán; Kocsis, Imre
    Central Bank Digital Currency (CBDC) systems are being developed around the world and production solutions can be expected in the near future. Should a central bank allow handling of CBDC on a ledger that is not under its supervision (via platform bridging), it may wish to specify certain conformance requirements regarding the transactions. We propose a novel audit scheme based on Zero-Knowledge Proofs, which allows the operator of the bridged ledger to prove its compliance to such requirements, without revealing details about the transactions (such as the exact participants, the direction of the transfer, or the transferred value). This scheme aims to resolve the conflict between banks having to audit how CBDC is used on the bridged blockchain and consortia trying to keep sensitive data private.
  • Item
    Using Dimension Reduction Methods on the Latent Space of Molecules
    (2022) Józsa, György; Sárközy, Péter
    De novo molecule design is the process of generating novel chemicals based on a dataset of drug-like molecules. This method has gained popularity in recent decades. Developing drug-like molecules is both costly and time-consuming. To speed the process up, machine learning and deep neural networks have been used in the last three decades. A particularly popular method is using a variational autoencoder to generate a latent space of drug-like molecules suitable for targeted searching. Quantifying the quality of such a latent space is vital for effective usage. This task is not trivial however, as the chemical structure of molecules cannot be easily quantized and such latent spaces tend to be high-dimensional, leading to the need for dimension reducing visualization algorithms to be applied. Many dimension reduction and visualization algorithms have been developed in recent decades. In this paper, we evaluate five recent algorithms – PCA, t-SNE, UMAP, TriMAP and PaCMAP – to see how well they perform on a given dataset. We examine each algorithm on its ability to transform a 64-dimensional latent space such that the resulting two-dimensional space is smooth over chemical structure. We optimize the hyperparameters of each algorithm to see how they transform the resulting embedding and perform a linear interpolation test to see how they map the latent space into two dimensions. We examine the invertibility and extensibility of each algorithm, as this can make targeted searching much easier to execute.
  • Item
    The Effect of Transition Granularity in the Model Checking of Reactive Systems
    (2022) Szkupien, Péter; Molnár, Vince
    The Theta model checking framework offers the eXtended Symbolic Transition System (XSTS) formalism as a target language for the transformation of high-level models to verify. In XSTS, multiple symbolic transitions can be defined by imperative and declarative statements. The language is flexible enough to offer a broad variety of expressing the semantics of high-level models (e.g., statecharts). Two extremes are i) encoding every (possibly non-deterministic) atomic behavior of the high-level model into a single transition (big steps with only stable states) or ii) modeling the control flow of the computation of the next state (small steps with transient states). Experience shows that big steps are efficient in reducing the state space but sometimes yield transitions that are too complex to handle. Furthermore, internal non-determinism in "big-step" transitions is hard to back-annotate from a counterexample to the high-level model. We examine the effect of transition granularity on model checking by applying a post-processing step that can split "big-step" transitions.
  • Item
    Using Invertible Plugins in Autoencoders for Fast and Customizable Post-training Optimization
    (2022) Pogány, Domonkos; Sárközy, Péter
    One of the main motivations for modern drug research is the production of new compounds that act as drugs, however developing a new drug is an excessively time and resource intensive process. Deep generative neural networks might provide a solution. With their help, we may be able to search in a continuous latent space to find drug molecules that are not yet known but have suitable chemical and structural properties (e.g. solubility, interaction with a given target protein). In this paper, we propose a model which can generate novel drug candidates, that are suitable for a pre-specified objective function of arbitrary properties. The model consists of a generative network and a predictor. The former is an autoencoder which utilizes attention to handle the textual representation of molecules, while the latter uses matrix factorization to predict drug-target interactions (DTI). With a genetic algorithm we can generate novel compounds from the continuous latent space, but if there are changes in the objective function, we may need to train the whole model again. This problem is typical of conditional generative models, to address it, we separated the predictor from the pretrained autoencoder thus forming the plugin. In addition to getting a flexible architecture without any deterioration in the so far achieved results, our model can also be used in a distributed setup by concatenating the plugins. In this way, the objective function can be broken down to smaller subtasks, which can be solved by different plugins without sharing any data.
  • Item
    Towards Hand-Over-Face Gesture Detection
    (2022) Révy, Gábor; Hadházi, Dániel; Hullám, Gábor
    Facial microexpressions are immediately appearing reactions on the face that indicate various details about people's mental and emotional states. Their most important property is that their interpretation is identical or very similar for people all over the world. At present, their identification requires a psychologist expert. Thus automating this task would enable a broader application. The goal of this research is the detection of microexpressions using hybrid expert algorithms. Our algorithms mainly rely on landmark point detectors. Based on their output, several expert algorithms are utilized to extract key features and changes appearing on the face of a subject. These algorithms usually include several steps of image processing and time series analysis algorithms. In this paper, a component responsible for detecting hand gestures and hand pose is introduced. This component helps other algorithms to eliminate false positive detections by detecting the hands over the face. In addition, the recognizability of hand-over-face gestures is investigated. Finally, the implemented face occlusion detector method is evaluated on videos.
  • Item
    Semantically Enabled Design for Edge Cyber Physical Systems
    (2022) Lengyel, Nándor; Kocsis, Imre
    Sensor and computational diversity and redundancy enable radically different implementations of the same functionality in the field and edge domains of cyber-physical systems (CPSs). This diversity and redundancy will also become cornerstones of reconfiguration-based resilience in CPSs, but to truly exploit them, the various provided and required services must be matched semantically. We present the prototype of an integrated, semantically enabled design toolset for data flow-centric CPS systems. A data stream graph DSL based on the W3C SOSA/SSN ontologies serves for high-level specification; semantically enabled function allocation in the Stardog knowledge graph platform creates high-level deployment models, exported in the TOSCA format. Based on the data flow semantics, implementation artifacts for the TOSCA model are automatically generated. These constitute Kubernetes containers for stream processing, and OMG DDS or Hyperledger Fabric for data stream graph nodes.
  • Item
    Physical Activity Recognition Based on Machine Learning
    (2022) Jurčić, Krunoslav; Magjarević, Ratko
    The following paper presents a comparison study of various machine learning techniques in recognition of activities of daily living (ADL), with special attention being given to movements during human falling and the distinction among various types of falls. The motivation for the development of physical activity recognition algorithm includes keeping track of users' activities in real-time, and possible diagnostics of unwanted and unexpected movements and/or events. The activities recorded and processed in this study include various types of daily activities, such as walking, running, etc., while fall activities include falling forward, falling backward, falling left and right (front fall, back fall and side fall). The algorithm was trained on two publicly available datasets containing signals from an accelerometer, a magnetometer and a gyroscope.
  • Item
    Pole Optimization of IIR Filters Using Backpropagation
    (2022) Horváth, Kristóf; Bank, Balázs
    Audio signal processing is a field where specialized techniques are used to account for the characteristics of hearing. In filter design the resulting transfer function need to follow the specification on an approximately logarithmic frequency scale, which can be done via methods such as frequency warping or fixed-pole parallel filters. Although these IIR filter design techniques are proven in practice, they do not produce optimal pole sets for the given specification. In this paper we present the first experiments of using a gradient-based pole optimization framework implemented in TensorFlow by realizing the IIR filter as a recurrent neural network (RNN). The method can improve the pole set of a filter compared to the initial pole set, resulting in a smaller approximation error. The proposed method is demonstrated using four example filter specifications.
  • Item
    The Conceptual Framework of a Privacy-Aware Federated Data Collecting and Learning System
    (2022) Alekszejenkó, Levente; Dobrowiecki, Tadeusz P.
    The federated learning methods offer a strong background of fusing and publishing simultaneously collected data. One of the most challenging problems in federated learning is to hide the identity of the participants. Privacy-preserving techniques try to perturb the participants' local data to match its distribution to the global data. In this paper, we consider that agents collect local environmental data. Neighboring agents can share some of their raw data to support real-time decisions and reduce deviation from the global data distribution. The agents will fuse their collected data into a global model that supports the long-term decision and plan making. We assume the specific situation where the necessary communication protocol between the agents may lead to sharing too much local raw data uncovering private and sensitive attributes of the data sharers. To handle privacy issues, we introduce a privacy-aware framework. Within this framework, local participants balance the amount of the shared raw data to make it informative enough yet not revealing, effectively bounding the loss of privacy. In this study, we use autonomous vehicle agents as an example to demonstrate the concepts of the proposed framework.
  • Item
    Investigating the Combined Application of Mendelian Randomization and Constraint-Based Causal Discovery Methods
    (2022) Vetró, Mihály; Bankó, Márton Bendegúz; Hullám, Gábor
    Mendelian randomization (MR) is often used in medical studies and biostatistics, to reveal direct causation effects between exposures and diseases, typically the effect of some exposure (like chemicals, habits and other factors) to a known disease or disorder. However, this procedure has some strict prerequisites, which often do not comply with the known variables, or the exact causal structure of the variables is not known in advance. In this study, we investigate the use of constraint-based causal discovery algorithms (PC, FCI and RFCI) to produce a sufficient causal structure from the known observations, to aid us in finding variable triplets, upon which MR can be performed. In addition, we show that the validity of MR cannot always be determined based on its results alone. Finally, we investigate the application of the MR principle to determine the direction of causality between variable-pairs, which is a problem most constraint-based causal discovery methods struggle with.
  • Item
    Intrabody Communication Methods – A Short Overview
    (2022) Roglić, Matija; Lučev Vasić, Željka
    This paper is an introduction to underlying mechanisms and the current state of technologies in the field of intrabody communication (IBC). IBC technologies utilize the human body as a communication channel to achieve communication between different devices that can be positioned inside or on the surface of the body. Current developments in the field of mobile gadgets, smartwatches and medical devices make this field of particular interest due to very low energy expenditure, security, and ability to protect private data. Since there are multiple subfields in the field of IBC technologies, this paper will focus on summarizing the principles of work for galvanic coupling and capacitive coupling and compare the tradeoffs of using one approach over the other.
  • Item
    Heterogeneous Federated CubeSat System: Problems, Constraints and Capabilities
    (2022) Batista, Carlos; Mattiello-Francisco, Fatima; Pataricza, András
    Different arguments were being presented in the last decade about CubeSats and their applications. Some of them address wireless communication (5G and 6G technologies) trying to achieve better characteristics as coverage and connectivity. Some arrived with terms as IoST (Internet of Space Things), Internet of Satellites (IoSat), DSS (Distributed Space Systems), and FSS (Federated Satellite Systems). All of them aim to use Small/NanoSatellites as constellations/swarms is to provide specific services, share unused resources, and evolve the concept of satellites-as-a-service (SaS). This paper aims to emophasize performance attributes of such cyber-physical systems, model their inherent operational constraints and at the very end, evaluate the quality of service in terms of figures of merit for the entering/leaving of new heterogeneous constituent systems, a.k.a satellites, to the constellation. This "whitepaper"-styled work focuses on presenting the definitions of this heterogeneous constellation problem, aims at its main capabilities and constraints, and proposes modeling approaches for this system representation and evaluation.
  • Item
    Design of an Audio Frequency Range Distributed Data Acquisition System Prototype
    (2022) Wiesner, András
    Use of packet-based real-time audio and video transmission gets more and more common nowadays. These systems consist of precisely synchronized distributed nodes, the synchronization is usually done using the IEEE 1588 Precision Time Protocol. For example, the well-known Dante system also utilizes PTP as the synchronization solution, and Audio Video Bridging (IEEE 802.1BA-2011) as well. In this paper I introduce a prototype system designed for hardware-level sampling synchronization based on the STM32H743 480MHz Cortex-M7 microcontroller with the aim of describing the synchronization algorithm. A custom extension board has been also made tailored to the NUCLEO development board featuring the MCU to provide us with the required analog and the synchronized I2S-interface This board gives a place for the TLV320AIC23BPW stereo CODEC performing A/D and D/A conversions. This system is designed for audio-frequency range – that's why I use audio a CODEC instead of discrete A/D and D/A converters.
  • Item
    Approximate Time-Optimal Model Predictive Control of a SCARA Robot: A Case Study
    (2022) Cseppentő, Bence; Swevers, Jan; Kollár, Zsolt
    This paper investigates, based on a case study of a SCARA robot, how time-optimal point-to-point motion can be approximately realized using a model predictive control formulation that has low computational complexity. Time-optimality is realized by an indirect formulation and different objective functions are compared. Using the ℓ1-norm instead of a quadratic penalty mimics time-optimality, however, to reduce computational complexity and evade some of its disadvantages, the Huber-norm is introduced as a velocity penalty while the position is penalized quadratically. The contribution of the sampling rate and the length of the prediction horizon is also examined, as the sampling rate poses a limit on the available computation time, while the prediction horizon influences computational complexity. Simulations were carried out in a MATLAB environment using the CasADi toolbox.
  • Item
    Dependability Modeling of Cyber-Physical Systems in the Gamma Framework
    (2022) Szabó, Richárd; Vörös, András
    Cyber-physical systems (CPS) can be found everywhere: smart homes, autonomous vehicles, aircrafts, healthcare, agriculture, and industrial production lines. CPSs are often critical, as system failure can cause serious damage to property and human lives. Today's cyber-physical systems are extremely complex, heterogeneous systems, so rigorous engineering approaches are needed both at design and runtime. On one hand, model-based techniques support the efficient system design, and on the other hand, fault-tolerant middleware and communication technologies support the reliable operation of critical CPS. However, modeling dependability-related system aspects is far from trivial. In this paper, our goal is to show a methodology that introduces design patterns for dependability modeling in the Gamma modeling framework to take a step towards the efficient design of dependable CPSs.
  • Item
    Bayesian Analysis of Multi-Target Genetic Markers Using Hierarchical Phenotypic Data
    (2022) Nagy, Tamás; Eszlári, Nóra; Juhász, Gabriella; Antal, Péter
    Hierarchical data is ubiquitous in healthcare, but taking advantage of hierarchic information is still an open problem. We demonstrate the strengths and weaknesses of Bayesian multilevel analysis (BMLA) in this scenario by characterizing the multi-target relationships between genetic and phenotypic variables. The BMLA method does not scale well with the number of variables, thus we performed filtering using standard pairwise genome-wide association analysis. In this first GWAS phase, we selected only the most significant genotypic variables for further screening. We worked with various thresholds, resulting in 274, 12, and 2 genetic variables, these were treated as interventional variables in the second phase of data analysis using BMLA. The hierarchy of the phenotypic variables was given a priori, thus, we could estimate the posteriors of the relevance, i.e., for direct, non-mediated interaction between these broader categories and genetic markers. Additionally, we could approximate the joint relevance of genetic markers for multiple phenotypic groups, such as for mental health, metabolic and cardiovascular descriptors. The existence of such multi-target (pleiotropic) genetic factors is already indicated by significant genetic correlation between disease groups, i.e., by the overlap of their genetic background. Using this two-stage Bayesian systems-based method, we could robustly induce posteriors for these multivariate and multi-target relevance relations. The systematic investigation of a varying number of genetic and phenotypic factors confirmed that the method is sensitive enough to highlight the weak effects of genetic variables that can be easily overshadowed by the strong phenotypic correlations.
  • Item
    An Initial Performance Analysis of Graph Predicate Evaluation over Partial Models
    (2022) Ficsor, Attila; Semeráth, Oszkár
    Graph-based modeling tools are widely used during the design, analysis and verification of complex critical systems. Those tools enables the automation of several design steps (e.g., by model transformation), and the early analysis of system designs (e.g. by test generation). The evaluation of complex graph predicates (or graph pattern matching) is a core technique in modeling and model transformation, and essential in scalable graph generation. This motivated the integration of industrial graph pattern matching tools directly to advanced data structures used in model checking and logic reasoning algorithms. In this paper we provide a report of a preliminary performance benchmark combining the incremental graph pattern matching algorithm of the Viatra framework with hash tries used for state space exploration on partial models.
  • Item
    Application of Coherence Function to the Analysis of Compressive Sensing
    (2022) Palkó, András; Sujbert, László
    Compressive sensing has been developed for the sampling of sparse or compressible signals. Strong theorems state that when a signal is sufficiently sparse, its samples can be accurately recovered from random sub-Nyquist measurements. As a consequence, compressive sensing is emerging as a part of various applications, such as image processing, biomedical problems or audio signal processing. Designing a compressive sensing application comprises the selection of many parameters, e.g. data acquisition scheme, compression ratio, reconstruction algorithm, etc. To make these decisions experimentally, a simple criterion to compare several options can prove to be helpful. This paper proposes to use the coherence function as a criterion to evaluate the quality of a signal transmission via compressive sensing. After a brief review of compressive sensing, the usage of the coherence function is presented. Simulation examples illustrate how it can help making the design decisions.