Position:

TU Wien is Austria's largest institution of research and higher education in the fields of technology and natural sciences. With over 26,000 students and more than 4000 scientists, research, teaching and learning dedicated to the advancement of science and technology have been conducted here for more than 200 years, guided by the motto "Technology for People". As a driver of innovation, TU Wien fosters close collaboration with business and industry and contributes to the prosperity of society.

The aim of this announcement is to look for and attract exceptional candidates, in all topics and areas in Information and Communication Technology. https://www.wwtf.at/funding/programmes/vrg/index.php?lang=EN#VRG24 , öffnet eine externe URL in einem neuen Fenster

Applications are being invited for outstanding early-career scientists (up to eight years after PhD (counting from the submission deadline), interested in building up their independent research group in the field of "Information and Communication Technology" at the TU Wien. The proposed field of research should contribute to the scientific advancement of the ICT field and demonstrate relevance to industry with a strong potential for mid-term technological and societal impact." mit dem "The proposed field of research should contribute to fundamental scientific advancement in the area of AI/ML. A special emphasis of this call is to link AI theory with data and knowledge of other research fields.

The selection follows a two-stage process: In stage one applicants apply for a tenure track professorship at the Technische Universität Wien (Submission deadline: 19.01.2024). In stage two, applicants apply for a WWTF grant together with a proponent from the Technische Universität WIen of the applicant’s choice (deadline on March 19, 2024, see https://www.wwtf.at/funding/programmes/vrg/index.php?lang=EN#VRG24, öffnet eine externe URL in einem neuen Fenster).

Vienna Research Groups for Young Investigators 2024

The 15th Vienna Research Groups for Young Investigators call 2024 is issued for up to three group leader positions with up to € 1.6 million per research group for six to eight years of funding as part of the WWTF’s Information and Communication Technology programme.

  • It addresses Vienna-based universities and research institutions that intend to hire an excellent young researcher from abroad for setting-up and managing an independent research group. Active recruitment procedures are mandatory.
  • The proposed field of research should contribute to fundamental scientific advancement in the area of AI/ML. A special emphasis of this call is to link AI theory with data and knowledge of other research fields.
  • WWTF especially encourages Vienna-based research institutions to propose female group leaders.
  • WWTF takes unconventional research careers into consideration.

Submission deadline: 2pm CET, March 19th 2024.

An online information session will take place on January 19th 2024 from 10:00-12:00am CET on Zoom. Link to the session:

https://us02web.zoom.us/j/88333953868?pwd=T3BSelFGeHZ1YkJXQXRYMFJTRnlmdz09, öffnet eine externe URL in einem neuen Fenster

 

Please apply here:

Link: https://jobs.tuwien.ac.at/Job/223552, öffnet eine externe URL in einem neuen Fenster 

For further information please contact:

  • Scientific questions: the respective group hosts – please see “scientific contact” in each group description
  • Administrative issues: wwtf@tuwien.ac.at 

Possible hosting labs at TU Wien

(in alphabetical order)

Scientific contact person: Sabine Andergassen (194-06)
Webpage of the Research Unit Machine Learning (194-06): https://informatics.tuwien.ac.at/orgs/e194-06, öffnet eine externe URL in einem neuen Fenster

Our main research interests are efficient and effective machine learning and data mining algorithms. Machine learning considers the problem of extracting useful functional or probabilistic dependencies from a sample of data. Such dependencies can then, for instance, be used to predict properties of partially observed data. Data mining is often used in a broader sense and includes several different computational problems, for instance, finding regularities or patterns in data. By efficiency we mean on the one hand the classical computational complexity of decision, enumeration, etc. problems but on the other hand also a satisfactory response time that allows for effectiveness. By effectiveness we mean
how well an algorithm helps to solve a real world problem.

To demonstrate the practical effectiveness of novel learning algorithms, we apply them in various fields such as Chemistry, Material Science, Electrical Engineering, Physics.

The goal of the new research group would be to develop new artificial intelligence methods that contribute to physics at a conceptual level. The vision should be to utilize AI not just as a computational tool, but as a means through which humans can discover new scientific ideas and concepts. Compelling areas of application and research should include the creation of foundational algorithms ranging from scientific discovery of new physics experiments to the analysis of scientific papers to predict and propose new, impactful research directions.

Scientific contact person: Thomas Eiter <thomas.eiter@tuwien.ac.at

Webpage of your present research group: https://informatics.tuwien.ac.at/orgs/e192-03, öffnet eine externe URL in einem neuen Fenster <https://informatics.tuwien.ac.at/orgs/e192-03, öffnet eine externe URL in einem neuen Fenster>

The current AI landscape consists of symbolic approaches (e.g. rule-based systems such as logic programming) and sub-symbolic approaches (e.g.neural networks) which have complementary benefits and downsides.Combining them in so-called hybrid AI has proven fruitful, but challenging. A major bottleneck is the fact that many importantdevelopments are restricted to specific approaches or formal languages and it is not clear how they develop to generalize or adapt these developments.

This research group aims to develop a general methodology for designing hybrid AI components with general quality assurances. This is possible by taking an algebraic perspective, which has previously proven useful forthe study of several concepts in symbolic AI. Such a language-independentmethodology allows the widespread use of hybrid AI without the need fortechnical expertise in both symbolic and sub-symbolic AI.

 

Scientific contact person: Prof. Dr. Johannes Böhm

Webpage of your present research group:

www.tuwien.at/mg/geo/hg, öffnet eine externe URL in einem neuen Fenster

The research group Higher Geodesy at the Department of Geodesy and Geoinformation deals with dynamic processes in the Earth system as manifested in changes of its form, rotation, gravity field, and atmosphere. We use space geodetic observations to and from satellites, most notably the Global Navigation Satellite Systems (GNSS), but also observations with radio telescopes to extragalactic radio sources.

The research group's primary focus revolves around the comprehensive handling and in-depth analysis of geodetic observations. These observations encompass GNSS networks comprising tens of thousands of permanent stations and potentially billions of low-cost GNSS receivers integrated into ubiquitous smartphones. The amalgamation of this vast and diverse data pool with datasets sourced from non-geodetic domains, notably Earth observing data, facilitates precise probing of the troposphere and ionosphere. This endeavor holds promise for significantly enhancing weather forecasting capabilities and establishing preemptive systems for tsunami early warnings, while concurrently facilitating the study of climate change variables. Moreover, leveraging geodetic datasets unveils opportunities for refining Earth's spatial orientation prediction, a crucial facet in space navigation and astronomy. The pioneering advancement of artificial intelligence methodologies and their strategic deployment across diverse datasets stands poised to unravel deeper insights into the dynamic processes governing Earth. This synergy between cutting-edge AI methodologies and multi-faceted datasets promises a paradigm shift, fostering a comprehensive understanding of Earth's intricate systems and unlocking novel perspectives into its evolving dynamics.

 

Scientific contact person: Ronald Blab, ronald.blab@tuwien.ac.at 

Webpage of your present research group: https://www.tuwien.at/en/cee/ibb/zdb/about-ushttps://www.tuwien.at/en/cee/hib/integral

 

AI-driven research requires a way to effectively put “humans-in-the-loop”, so that automated research can be directed in a creative human way. A natural environment for human-centred AI-applications are Digital Twins as they encapsulate real-world processes in a practical digital format and make them machine-readable. Bringing digital twins and AI together has an unrealised synergy potential for human-centred AI-driven research. On the one hand, most AI applications just use data as input and provide some output instead of running in a structured environment like a digital twin. On the other hand, digital twins are ungeneralised artefacts for singular purposes that can rarely be repurposed and sometimes use AI-based tools. Digital Twins both provide an application ground and an execution environment for AI systems in numerous civil engineering applications.

Scientific contact person: Stefan Woltran <stefan.woltran@tuwien.ac.at>

Webpage of your present research group: https://dbai.tuwien.ac.at/, öffnet eine externe URL in einem neuen Fenster

The main topic is Structured argumentation which provides a natural match with research unit DBAI where abstract argumentation is mainly researched.

Structured Argumentation provides the necessary intermediate step between argument mining and core computational methods of argumentation. Main goal of the VRG is to equip arugmentation systems with an explainability layer which is key for several applications domains, e.g. mediation processes or deliberation platforms.

Scientific contact persons: Prof. Wolfgang Wagner
Webpage: https://www.tuwien.at/en/mg/geo/rs

The use of Artificial Intelligence (AI) to model the Earth’s weather and climate system has recently made big headlines, with some research teams and companies having demonstrating that AI-based forecast can be more accurate and faster than physics-based predictions at a fraction of the compute power. This has also spurred the expectation that it will be possible to use AI for combining Earth observation data collected by satellites with Earth system models to improve the spatial resolution of digital replica of the Earth to 1km and better. This is deemed an essential step for being able to model the combined effect of natural and anthropogenic processes, which is e.g. necessary for planning climate change adaptation and mitigation measures. However, the use of large AI models in Earth observation and Earth system modelling can be problematic for many reasons, including a tendency for overfitting, high correlation between input data sets, and the often strong sensitivity of AI models to unknown processes that are neither described by the Earth system models nor are captured by the satellite observations. These problems are compounded by the fact that AI model outputs are notoriously difficult to interpret, and do not offer easy routes for making them explainable. Therefore, we propose to establish a new Vienna Research Group that carries out basic research for making AI models used to model Earth observation data explainable. This is crucial in applications where transparency, interpretability, and accountability are essential, such as forecasting climatic extremes or quantifying the effectiveness of climate change adaptation and mitigation measures.

 

Scientific contact person: Univ.–Prof. Dr. Gerald Matz

Webpage of your present research group: https://tiss.tuwien.ac.at/fpl/research-unit/index.xhtml?cmnOrgUnitId=1000643, öffnet eine externe URL in einem neuen Fenster

Signal processing has been at the heart of numerous modern approaches for machine learning and artificial intelligence. Examples include compressed sensing for sparse learning, Bayesian statistical signal processing for target tracking and reinforcement learning, data compression and transmission for federated learning, and graph signal processing for learning from big data. The research division “Signal Processing” at the Institute of Telecommunications of TU Wien has a proven track record of internationally competitive research in these areas, covering theoretical analyses, methodological developments, and real-world applications. We now would like to expand the scientific scope of the division by establishing a new research group on graph-based learning with a focus on the design of interpretable and robust graph neural network (GNN) models. This class of neural networks have recently shown great promise in applications ranging from recommender systems to protein folding. The new VRG group “Graph-based Learning for Biomedicine” will be specifically dedicated to generalizing existing modelling paradigms (e.g., convolutional layers, attention mechanisms, transformers) to the GNN context and to exploring applications of this framework in biomedicine and healthcare.

Scientific contact persons: Ass.Prof. Astrid Weiss, E-mail: astrid.weiss@tuwien.ac.at

Webpage of your present research group: http://igw.tuwien.ac.at/hci/, öffnet eine externe URL in einem neuen Fenster

Our Human-Computer Interaction (HCI) group seeks to merge relevant technical / engineering and social sciences research with a practical contribution to the design of technology particularly mobile, tangible and sensor-based technologies. The multidisciplinary group combines disciplines like informatics, engineering, psychology, sociology, medical-informatics, design and media studies. The research application areas are on end-user's inclusion & participation, acceptance & adoption of new technologies, motivations and experiences of users, ethics and social impact of information & communication technologies.The group applies interdisciplinary, cutting-edge methodologies in order to conduct user requirement analysis, co-design new technologies through participative design processes and evaluate technologies in use through a mix of lab and field based, short and long-term, user studies.

The new VRG group "Human Augmentation" should complement the HCI group by means of research on the relation of generative AI and our bodily experiences. In the digital age, we can easily outsource our cognitive labour to our devices, apps and AI. We can augment our senses which affects our lived experiences and subsequently our social interactions with other alive, embodied, and affective agents. How can we augment people through AI and ML supported solutions, but keep them active and engaged prosocial members of society.

Scientific contact person: Florian Michahelles
Webpage of your present research group: http://media.tuwien.ac.at, öffnet eine externe URL in einem neuen Fenster

Advances in sensing, processing, and machine learning enable machines to take initiative of interactions and to proactively approach their users, while gradually fading into sensitive application areas with societal and ethical impact. This development is propelled by new modalities of interaction, such as voice, haptic feedback, virtual/augmented reality, emerging sensing opportunities introspecting users and new capabilities of predicting user needs and intentions.
A new group focusing on human-centred AI would implement and evaluate use cases of humans and machines collaborating on a joint goal where machine repetitiveness and human ingenuity would complement each other. Compelling areas of application and research would range from cognitive tasks to physical tasks.

 

 

 Scientific contact person: Agata Ciabattoni <agata@logic.at> and Chris Fermüller <chrisf@logic.at>

Webpage of your present research group: https://informatics.tuwien.ac.at/orgs/e192, öffnet eine externe URL in einem neuen Fenster

Logic offers precision and trust and it is at the base of many applications in various fields. As Machine Learning (ML) models become more developed, we are faced with the problem of understanding how the algorithm found a specific result. Typically the  calculation process is  what is commonly referred to as a “black box", created directly from the data, that we are unable to interpret. ML models on occasions catastrophically fail and can even support poor decisions due to bias.

The use of logic to formalize classification problems in ML and  to provide explanations that can be read by a human is a promising approach. Most of the attention has been given to classical propositional logic so far. But this is hardly a very suitable language to model quantitative problems. Hence, we will mainly explore and apply more appropriate non-classical logics.

Scientific contact person: Thomas Gärtner thomas.gaertner@tuwien.ac.at

Webpage of the the present group: https://ml-tuw.github.io/, öffnet eine externe URL in einem neuen Fenster

Machine learning has great potential to aid medical diagnosis and decision-making. Advances in machine learning can reduce misdiagnoses, avoid unnecessary procedures, provide personalized treatment, and accurately assess medical risks. However, there are several challenges in integraing machine learning in healthcare. They range from dataset curation to the development of transparent and interpretable algorithms. Other challenges include effective communication with medical practitioners, the development of strong theoretical guarantees and the validation using medical standards.

The goal of the research group is to develop solid foundations of data science and machine learning that meet the demands of the medical field. It aims to achieve this through the lens of learning and statistical theory, developing and analyzing state-of-the-art machine learning methods, and working with medical doctors to use them in high-impact clinical applications. It intends to bridge the gap between machine learning and clinical research. On the one hand, it is expected to study and develop new machine learning methods that are robust and interpretable. On the application side, to work with doctors to develop methods for automatic diagnosis and identification of new risk factors from medical diagnosis.

Scientific contact person: Silvia Miksch
Webpage of your present research group: https://www.cvast.tuwien.ac.at/, öffnet eine externe URL in einem neuen Fenster

The Visual Analytics  Group (CVAST) is a research unit of the Institute of Visual Computing and Human-Centered Technology at TU Wien Informatics. Visual Analytics as ”the science of analytical reasoning facilitated by interactive visual interfaces” aims to enable the exploration and the understanding of large and complex data sets by intertwining interactive visualization, data analysis, human-computer interaction, and cognitive and perceptual science.
CVAST  is concerned with computer-tools, methods, and concepts that support humans in coping with complex information spaces. We strive to make complex information structures more comprehensible, facilitate new insights, and enable process of information and knowledge discovery. At this, human abilities as well as users' needs and tasks are central issues to assist in situations where complex decisions need to be made. Data and information are a broad field – we focus particularly on the temporal, spatial and spatial-temporal dimensions.

The new WWTF VRG group will complement CVAST by means of mixed-initiative Artificial Intelligence (AI) and Visualization strategies to push our problem-solving and decision-making capabilities beyond the traditional AI and Machine Learning (ML) approaches (in the vision of Human-Centered AI, explainable AI, and interactive ML). Such an approach strategically harnesses the collective strengths of human expertise and cutting-edge AI/ML and Visualization techniques to forge a path toward more sophisticated and dynamic problem-solving methodologies. Possible application domains are Health Care/Medicine, Cultural Heritage, Business Intelligence, and Financial Sectors.  

The WWTF YRG will be embedding in the research unit "Visual Analytics" # 193-07
 

Scientific contact person: Thomas Gärtner thomas.gaertner@tuwien.ac.at

Webpage of the the present group: https://ml-tuw.github.io/, öffnet eine externe URL in einem neuen Fenster

This research group aims to improve the efficiency of the entire LHC simulation and analysis chain by developing flexible, efficient, controllable, and precise ML-based methods. In particular, we plan to investigate how current state-of-the-art generative models can be adapted and generalized to deal with structurally different particle physics datasets, which contain local rich-peaking structures and complex far-reaching correlations. Crucially, these methods need to achieve the necessary speed gains while being precise to at least 1\% level. More specifically, achieving precision is about ensuring accuracy not in single events but in the distribution of thousands or millions of events, owing to the quantum nature of physics, which is very different from many other ML applications. Anticipating a significantly improved generator, we plan to develop fast and accurate simulation-based inference (SBI) methods complementing current LHC analyses. Simultaneously, we plan to seamlessly integrate our newly developed methods into a publicly available ML and event generator framework and deploy it to domain experts from particle physics. Importantly, all developed techniques require a comprehensive uncertainty treatment and understanding of regions of low precision. Leveraging these ML techniques, not as mere tools but as an integral research component, we will crucially contribute toward explainable AI by adopting innovations from the AI community and integrating physicists' expertise in uncertainties and symmetries.

For effective dissemination, this group will publish on open-access platforms like ArXiv and in high-impact peer-reviewed journals. Further, they will present the findings and research results at international conferences, such as NeurIPS and ICML (computer science), or ML4Jets, Boost, and ACAT (particle physics), amplifying their reach within the scientific community. In the mid to long term, they will encourage future synergies between both disciplines, ensuring a seamless knowledge exchange and fostering innovative research directions. This collaborative approach is pivotal for realizing the full potential of data-driven techniques and achieving groundbreaking advancements in our understanding of the universe.

Scientific contact person: Thomas Lukasiewicz <thomas.lukasiewicz@tuwien.ac.at>
Webpage of your present research group: https://informatics.tuwien.ac.at/orgs/e192-07, öffnet eine externe URL in einem neuen Fenster

The last years have been landmark years in artificial intelligence (AI): Deep learning, an advanced machine learning technology based on neural networks, is achieving revolutionary results in tasks such as speech recognition/generation, computer vision, language-related tasks, game playing, and self-driving vehicles. Compared to information processing in the human brain, however, there are still at least two very striking differences and serious drawbacks of deep learning, namely, that the human brain is highly universal and extremely energy-efficient, both very much in contrast to current deep learning technologies. This suggests that while deep learning certainly seems to realize some facets of human intelligence, there are still crucial aspects missing from how information processing in the human brain really works. A very prominent theory of how the human brain may work is the free-energy principle, which is closely connected to the Bayesian brain hypothesis, and which is partially realized via predictive coding (PC), a theory of brain function in which the brain is constantly generating and updating a mental model of the environment. PC originated in theoretical neuroscience as a model of information processing in the cortex. While these abstract theories of how the brain may work seem very appealing, they still have no concrete implementations to date. However, there has recently been substantial progress on the understanding of PC. In particular, recent work has developed PC into a general-purpose algorithm able to train neural networks using only local computations. In this task, PC shows a substantially greater flexibility against deep learning, since PC-trained networks can function
as classifiers, generators, and associative memories simultaneously, and they can also be defined on arbitrary graph topologies (including cycles). Furthermore, PC-trained networks show a similar performance as deep neural networks in classification tasks, and PC is also closely connected to control theory and applications in robotics. Interestingly, PC is based on a novel form of credit assignment (called prospective configuration), differently from backpropagation-based deep learning, which moves PC closer to how the human brain may work. Furthermore, PC has an efficient approximation based only on local computations without external control, called parallel PC, which may be efficiently implemented on neuromorphic hardware. This VRG will explore PC as an alternative approach to train deep neural networks, especially towards more human-level AI and higher energy efficiency.

The VRG will be embedded in the research unit "Artificial Intelligence Techniques" (E192-07).
 

Scientific contact person: Thomas Gärtner thomas.gaertner@tuwien.ac.at

Webpage of the the present group: https://ml-tuw.github.io/, öffnet eine externe URL in einem neuen Fenster

Contemporary information societies constitute complex adaptive systems that are strongly shaped by two interacting sets of rules: law and code. Recent advances in computing methods and technologies, combined with the increasing availability of data concerning every aspect of our lives, create unprecedented opportunities to tackle our world's biggest challenges, but also unprecedented risks for individuals, societies, and our planet at large. To seize the opportunities and mitigate the risks, we need a productive exchange between computer scientists and legal scholars.

This research group will establish such an exchange in the area of machine learning, with a focus on connecting two complementary movements, responsible machine learning and data-centric machine learning. In particular, we aim to (1) understand the fundamental trade-offs between (various formalizations of) different legal desiderata for machine-learning applications, such as fairness, accountability, explainability, privacy, and security; (2) theoretically and empirically analyze the role of data in building machine-learning systems that align with humanist values and meet human needs; and (3) develop responsible, data-centric machine-learning methods that embrace regulation by design. Honoring the nature of contemporary information societies as complex adaptive systems, our research will pay particular attention to the challenges of learning with relational data (i.e., graphs and networks).

The overarching goal of this research group is to advance responsible data-centric machine learning with the highest international visibility. To this end, we seek (1) to publish at top venues in computer science as well as in top interdisciplinary outlets, (2) to establish and mediate collaborations between computer scientists and domain scientists (especially in law, the social sciences, and the humanities), and (3) to engage with a broad range of stakeholders within and beyond academia (including government, industry, and civil-society actors).

Scientific contact person: Prof. Dr. Ioannis Giannopoulos igiannopoulos@geo.tuwien.ac.at

Webpage of the the present group: https://geoinfo.geo.tuwien.ac.at/, öffnet eine externe URL in einem neuen Fenster

The amount of human geo-data is growing exponentially because of recent developments in areas of IoT and urban computing. At the same time, AI is taking off as a tool in spatio-temporal analysis providing new insights but lacking in transparency and ease of access. To address both the questions on the data and the analysis thereof, Digital Twins could offer a prime venue to enable accessible, reproducible, and explainable Geo-AI. However, Digital Twins are also a buzz word that requires more rigorous definition to underpin Geo-AI research. Nowadays, Digital Twins are specific artifacts created to represent one particular system instead of a generic tool that helps researchers to manage complexity in their work. The new group would develop a consistent theoretical approach for AI-driven Digital Twins and their application as a framework for accessible, reproducible, and explainable Geo-AI on concrete examples of human activity on a geographic scale.

Scientific contact person: Thomas Gärtner thomas.gaertner@tuwien.ac.at

Webpage of the the present group: https://ml-tuw.github.io/, öffnet eine externe URL in einem neuen Fenster

The field of security enables the use of digital systems while decreasing the possibility of an attacker tampering with these systems. Protecting them has become a significant financial factor: companies spend up to 10% of their IT budget on preventative measures. With the recent rise of machine learning (ML), the security and trustworthiness of ML have come into focus. Recent incidents, for example in the area of chatbots, entail the failure of safeguards, data leakage, and learning offensive language from users. Also, Malware, Spam, and CV classification systems are frequently reverse-engineered and circumvented.

Consequently, a large body of work has accumulated in ML security. Yet, the practicality of this research has been questioned: There is little knowledge about ML usage in the industry and corresponding practical threat models. Furthermore, it remains unclear how the exact practical conditions interact with possible defenses. Addressing these research questions requires a multidisciplinary endeavor incorporating human subject research for a deep understanding of industrial settings, the human factor of ML security, and finally domain knowledge to understand specifics of applications that enable or prevent specific types of attacks and defenses. The new group's goal is thus to conduct high-quality research to tackle the problem of practical ML security. This research will encompass existing attacks, applicational differences, human factors, and implementable mitigations and defenses.

Scientific contact person: Univ. Prof. Dr. Ivona Brandic

Webpage of your present research group: http://hpc.ec.tuwien.ac.at/, öffnet eine externe URL in einem neuen Fenster

Short Description: We are inviting VRG applicants with challenging cutting edge proposals targeting utilization of AI/ML to facilitate sustainable and reliable ICT infrastructures linking with data and knowledge from

  • High Performance Computing (HPC) Systems
  • Cloud Computing Systems
  • Edge Computing Systems
  • Hybrid Classic-Quantum Systems