- Kosice, Kosice, Slovakia
Tomas Horvath
Univerzita Pavla Jozefa Šafárika, Computer Science, Faculty Member
Research Interests:
Research Interests:
Research Interests:
Pig farming is largely characterized by closed, large-scale housing technology. These systems are driven by resource efficiency. In intensive technologies, humans control almost completely. However, there are pig farming systems where... more
Pig farming is largely characterized by closed, large-scale housing technology. These systems are driven by resource efficiency. In intensive technologies, humans control almost completely. However, there are pig farming systems where humans have just little control. These free-range technologies are called organic pig farming systems in which the quality characteristics of the produced meat sold on a premium price are primary. We present the practical difficulties that are challenging in implementing precision pig farming. We characterize the data science methods that determine the reliability our conclusions. This chapter describes the literature on the behavior and production results of pigs, social aspects, and the possibilities of the certified pig meat supply chain. Digital solutions can be implemented to verify and trace the origin of meat products. In our project, Mangalica breeding sows were tagged with passive Radio Frequency Identification (RFID) tags, and a research zone...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Federated learning (FL) is an emerging branch of machine learning (ML) research, that is examining the methods for scenarios, where individual nodes possess parts of the data, and the task is to form a single common model that fits to the... more
Federated learning (FL) is an emerging branch of machine learning (ML) research, that is examining the methods for scenarios, where individual nodes possess parts of the data, and the task is to form a single common model that fits to the whole distribution. In FL, we generally use mini batch gradient descent for optimizing weights of the models that appears to work very well for federated scenarios. For traditional machine learning setups, a number of modifications has been proposed to accelerate the learning process and to help to get over challenges posed by the high dimensionality and nonconvexity of search spaces of the parameters. In this paper we present our experiments on applying different popular optimization methods for training neural networks in a federated manner. 1 Federated Learning Federated learning (FL) [1] is a new paradigm in Machine Learning (ML), that is dealing with an increasingly important distributed optimization setting, that came into view with the sprea...
Research Interests:
With the spread of digitalization across every aspects of society and economy, the amount of data generated keeps increasing. In some domains, this generation happens in such a massively distributed fashion that poses challenges for even... more
With the spread of digitalization across every aspects of society and economy, the amount of data generated keeps increasing. In some domains, this generation happens in such a massively distributed fashion that poses challenges for even collecting the data to build machine learning (ML) models on it, not to mention the processing power necessary for training. An important aspect of processing information that has been generated at users is privacy concerns, that is, users might be unwilling to expose anything that would enable one to draw any conclusion regarding to confidential information they possess. In this work, we present a experiment on a genetic algorithm based federated learning (FL) algorithm, that reduces the data transfer from individual users to the learner to a single fitness value.
Research Interests:
The goal of this paper is to present an efficient load forecasting algorithm for large electric power systems. It uses a combination of nearest neighbor-based load profile clustering and rule-based load forecasting. The load data was... more
The goal of this paper is to present an efficient load forecasting algorithm for large electric power systems. It uses a combination of nearest neighbor-based load profile clustering and rule-based load forecasting. The load data was sliced into daily load curves, which were K-Means-clustered, thereby compressing data and simplifying the solution. K-Means was chosen in the proof of concept phase and will be substituted with more precise solutions later. In the forecasting phase the daily load profile is predicted based on the forecast date, day type (e.g. weekday or weekend) and historical consumption data for similar days in the past. The solution was tested on a large dataset consisting of one year-long, 5-minute measurement data in a 1900-power-line system. The solution showed excellent performance in both the training and forecast phases. It produced meaningful forecasts even when the input data contained significant amounts of anomalies. An additional advantage of the presented...
Research Interests:
Convolutional Neural Network (CNN) has become one of the most popular techniques in image classification. Usually CNN models are trained on a large amount of data, but in this paper, it is discussed CNN usage on data shortage and class... more
Convolutional Neural Network (CNN) has become one of the most popular techniques in image classification. Usually CNN models are trained on a large amount of data, but in this paper, it is discussed CNN usage on data shortage and class imbalance issues. The study is conducted on a small dataset of vine leaves images on a classification task with five classes using two different approaches. In the first approach, a simple CNN model is used, while in the second approach, the Visual Geometry Group (VGG) model with transfer learning is used. It is shown that using different deep learning techniques such as transfer learning, stratified sampling, data augmentation, and the state of arts CNN models such as VGG gives a relatively very good model performance with up to 87% accuracy.
Research Interests:
The field of Artificial Intelligence has always strove towards solving problems with computers that were previously only solvable by humans. An interesting challenge we have these years is extracting information from printed documents. In... more
The field of Artificial Intelligence has always strove towards solving problems with computers that were previously only solvable by humans. An interesting challenge we have these years is extracting information from printed documents. In this publication we focus on the sub-domain of classifying pieces of information on printed invoices. Our goal here was to create a solution capable of finding information on scanned invoices without knowing the template of the invoice. The template-less design is important as invoices can have many different structure based on the issuer. First we feed the invoice image to a commercially available Optical Character Recognition (OCR) engine which returns the extracted texts with their bounding boxes. This information itself wouldn’t be enough so we enrich it with feature engineering. The engineered features give information about the content of the text and the context of the surrounding of the bounding box as well as meta information about the ent...
Research Interests:
Research Interests:
Abstract. In this paper we introduce the method for the integration of ranked distributed data with use of dynamically learned monotone aggregation function. We use the method of monotone graded classification to learn the new aggregation... more
Abstract. In this paper we introduce the method for the integration of ranked distributed data with use of dynamically learned monotone aggregation function. We use the method of monotone graded classification to learn the new aggregation functions in dependency of user preferences on returned objects. This method helps user to specify his/her requirements without any knowledge about the values of attributes of the objects and retrieve more relevant object consecutively. We show our method on an illustrative example.
Research Interests:
Educational assessment plays a central role in the teaching-learning process as a tool for evaluating students’ knowledge of the concepts associated with the learning objectives. The evaluation and scoring of essay answers is a process,... more
Educational assessment plays a central role in the teaching-learning process as a tool for evaluating students’ knowledge of the concepts associated with the learning objectives. The evaluation and scoring of essay answers is a process, besides being costly in terms of time spent by teachers, what may lead to inequities due to the difficulty in applying the same evaluation criteria to all answers. In this work, we present a system for online essay exam evaluation and scoring which is composed of different modules and helps teachers in creating, evaluating and giving textual feedbacks on essay exam solutions provided by students. The system automatically scores essays, semantically, using pair-wise approach. Using the system, the teacher can also give an unlimited number of textual feedbacks by selecting a phrase, a sentence or a paragraph on a given student’s essay. We performed a survey to assess the usability of the system with regard to the time saved during grading, an overall level of satisfaction, fairness in grading and simplification of essay evaluation. Around 80% of the users responded that the system helps them to grade essays more fairly and easily.
Research Interests:
The bottleneck of event recommender systems is the availability of actual, up-to-date information on events. Usually, there is no single data feed, thus information on events must be crawled from numerous sources. Ranking these sources... more
The bottleneck of event recommender systems is the availability of actual, up-to-date information on events. Usually, there is no single data feed, thus information on events must be crawled from numerous sources. Ranking these sources helps the system to decide which sources to crawl and how often. In this paper, a model for event source evaluation and ranking is proposed based on well-known centrality measures from social network analysis. Experiments made on real data, crawled from Budapest event sources, shows interesting results for further research.
Research Interests:
Collaborative filtering one of the recommendation techniques has been applied for e-learning recently. This techniqu e makes an assumption that each user rates for an item once. However, in educa tional environment, each student may... more
Collaborative filtering one of the recommendation techniques has been applied for e-learning recently. This techniqu e makes an assumption that each user rates for an item once. However, in educa tional environment, each student may perform a task (problem) several times. Thus, applying original collaborative filtering for student's task recommen dation may produce unsatisfied results. We propose using context-aware models to utilize all interactions (performances) of the given student-ta sk pairs. This approach can be applied not only for personalized learning envir onment (e.g., recommending tasks to students) but also for predicting student performance. Evaluation results show that the proposed approach works bette r than the none-context method, which only uses one recent performance.
Personalization approaches in learning environments are crucial to foster effective, active, efficient, and satisfactory learning. They can be addressed from different perspectives and also in various educational settings, including... more
Personalization approaches in learning environments are crucial to foster effective, active, efficient, and satisfactory learning. They can be addressed from different perspectives and also in various educational settings, including formal, informal, workplace, lifelong, mobile, contextualized, and selfregulated learning. PALE workshop offers an opportunity to present and discuss a wide spectrum of issues and solutions. In particular, this fifth edition includes 6 papers dealing with adapting the study plan (with highlighting), student’s performance (i.e., academic distress), self-regulating learning skills, interoperability in learner modelling by integrating standards (i.e., IMS specification), confusion detection by monitoring mouse movements in a computer game, and knowledge acquisition of mathematical concepts.
Background Classification of EEG signals is the common theoretical background of various EEG-related recognition tasks, such as the recognition of symptoms of diseases. We consider these tasks as time series classification tasks for which... more
Background Classification of EEG signals is the common theoretical background of various EEG-related recognition tasks, such as the recognition of symptoms of diseases. We consider these tasks as time series classification tasks for which models based on dynamic time warping (DTW) are popular and effective ( Dau et al., 2018 , Buza et al., 2015 ). According to a recent study ( Dau et al., 2018 ), setting the appropriate warping window size (WWS) is crucial for the accuracy in various applications. Materials and methods Using a publicly available EEG dataset ( https://archive.ics.uci.edu/ml/datasets/eeg+database ), we examined whether the WWS is crucial in case of EEG classification. We considered two DTW-based methods, in particular, the wide-spread nearest neighbor and a more advanced approach, PROCESS ( Buza et al., 2015 ). We performed disease recognition experiments according to patient-based 10-fold cross-validation and measured the area under receiver operator characteristic curve (AUC) for both models with various WWS. Results A reasonably high accuracy is achieved in a relatively wide rage of WWS. Most importantly, WWS should not be set to zero. The AUC of the best examined model was around 0.875 which we achieved with relatively low WWS corresponding to approximately 170 ms. Conclusions Our observations show that in case of EEG classification, the examined classifiers are much less sensitive to the WWS than suggested by Dau et al. (2018) .
Research Interests:
Machine learning algorithms often contain many hyperparameters whose values affect the predictive performance of the induced models in intricate ways. Due to the high number of possibilities for these hyperparameter configurations, and... more
Machine learning algorithms often contain many hyperparameters whose values affect the predictive performance of the induced models in intricate ways. Due to the high number of possibilities for these hyperparameter configurations, and their complex interactions, it is common to use optimization techniques to find settings that lead to high predictive accuracy. However, we lack insight into how to efficiently explore this vast space of configurations: which are the best optimization techniques, how should we use them, and how significant is their effect on predictive or runtime performance? This paper provides a comprehensive approach for investigating the effects of hyperparameter tuning on three Decision Tree induction algorithms, CART, C4.5 and CTree. These algorithms were selected because they are based on similar principles, have presented a high predictive performance in several previous works and induce interpretable classification models. Additionally, they contain many inte...
Research Interests:
The procure to pay process (P2P) in large enterprises is a back-end business process which deals with the procurement of products and services for enterprise operations. Procurement is done by issuing purchase orders to impaneled vendors... more
The procure to pay process (P2P) in large enterprises is a back-end business process which deals with the procurement of products and services for enterprise operations. Procurement is done by issuing purchase orders to impaneled vendors and invoices submitted by vendors are paid after they go through a rigorous validation process. Agents orchestrating P2P process often encounter the problem of matching a product or service descriptions in the invoice to those in purchase order and verify if the ordered items are what have been supplied or serviced. For example, the description in the invoice and purchase order could be TRES 739mL CD KER Smooth and TRES 0.739L CD KER Smth which look different at word level but refer to the same item. In a typical P2P process, agents are asked to manually select the products which are similar before invoices are posted for payment. This step in the business process is manual, repetitive, cumbersome, and costly. Since descriptions are not well-formed ...
Research Interests:
This work integrates two diploma theses: Logic Programming on Ranked RDF Data and Fuzzy ILP on RDF Data. Both work with fuzzy logic and RDF data, the first one from inductive and the second from deductive point of view. We analyze the... more
This work integrates two diploma theses: Logic Programming on Ranked RDF Data and Fuzzy ILP on RDF Data. Both work with fuzzy logic and RDF data, the first one from inductive and the second from deductive point of view. We analyze the possibilities of using RDF for the purpose of logic programming. This includes defining rules for user ranking, transforming them to database select queries, taking the results as positive examples for ILP and finally learning the rules from data.
In this paper, an experimental comparison of publicly avail- able algorithms for computing intents of all formal concepts and mining frequent closed itemsets is provided. Experiments are performed on real data sets from UCI Machine... more
In this paper, an experimental comparison of publicly avail- able algorithms for computing intents of all formal concepts and mining frequent closed itemsets is provided. Experiments are performed on real data sets from UCI Machine Learning Repository and FIMI Repository. Results of experiments are discussed at the end of the paper.
This paper focuses to a formal model of user preference learning for content-based recommender systems. First, some fundamental and special requirements to user preference learning are identified and proposed. Three learning tasks are... more
This paper focuses to a formal model of user preference learning for content-based recommender systems. First, some fundamental and special requirements to user preference learning are identified and proposed. Three learning tasks are introduced as the exact, the order preserving and the iterative user preference learning tasks. The first two tasks concern the situation where we have the user's rating available for a large part of objects. The third task does not require any prior knowledge about the user's ratings (i.e. the user's rating history). Local and global preferences are distinguished in the presented model. Methods for learning these preferences are discussed. Finally, experiments and future work will be described.
Research Interests:
Traffic itself can be a huge challenge for most commuters regardless of the transportation method of their choice. For example, it is inevitable to experience delays and congestion during rush hours. All commute methods have their own... more
Traffic itself can be a huge challenge for most commuters regardless of the transportation method of their choice. For example, it is inevitable to experience delays and congestion during rush hours. All commute methods have their own specific characteristics when it comes to delays cars and buses suffer from traffic jams and similar principles apply to railways as well. However, the causes of railway delays are not that straightforward and they need further investigation. According to our personal experiences most passengers are not aware of the reasons behind train delays even though they are usually encountered multiple times a day. In this paper I will present possible answers based on the data collected from the publicly available APIs of Hungarian State Railways over the past 1.5 years.
BlaBoO is an open source software initiative, that aims to help find optimal solution for the widest possible range of optimization problems such as, amongst others, hyperparameter tuning of machine learning algorithms. The software is... more
BlaBoO is an open source software initiative, that aims to help find optimal solution for the widest possible range of optimization problems such as, amongst others, hyperparameter tuning of machine learning algorithms. The software is able to optimize the parameters of any black-box function runnable in command line. It needs no cumbersome installation, launching the application requires running a Java executable file because of what the application runs under various platforms. BlaBoO provides a GUI as well as a batch mode processing abilities. It is easy to extend it by new optimization algorithms. BlaBoO is intended not only for managing black-box optimization experiments but also for education purposes. Its features make it a valuable tool for (not only beginner) machine learning practitioners, students and teachers as well.
A technique to detect patterns in student’s program source codes. First, we represent a source code in the form of an Abstract Syntax Tree (AST). The detection of patterns is done with the SLEUTH algorithm for frequent subgraph mining on... more
A technique to detect patterns in student’s program source codes. First, we represent a source code in the form of an Abstract Syntax Tree (AST). The detection of patterns is done with the SLEUTH algorithm for frequent subgraph mining on trees. We provide experiments using real data from a programming course at our university. In the paper, we discuss the relation between patterns and skills as well as some use cases and further directions of our research.
Formal Concept Analysis aims at finding clusters (concepts) with given properties in data. Most techniques of concept analysis require a dense matrix with no missing values on the input. However, real data are often incomplete or... more
Formal Concept Analysis aims at finding clusters (concepts) with given properties in data. Most techniques of concept analysis require a dense matrix with no missing values on the input. However, real data are often incomplete or inaccurate due to the noise or other unforeseen reasons. This paper focuses on using matrix factorization methods to complete the missing values in the input data such that it can be used with arbitrary concept analysis technique. The used matrix factorization model approximates the sparse object-item data matrix by a product of two dense factor matrices, thus, mapping objects and items to a common latent space. The mentioned object-factor and item-factor matrices are obtained by a simple stochastic gradient optimization method. We also investigate how the amount of missing values influences the output of the concept analysis. Two measures, well-known in the information retrieval community, have been used for the evaluation of the proposed framework. Real d...
Research Interests:
In this study, we present a multimodal emotion recognition architecture that uses both feature-level attention (sequential co-attention) and modality attention (weighted modality fusion) to classify emotion in art. The proposed... more
In this study, we present a multimodal emotion recognition architecture that uses both feature-level attention (sequential co-attention) and modality attention (weighted modality fusion) to classify emotion in art. The proposed architecture helps the model to focus on learning informative and refined representations for both feature extraction and modality fusion. The resulting system can be used to categorize artworks according to the emotions they evoke; recommend paintings that accentuate or balance a particular mood; search for paintings of a particular style or genre that represents custom content in a custom state of impact. Experimental results on the WikiArt emotion dataset showed the efficiency of the approach proposed and the usefulness of three modalities in emotion recognition.
Research Interests:
Motivation for this paper are classification problems in which data can not be clearly divided into positive and negative examples, especially data in which there is a monotone hierarchy (degree, preference) of more or less positive... more
Motivation for this paper are classification problems in which data can not be clearly divided into positive and negative examples, especially data in which there is a monotone hierarchy (degree, preference) of more or less positive (negative) examples. We use data expressing the impact of information systems on business competitiveness in a graded way. The research was conducted on a sample of more than 200 Slovak companies. Competitiveness is estimated by Porter’s model. The induction is achieved via multiple use of two valued induction on alpha-cuts of graded examples with monotonicity axioms in background knowledge. We present results of ILP system ALEPH on above data interpreted as annotated rules. We comment on relations of our results to some statistical models.
Research Interests:
ABSTRACT Ground penetrating radar is used to scan the shallow subsurface for detecting buried objects like pipes without corrupting the road surface. Buried objects are represented by hyperbola branches on GPR radargram images. As the... more
ABSTRACT Ground penetrating radar is used to scan the shallow subsurface for detecting buried objects like pipes without corrupting the road surface. Buried objects are represented by hyperbola branches on GPR radargram images. As the manually interpretation of such radargrams is expensive and time consuming, an important goal in this field is to automatize the pipe localization process. A novel approach which is able to automatically estimate the number of buried objects, their horizontal position and their depth from a radargram is presented in this paper. Additionally, this approach delivers the reflection patterns of the buried objects, which may help to estimate their material. The core of our method is a specialised clustering algorithm which detects apexes with appendant hyperbola branch shaped clusters and ignores noise. A hyperbola fitting algorithm delivering hyperbola parameters is applied to this clusters. The pipe locations are estimated by means of the clusters found and the apexes of the fitted hyperbola branches.
... In [16] it is shown that FLP is more suitable for deductive data models and the purpose of ... We base our procedural semantics on the "backward usage of modus ponens" (no refutation nor... more
... In [16] it is shown that FLP is more suitable for deductive data models and the purpose of ... We base our procedural semantics on the "backward usage of modus ponens" (no refutation nor resolution is applied ... P(gm = 0|ca = 0.45, ch = 0.6, co = 0) = 0 that is that the probability is 1 ...
Research Interests:
The 12th conference ITAT’12 Information Technologies–Application and Theory was held at the hotel Magura located in Monková dolina, near Ždiar, Belianske Tatry, Slovakia in September 17-21th, 2012. The conference is a traditional place of meetings for Czech and Slovak computer science communities...more