Tamás Matuszka | Eötvös Loránd University (original) (raw)

Papers by Tamás Matuszka

NeurIPS 2023 Workshop on Adaptive Experimental Design and Active Learning in the Real World, Jan 14, 2023

Active learning, a powerful paradigm in machine learning, aims at reducing labeling costs by sel... more Active learning, a powerful paradigm in machine learning, aims at reducing labeling costs
by selecting the most informative samples from an unlabeled dataset. However, the traditional active learning process often demands extensive computational resources, hindering scalability and efficiency. In this paper, we address this critical issue by presenting a novel method designed to alleviate the computational burden associated with active learning on massive datasets. To achieve this goal, we introduce a simple, yet effective method-agnostic framework that outlines how to strategically choose and annotate data points, optimizing the process for efficiency while maintaining model performance. Through case studies, we demonstrate the effectiveness of our proposed method in reducing computational costs while maintaining or, in some cases, even surpassing baseline model outcomes. Code is available at https://github.com/aimotive/Compute-Efficient-Active-Learning

Lecture Notes in Computer Science, Jul 20, 2022

Training neural networks to perform 3D object detection for autonomous driving requires a large a... more Training neural networks to perform 3D object detection for autonomous driving requires a large amount of diverse annotated data. However, obtaining training data with sufficient quality and quantity is expensive and sometimes impossible due to human and sensor constraints. Therefore, a novel solution is needed for extending current training methods to overcome this limitation and enable accurate 3D object detection. Our solution for the above-mentioned problem combines semipseudo-labeling and novel 3D augmentations. For demonstrating the applicability of the proposed method, we have designed a convolutional neural network for 3D object detection which can significantly increase the detection range in comparison with the training data distribution.

Cornell University - arXiv, Nov 17, 2022

Autonomous driving is a popular research area within the computer vision research community. Sinc... more Autonomous driving is a popular research area within the computer vision research community. Since autonomous vehicles are highly safety-critical, ensuring robustness is essential for real-world deployment. While several public multimodal datasets are accessible, they mainly comprise two sensor modalities (camera, LiDAR) which are not well suited for adverse weather. In addition, they lack farrange annotations, making it harder to train neural networks that are the base of a highway assistant function of an autonomous vehicle. Therefore, we introduce a multimodal dataset for robust autonomous driving with long-range perception. The dataset consists of 176 scenes with synchronized and calibrated LiDAR, camera, and radar sensors covering a 360-degree field of view. The collected data was captured in highway, urban, and suburban areas during daytime, night, and rain and is annotated with 3D bounding boxes with consistent identifiers across frames. Furthermore, we trained unimodal and multimodal baseline models for 3D object detection. Data are available at https: //github.com/aimotive/aimotive_dataset.

Lecture Notes in Computer Science, 2016

In this paper, we introduce the K-Culture Time Machine Project, which develops a mobile AR platfo... more In this paper, we introduce the K-Culture Time Machine Project, which develops a mobile AR platform for visualizing time-space connected cultural contents of Korea. Existing AR Applications in cultural heritage domains are currently not interoperable and cannot reuse content. To solve this problem, we developed the modified AR reference model as generic framework of a context-aware AR platform; and we developed the context-aware AR platform with several core technologies according to this model. For back-end, we established the Korean Cultural Heritage Data Model (KCHDM) to aggregate the heterogeneous cultural heritage databases in Korea. We also developed the semi-auto time-space correlation generation module for domain experts. Moreover, we developed the authoring tool to generate the time-space connected AR contents. For the front-end, we developed the vision- and sensor-based spatial data composition technology to perform the solid tracking in outdoor environment and context-aware AR framework. Through the UI/UX and 3D contents, whole technologies are packaged into the mobile AR platform. As a validation process for the application, a mobile AR application for the Korean world cultural heritage was developed. Based on this project, an interoperable AR platform that responds with heterogeneous database would be developed and smart tour guide of the cultural heritage site would be possible.

Lecture Notes in Computer Science, 2016

Lecture Notes in Computer Science, 2013

Lecture Notes in Computer Science, 2013

Mobile Web Information Systems, 2014

An enormous amount of information stored in semantic format has become available nowadays. In ord... more An enormous amount of information stored in semantic format has become available nowadays. In order to browse this huge data, the development of different browsers has become necessary. Today, many browsers available, but these are typically desktop applications. Vast knowledge bases have been created by means of linking the different public datasets. The efficient query of these data is a difficult problem. The currently available browsers typically do not allow browsing over the federated datasets, generally displaying only the information of a specified dataset. In this paper a mobile semantic web information system is presented, which allows smartphones to browse federated semantic datasets. In addition, the details of data integrator middleware and the Android-based client and the evaluation are described.

The Semantic Web. Latest Advances and New Domains, 2015

Due the proliferation of smartphones in everyday use, several different outdoor navigation system... more Due the proliferation of smartphones in everyday use, several different outdoor navigation systems have become available. Since these smartphones are able to connect to the Internet, the users can obtain location-based information during the navigation as well. The users could interactively get to know the specifics of a particular area (for instance, ancient cultural area, Statue Park, cemetery) with the help of thus obtained information. In this paper, we present an Augmented Reality system which uses Semantic Web technologies and is based on the interaction between the user and the smartphone. The system allows navigating through a specific area and provides information and details about the sight an interactive manner. Keywords—Augmented Reality, Semantic Web, Human Computer Interaction, Mobile Application.

2016 12th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2016

In this study, we present a novel Augmented Reality (AR) application for providing contextual inf... more In this study, we present a novel Augmented Reality (AR) application for providing contextual information of cultural heritage sites. Whereas most location-based AR systems are designed to show simple information on a Point of Interest (POI) in the real world, the suggested application offers information related to various cultural heritages including tangible and intangible heritages linked to the POI. is is accomplished by a cultural heritage ontology that aggregates heterogeneous data and reorganizes information in themes and relationships. The users can browse information such as the person who created the POI and events that took place at the location as separate entities. We implemented the application for Injeongjeon and the vicinity of Changdeokgung palace and conducted user studies to determine how people explore and consume contextual information at the heritage site. We provide directions that will be useful for designing information to support heritage site visiting, and...

The paper presents Augmented Reality (AR) as a step b etween reality and virtual reality for the ... more The paper presents Augmented Reality (AR) as a step b etween reality and virtual reality for the benifit of education. After some in troductory example of where AR heads, the technology itself is explained. Then, we descri be the modules of the AR portal that we have developed and how it can be used. We give exam ples of its use in university courses and activities. Furthermore we describe an experime nt that we set up at a secondary school to help in popularizing chemistry and raising grade s of students, besides making then enjoy involvement with the topic. Our results are very po sitive, so we consider our efforts to be a starting point for further developments.

Abstract. In the paper we report on the parallel enumeration of the degree sequences (their numbe... more Abstract. In the paper we report on the parallel enumeration of the degree sequences (their number is denoted by G(n)) and zerofree degree sequences (their number is denoted by (G z (n)) of simple graphs on n = 30 and n = 31 vertices. Among others we obtained that the number of zerofree degree sequences of graphs on n = 30 vertices is G z (30) = 5 876 236 938 019 300 and on n = 31 vertices is G z (31) = 22 974 847 474 172 374. Due to Corollary 21 in

Let a, b, m, and n be integers (0 � ab, 1 � mn). An (a,b,n)-tournament (9) is a directed loopless... more Let a, b, m, and n be integers (0 � ab, 1 � mn). An (a,b,n)-tournament (9) is a directed loopless multigraph T = (V,A), where V = {V1,...,Vn} and if 1 � i < jn, then Vi and Vj are connected with at least a and at most b arcs. The score sequence of T is the non- decreasing sequence of its outdegrees and the score set D = {d1,...,dm} of T is the increasingly ordered set of its outdegrees. We propose four al- gorithms generating score sequences corresponding to any D: Balancing reconstructs the majority of the score sets; Shortening reconstructs all score sets containing at most seven elements and so improves the theorem of Hager (7); Sequencing finds a shortest score sequence corresponding to D, while Diophantine generates all score sequences corresponding to D. The algorithms are based on a new, extended version of the Reid-Yao theorem (25, 34).

Due the proliferation of smartphones in everyday use, several different outdoor navigation system... more Due the proliferation of smartphones in everyday use, several different outdoor navigation systems have become available. Since these smartphones are able to connect to the Internet, the users can obtain location-based information during the navigation as well. The users could interactively get to know the specifics of a particular area (for instance, ancient cultural area, Statue Park, cemetery) with the help of thus obtained information. In this paper, we present an Augmented Reality system which uses Semantic Web technologies and is based on the interaction between the user and the smartphone. The system allows navigating through a specific area and provides information and details about the sight an interactive manner. Keywords—Augmented Reality, Semantic Web, Human Computer Interaction, Mobile Application.

Multimedia Tools and Applications

Acta Cybernetica, 2014

E-Science relies heavily on manipulating massive amounts of data for research purposes. Researche... more E-Science relies heavily on manipulating massive amounts of data for research purposes. Researchers should be able to contribute their own data and methods, thus making their results accessible and reproducible by others worldwide. They need an environment which they can use anytime and anywhere to perform data-intensive computations. Virtual observatories serve this purpose. With the advance of the Semantic Web, more and more data is available in Resource Description Framework based databases. It is often desirable to have the ability to link data from local sources to these public data sets. We present a prototype system, which satisfies the requirements of a virtual observatory over semantic databases, such as user roles, data import, query execution, visualization, exporting result, etc. The system has special features which facilitate working with semantic data: visual query editor, use of ontologies, knowledge inference, querying remote endpoints, linking remote data with local data, extracting data from web pages.

2013 IEEE 4th International Conference on Cognitive Infocommunications (CogInfoCom), 2013

ABSTRACT Social networks like Twitter and Facebook have gained a significant popularity with peop... more ABSTRACT Social networks like Twitter and Facebook have gained a significant popularity with people from all parts of the society in the past decade, providing a new kind of data source for novel social-aware applications. A great majority of the users are online all the time, posting real-time information on various topics including unpredicted events. An accident or a natural disaster is often posted on social networks hours before appearing in traditional news. In this paper, we outline a framework for real-time event detection in Twitter data. In contrast to prior works where the absolute or relative changes in the frequencies of some predefined keywords are taken into account, we introduce a lifecycle for each keyword to be observed, expressing their average behavior (e.g. average frequency changes) over time. As a motivation, we show that some keywords exhibit periodic behavior that can be handled by our model. The proposed lifecycle model enables us to define novel temporal features used by our framework in real-time event detection.

Lecture Notes in Computer Science, 2014

Owing to the remarkable advancement of smartphones, Augmented Reality applications have become pa... more Owing to the remarkable advancement of smartphones, Augmented Reality applications have become part of everyday life. Augmented Reality browsers are the most commonly used among these applications. The users can search and display interesting places from the physical environment surrounding them by means of these browsers. Some of the most popular AR browsers use only one data source and the openly available datasets are not used. In contrast, the main objective of Linked Open Data community project is to link knowledge from different data sources. This pursuit makes it easier to retrieval information, among others. In this paper, an Augmented Reality browser was presented. Information derived from Linked Open Data was used by the browser as data source. Due to this, the system is able to handle more data sources.

NeurIPS 2023 Workshop on Adaptive Experimental Design and Active Learning in the Real World, Jan 14, 2023

Active learning, a powerful paradigm in machine learning, aims at reducing labeling costs by sel... more Active learning, a powerful paradigm in machine learning, aims at reducing labeling costs
by selecting the most informative samples from an unlabeled dataset. However, the traditional active learning process often demands extensive computational resources, hindering scalability and efficiency. In this paper, we address this critical issue by presenting a novel method designed to alleviate the computational burden associated with active learning on massive datasets. To achieve this goal, we introduce a simple, yet effective method-agnostic framework that outlines how to strategically choose and annotate data points, optimizing the process for efficiency while maintaining model performance. Through case studies, we demonstrate the effectiveness of our proposed method in reducing computational costs while maintaining or, in some cases, even surpassing baseline model outcomes. Code is available at https://github.com/aimotive/Compute-Efficient-Active-Learning

Lecture Notes in Computer Science, Jul 20, 2022

Training neural networks to perform 3D object detection for autonomous driving requires a large a... more Training neural networks to perform 3D object detection for autonomous driving requires a large amount of diverse annotated data. However, obtaining training data with sufficient quality and quantity is expensive and sometimes impossible due to human and sensor constraints. Therefore, a novel solution is needed for extending current training methods to overcome this limitation and enable accurate 3D object detection. Our solution for the above-mentioned problem combines semipseudo-labeling and novel 3D augmentations. For demonstrating the applicability of the proposed method, we have designed a convolutional neural network for 3D object detection which can significantly increase the detection range in comparison with the training data distribution.

Cornell University - arXiv, Nov 17, 2022

Autonomous driving is a popular research area within the computer vision research community. Sinc... more Autonomous driving is a popular research area within the computer vision research community. Since autonomous vehicles are highly safety-critical, ensuring robustness is essential for real-world deployment. While several public multimodal datasets are accessible, they mainly comprise two sensor modalities (camera, LiDAR) which are not well suited for adverse weather. In addition, they lack farrange annotations, making it harder to train neural networks that are the base of a highway assistant function of an autonomous vehicle. Therefore, we introduce a multimodal dataset for robust autonomous driving with long-range perception. The dataset consists of 176 scenes with synchronized and calibrated LiDAR, camera, and radar sensors covering a 360-degree field of view. The collected data was captured in highway, urban, and suburban areas during daytime, night, and rain and is annotated with 3D bounding boxes with consistent identifiers across frames. Furthermore, we trained unimodal and multimodal baseline models for 3D object detection. Data are available at https: //github.com/aimotive/aimotive_dataset.

Lecture Notes in Computer Science, 2016

In this paper, we introduce the K-Culture Time Machine Project, which develops a mobile AR platfo... more In this paper, we introduce the K-Culture Time Machine Project, which develops a mobile AR platform for visualizing time-space connected cultural contents of Korea. Existing AR Applications in cultural heritage domains are currently not interoperable and cannot reuse content. To solve this problem, we developed the modified AR reference model as generic framework of a context-aware AR platform; and we developed the context-aware AR platform with several core technologies according to this model. For back-end, we established the Korean Cultural Heritage Data Model (KCHDM) to aggregate the heterogeneous cultural heritage databases in Korea. We also developed the semi-auto time-space correlation generation module for domain experts. Moreover, we developed the authoring tool to generate the time-space connected AR contents. For the front-end, we developed the vision- and sensor-based spatial data composition technology to perform the solid tracking in outdoor environment and context-aware AR framework. Through the UI/UX and 3D contents, whole technologies are packaged into the mobile AR platform. As a validation process for the application, a mobile AR application for the Korean world cultural heritage was developed. Based on this project, an interoperable AR platform that responds with heterogeneous database would be developed and smart tour guide of the cultural heritage site would be possible.

Lecture Notes in Computer Science, 2016

Lecture Notes in Computer Science, 2013

Lecture Notes in Computer Science, 2013

Mobile Web Information Systems, 2014

An enormous amount of information stored in semantic format has become available nowadays. In ord... more An enormous amount of information stored in semantic format has become available nowadays. In order to browse this huge data, the development of different browsers has become necessary. Today, many browsers available, but these are typically desktop applications. Vast knowledge bases have been created by means of linking the different public datasets. The efficient query of these data is a difficult problem. The currently available browsers typically do not allow browsing over the federated datasets, generally displaying only the information of a specified dataset. In this paper a mobile semantic web information system is presented, which allows smartphones to browse federated semantic datasets. In addition, the details of data integrator middleware and the Android-based client and the evaluation are described.

The Semantic Web. Latest Advances and New Domains, 2015

Due the proliferation of smartphones in everyday use, several different outdoor navigation system... more Due the proliferation of smartphones in everyday use, several different outdoor navigation systems have become available. Since these smartphones are able to connect to the Internet, the users can obtain location-based information during the navigation as well. The users could interactively get to know the specifics of a particular area (for instance, ancient cultural area, Statue Park, cemetery) with the help of thus obtained information. In this paper, we present an Augmented Reality system which uses Semantic Web technologies and is based on the interaction between the user and the smartphone. The system allows navigating through a specific area and provides information and details about the sight an interactive manner. Keywords—Augmented Reality, Semantic Web, Human Computer Interaction, Mobile Application.

2016 12th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2016

In this study, we present a novel Augmented Reality (AR) application for providing contextual inf... more In this study, we present a novel Augmented Reality (AR) application for providing contextual information of cultural heritage sites. Whereas most location-based AR systems are designed to show simple information on a Point of Interest (POI) in the real world, the suggested application offers information related to various cultural heritages including tangible and intangible heritages linked to the POI. is is accomplished by a cultural heritage ontology that aggregates heterogeneous data and reorganizes information in themes and relationships. The users can browse information such as the person who created the POI and events that took place at the location as separate entities. We implemented the application for Injeongjeon and the vicinity of Changdeokgung palace and conducted user studies to determine how people explore and consume contextual information at the heritage site. We provide directions that will be useful for designing information to support heritage site visiting, and...

The paper presents Augmented Reality (AR) as a step b etween reality and virtual reality for the ... more The paper presents Augmented Reality (AR) as a step b etween reality and virtual reality for the benifit of education. After some in troductory example of where AR heads, the technology itself is explained. Then, we descri be the modules of the AR portal that we have developed and how it can be used. We give exam ples of its use in university courses and activities. Furthermore we describe an experime nt that we set up at a secondary school to help in popularizing chemistry and raising grade s of students, besides making then enjoy involvement with the topic. Our results are very po sitive, so we consider our efforts to be a starting point for further developments.

Abstract. In the paper we report on the parallel enumeration of the degree sequences (their numbe... more Abstract. In the paper we report on the parallel enumeration of the degree sequences (their number is denoted by G(n)) and zerofree degree sequences (their number is denoted by (G z (n)) of simple graphs on n = 30 and n = 31 vertices. Among others we obtained that the number of zerofree degree sequences of graphs on n = 30 vertices is G z (30) = 5 876 236 938 019 300 and on n = 31 vertices is G z (31) = 22 974 847 474 172 374. Due to Corollary 21 in

Let a, b, m, and n be integers (0 � ab, 1 � mn). An (a,b,n)-tournament (9) is a directed loopless... more Let a, b, m, and n be integers (0 � ab, 1 � mn). An (a,b,n)-tournament (9) is a directed loopless multigraph T = (V,A), where V = {V1,...,Vn} and if 1 � i < jn, then Vi and Vj are connected with at least a and at most b arcs. The score sequence of T is the non- decreasing sequence of its outdegrees and the score set D = {d1,...,dm} of T is the increasingly ordered set of its outdegrees. We propose four al- gorithms generating score sequences corresponding to any D: Balancing reconstructs the majority of the score sets; Shortening reconstructs all score sets containing at most seven elements and so improves the theorem of Hager (7); Sequencing finds a shortest score sequence corresponding to D, while Diophantine generates all score sequences corresponding to D. The algorithms are based on a new, extended version of the Reid-Yao theorem (25, 34).

Due the proliferation of smartphones in everyday use, several different outdoor navigation system... more Due the proliferation of smartphones in everyday use, several different outdoor navigation systems have become available. Since these smartphones are able to connect to the Internet, the users can obtain location-based information during the navigation as well. The users could interactively get to know the specifics of a particular area (for instance, ancient cultural area, Statue Park, cemetery) with the help of thus obtained information. In this paper, we present an Augmented Reality system which uses Semantic Web technologies and is based on the interaction between the user and the smartphone. The system allows navigating through a specific area and provides information and details about the sight an interactive manner. Keywords—Augmented Reality, Semantic Web, Human Computer Interaction, Mobile Application.

Multimedia Tools and Applications

Acta Cybernetica, 2014

E-Science relies heavily on manipulating massive amounts of data for research purposes. Researche... more E-Science relies heavily on manipulating massive amounts of data for research purposes. Researchers should be able to contribute their own data and methods, thus making their results accessible and reproducible by others worldwide. They need an environment which they can use anytime and anywhere to perform data-intensive computations. Virtual observatories serve this purpose. With the advance of the Semantic Web, more and more data is available in Resource Description Framework based databases. It is often desirable to have the ability to link data from local sources to these public data sets. We present a prototype system, which satisfies the requirements of a virtual observatory over semantic databases, such as user roles, data import, query execution, visualization, exporting result, etc. The system has special features which facilitate working with semantic data: visual query editor, use of ontologies, knowledge inference, querying remote endpoints, linking remote data with local data, extracting data from web pages.

2013 IEEE 4th International Conference on Cognitive Infocommunications (CogInfoCom), 2013

ABSTRACT Social networks like Twitter and Facebook have gained a significant popularity with peop... more ABSTRACT Social networks like Twitter and Facebook have gained a significant popularity with people from all parts of the society in the past decade, providing a new kind of data source for novel social-aware applications. A great majority of the users are online all the time, posting real-time information on various topics including unpredicted events. An accident or a natural disaster is often posted on social networks hours before appearing in traditional news. In this paper, we outline a framework for real-time event detection in Twitter data. In contrast to prior works where the absolute or relative changes in the frequencies of some predefined keywords are taken into account, we introduce a lifecycle for each keyword to be observed, expressing their average behavior (e.g. average frequency changes) over time. As a motivation, we show that some keywords exhibit periodic behavior that can be handled by our model. The proposed lifecycle model enables us to define novel temporal features used by our framework in real-time event detection.

Lecture Notes in Computer Science, 2014

Owing to the remarkable advancement of smartphones, Augmented Reality applications have become pa... more Owing to the remarkable advancement of smartphones, Augmented Reality applications have become part of everyday life. Augmented Reality browsers are the most commonly used among these applications. The users can search and display interesting places from the physical environment surrounding them by means of these browsers. Some of the most popular AR browsers use only one data source and the openly available datasets are not used. In contrast, the main objective of Linked Open Data community project is to link knowledge from different data sources. This pursuit makes it easier to retrieval information, among others. In this paper, an Augmented Reality browser was presented. Information derived from Linked Open Data was used by the browser as data source. Due to this, the system is able to handle more data sources.