Speakers & Abstracts

Archimedes Science Awardee

Cordelia Schmid

Cordelia Schmid holds a M.S. degree in Computer Science from the University of Karlsruhe and a Ph.D. from the Institut National Polytechnique de Grenoble. She is a research director at Inria and has a part-time affiliation at Google DeepMind. Her research spans computer vision, machine learning, and robotics. Dr. Schmid is a member of the German National Academy of Sciences Leopoldina and has received several awards, including the Körber European Science Prize. 

"Artificial intelligence: past, present and future "

Artificial Intelligence (AI) advances the theory and development of computer systems capable of performing tasks that normally require human intelligence. This rapidly evolving field has key application domains including visual perception, natural language understanding, question answering, and robotics, where recent systems have demonstrated compelling results. 

Current leading approaches rely on two core components: (i) machine learning algorithms and (ii) data design and engineering. In this talk, we will explore the impact of both these elements across various areas of application. Furthermore, we will discuss the evolution of artificial intelligence, discussing its past, present capabilities, and future developments.

AI & Health

Anja Braune

Helmholtz-Zentrum Dresden-Rossendorf
Institute of Radiopharmaceutical Cancer Research

“Deep Learning in Radiopharmaceutical Imaging with PET: Enhancing Cancer Diagnostics through AI”

Cancer remains a leading cause of death worldwide. At the Institute of Radiopharmaceutical Cancer Research at HZDR, we aim to revolutionize cancer diagnosis and therapy through the development of innovative radiopharmaceuticals and advanced precision imaging methodologies. Positron emission tomography (PET) is a non-invasive, in vivo imaging modality that provides functional insights on molecular processes. It plays a crucial role in oncology, as well as in neurology and cardiology. However, increasing data complexity, growing demand for personalized medicine, and medical staff shortages challenge imaging techniques. To address these challenges, we focus on developing convolutional neural network (CNN)-based deep learning (DL) methods to enhance image quality and quantitative accuracy, accelerate analysis, streamline interpretation, and ultimately to promote individualized diagnostics. Selected applications include:
• DL-based respiratory motion correction:
A CNN-based approach compensates for breathing-induced motion artifacts by aligning motion-gated PET frames and without increasing image noise, thereby improving lesion detectability in moving organs while preserving quantitative accuracy.
• DL-based noise reduction:
A CNN-based denoising model replicates the edge-preserving properties of bilateral filtering while eliminating the need for manual parameter tuning, enhancing lesion visibility and usability.
• DL-based delineation and classification of tumor lesions:
a CNN-based model trained on annotated clinical PET datasets fully automatically and robustly detects, delineates, and differentiates metabolically active lesions as primary tumor and Lymph Node metastases —even in challenging cases.
DL-based methods offer large potential for improvement of PET image quality as well as for accelerating quantitative image evaluation. They enable reproducible, user-independent and fast image analysis and interpretation. This contributes to more sensitive diagnosis and promotes precision oncology.

Peter Horváth

HUN-REN Biological Research Centre Szeged, Helmholtz Munich, Hungary/Germany

“Life beyond the pixels: single-cell analysis using deep learning and image analysis methods”

In this talk I will give an overview of the computational steps in the analysis of single cell-based large- scale microscopy experiments. First, I will present a novel microscopic image correction method designed to eliminate illumination and uneven background effects which, left uncorrected, corrupt intensity-based measurements. New single-cell image segmentation methods will be presented using differential geometry, energy minimization and deep learning methods. I will discuss machine learning software tools capable of identifying cellular phenotypes based on features extracted from the image. For cases where discrete cell-based decisions are not suitable, we propose a method to use multi-parametric regression to analyze continuous biological phenomena. To improve the learning speed and accuracy, we propose an active learning scheme that selects the most informative cell samples. Our recently developed Deep Visual Proteomics method fpr single-cell isolation methods, based on laser-microcapturing and patch clamping, utilizes the selection and extraction of specific cell(s) using the above machine learning models. I will show that we successfully performed DNA and RNA sequencing, proteomics, lipidomics and targeted electrophysiology measurements on the selected cells and their usage in personalized precision cancer therapies.

Ivo Sbalzarini

Dresden University of Technology/Max Planck Institute for Molecular Cell Biology and Genetics, Germany

“The Rise of AI in the Life Sciences: Possibilities, Promises, Problems”

Living systems are immensely complex, and we still do not have a clear understanding of what brings matter to life. Artificial Intelligence, however, suggests a new perspective on living systems as information-processing, computing systems. Indeed, cells process information in chemical signaling circuits, and they communicate with one another. This defines an unusual compute architecture, which is different from electronic digital computers. In digital computers, only two states exist: 0 and 1. In cells, however, the state consists of the simultaneous concentrations of tens of thousands of molecules, rending it massively high-dimensional. Detecting patterns in this high-dimensional space is a monumental task for which AI provides a promising tool. This, however, will require integrating physical, chemical, and biological knowledge into the AI algorithms. Living systems exist in the physical world and follow laws of physics. Combining this structural knowledge with data is a main challenge. In this talk, I summarize the currently existing possibilities of using AI to understand living systems on the basis of physics, I speculate about future promises, and I mention some of the challenges and problems the field might encounter on the way.

Martin Sedlmayr

Dresden University of Technology/University Hospital Carl Gustav Carus Dresden, Germany

“From Data Foundations to Ecosystems: Enabling AI in Healthcare Across Sectors and Disciplines”

Artificial Intelligence (AI) is reshaping healthcare—but its potential depends on more than algorithms. It requires access to high-quality data, interoperable infrastructures, and sustainable collaboration across sectors. In Germany, the Medical Informatics Initiative (MII) and the Network University Medicine (NUM) have advanced data integration within academic medicine through semantic standards, governance structures, and data use pathways. As a core contributor to both programs, the University Medicines in Dresden and Leipzig help shape these developments nationally.
However, if AI is to reach patients and support care innovation, its foundations must extend beyond university hospitals. Through our Digital Progress Hubs, we connect primary and secondary care providers, bridging academic and non-academic actors via shared data and infrastructure services.
Our new initiative, KIMed, takes this a step further: as a multidisciplinary, cross-sectoral AI innovation network, it integrates academia, hospitals, public health institutions, computing centers, start-ups, and industry into an overarching ecosystem. The network fosters a secure, federated data environment, develops innovative demonstrators, and strengthens digital literacy through targeted training formats.
By aligning with Saxony’s regional innovation strategy and embedding regulatory, technical, and societal perspectives, KIMed sets an example for how AI readiness, innovation capacity, and digital sovereignty can be built—not through isolated projects, but through enduring ecosystems.

Virgilijus Ulozas

Lithuanian University of Health Sciences, Lithuania

“Artificial intelligence in clinical voice evaluation: from idea to practical implementation”

The contemporary booming of Artificial Intelligence (AI) methods and audio signal processing techniques open new perspectives for the use of voice signals to detect or monitor laryngeal diseases. The emerging results of multidisciplinary research in the field of AI application for clinical voice analysis will be presented in the lecture.  The developed mobile VoiceScreen application (available in 17 different languages now) is meant for automated acoustic voice pathology screening, clinical voice monitoring and assessment of therapeutical and surgical outcomes of laryngeal diseases, it is an accurate and robust tool for voice quality measurement and demonstrates the feasibility to be used in clinical settings. Another development represents the multiparametric Voice Wellness Index (VWI) that integrates voice-related data of two different information sources (i.e., acoustic voice analysis and a glottal function symptom-based questionnaire) for voice assessment.  The VWI was found to be highly efficient in describing differences in voice quality status and discriminating between normal and dysphonic voices based on clinical diagnosis, i.e., dysphonia type, implying the VWI’s reliable voice screening potential. The most recent development is the AI-driven SpeechEnhancer algorithm. It is a promising tool for speech rehabilitation after laryngeal oncosurgery. SpeechEnhancer demonstrates the potential for use in clinical settings both by healthcare professionals and patients following laryngeal carcinoma surgery.

Artur Yakimovich

University of Wroclaw, Poland/CASUS HZDR, Germany

“Generative AI for Inverse Problems in Biomedical Computational Microscopy”

Advanced microscopy techniques, including three-dimensional, super-resolution and quantitative phase microscopy remain at the forefront of biomedical discovery. These methods enable researchers to visualise complex molecular processes and interactions at the level of single molecules or molecular complexes, capturing yet unseen information and pushing the boundaries of our understanding of health and disease. These innovations have been made possible, among others, through rapid progress in biophotonics, as well as computational processing and analysis of image-based data. However, advanced biophotonics comes at the cost of complex equipment, as well as difficult and lengthy data acquisition and necessitates highly-trained personnel. We demonstrate in several works that this hurdle can be addressed using generative and discriminative AI algorithms by formulating the conversion from conventional microscopy modalities like widefield to advanced like super-resolution as a set of inverse problems. We show that incorporating nuance of the data domain into the algorithm design, as well as leveraging synthetic data pre-training, leads to better performance in these algorithms. Among other examples, we demonstrate how Generative AI algorithms can be utilised for Virtual Staining of virus infection in cultured cells, allowing for quasi-label-free detection of infected cells.

AI & Energy

Tomasz Bulik

University of Warsaw, Poland

“Einstein Telescope: status and challenges”

Einstein Telescope is the project to build a third generation gravitational wave observatory that will revolutionize the field of gravitational wave astronomy. I will present the current status of the project and sketch the path forward towards its realization. I will concentrate on the challenges that the Einstein Telescope faces including not only technical but also those related to data analysis. I will describe the possibilities of using modern data analysis techniques like machine learning and artificial intelligence in the data analysis of this future facility.

Werner Dobrautz

Dresden University of Technology/CASUS HZDR, Germany

“Quantum, Classical, and Machine Learning Synergies for Next-Generation Energy Materials”

Understanding and designing next-generation energy materials—such as molecular catalysts or unconventional superconductors—requires predictive modeling of complex quantum systems, often governed by strong electron correlations. These challenges remain at the frontier of physics, chemistry, and materials science due to the exponential computational cost of solving many-body quantum problems with classical methods alone.
In this talk, I will explore how synergistic advances in High-Performance Computing (HPC), Artificial Intelligence (AI)/Machine Learning (ML), and Quantum Computing are reshaping our ability to simulate and understand quantum materials relevant to energy science. HPC enables scalable simulations at unprecedented accuracy; AI/ML accelerates optimization, enables surrogate modeling, and uncovers hidden structure–property relationships; and quantum algorithms promise fundamentally new approaches to otherwise intractable problems.
Focusing on applications in transition metal complexes for catalysis and model systems for high-temperature superconductors, I will show how integrating these computational paradigms enables new insights into energy-relevant materials and complements experimental efforts. This convergence opens a transformative pathway toward AI- and quantum-enhanced design of functional materials for sustainable energy technologies.

András Zénó Gyöngyösi

MouldTech Systems Ltd. (industrial partner of the Széchenyi István University), Hungary

“The role of AI in a precision meteorological system, based on atmospheric-profilingdrone data: an environmental input information for the planning of weather-driven renewable energy production”

Environmental data is a key input for the electricity production industry due to the increasing share of weather-driven renewable energy. Vertical profile measurements of the atmospheric variables using light drones adds a significant value to the forecasts by numerical weather prediction models. In my lightning talk the drone data based — so-called – Precision Meteorological System will be introduced, in addition
to the possible use of Artificial Intelligence in the analysis of weather data in support of the planning of energy production, such as Artificial Intelligence powered weather forecasts, analysis of satellite and land-based weather information using machine learning and deep learning techniques, the use of neural network in the short-term prediction of aviation meteorology parameters (e.g.: visibility and low cloud ceiling), and the fine tuning of drone measurements in order to detect the structure of atmospheric variable profiles at a higher accuracy.

Pavel Kordik 

Czech Technical University Prague, Czech Republic

“Recent AI developments and Projects”

In this talk, I will highlight the latest developments in AI relevant to the energy sector, with a focus on how modern machine learning techniques—including deep learning, reinforcement learning, and recommender systems—can be applied to improve efficiency, resilience, and sustainability. I will also present current research initiatives at CTU FIT, including a joint research project with the Helmholtz Institute in Dresden.

Michael Kramer

Max Planck Institute for Radio Astronomy Bonn, Germany/University of Manchester, UK

“Surviving the Data Avalanche (and more) with AI – thanks to radio astronomy”

The next generation of telescopes, especially radio telescopes, will generate more data than the entire global internet. It is obvious that saving these data is not only impossible for many different reasons, but the required energy consumption would also be outrageous. We already face similar problems in radio astronomy today, and it is only thanks to AI that we can select the most relevant data and identify outliers in our data streams. While this enables exciting astronomical discoveries, it also provides valuable lessons for many other data-intensive applications, such as the Internet of Things or smart cities. This talk will illustrate the challenges and provides examples. 

Thomas Kühne

Helmholtz-Zentrum Dresden-Rossendorf

Head of CASUS – Center for Advanced Systems Understanding

“High-performace computing and artificial intelligence for the design of novel energy materials and sustainable reactions”

AI & Microelectronics

Andrius Katkevičius

Vilnius Gediminas Technical University, Lithuania

“Artificial Intelligence in Electronics and Embedded Systems: Applications, Challenges and Future Directions”

Vilnius Tech hosts strong research groups in electronics, microelectronics and artificial intelligence (AI). Currently, collaboration among these groups is becoming increasingly integrated, encompassing both scientific research and educational programs.

This presentation will explore various examples of AI applications across ongoing research activities, including the synthesis and analysis of microwave devices, electrical circuit analysis, sensor data processing and thermal image analysis in the resource-constrained embedded systems. These applications are especially relevant in the context of defense technologies. Neural network-based solutions enable the replacement of traditional analytical and numerical methods, which rely on complex mathematical models, thereby accelerating signal processing and device modeling procedures while maintaining the desired accuracy. This allows for real-time data processing, which is often challenging in embedded systems with limited resources. The presentation will also address how advancements in embedded electronics facilitate the deployment of AI in new application domains.

Vilnius Tech is making significant investments in infrastructure to support research in electronics and AI, including electronic circuit prototyping equipment and servers for training artificial neural networks. These efforts also involve the establishment of competence centers and the development of international partnerships and collaborative projects. Several key recent initiatives will be highlighted.

Akash Kumar

Ruhr University Bochum, Germany

“Reducing AI Energy Footprint: Challenges, Trends and Opportunities“

AI is shifting rapidly to the edge, where privacy, security, and real-time responsiveness matter most. Yet on battery-powered devices, energy is the ultimate constraint. From federated learning to always-on inference, the question is simple but urgent: how can we deliver powerful AI without draining the device?
The answer lies in rethinking how we compute. Emerging approaches—neuromorphic, stochastic, near-memory, and especially approximate computing—offer new ways to balance accuracy, energy, and latency. By pruning and quantizing models, building approximate accelerators, and designing energy-aware operators, we can dramatically cut the cost of the multiply-accumulate operations that dominate AI workloads.
But the design space is enormous. Here, machine learning itself becomes a design partner, guiding exploration, predicting quality, and accelerating convergence toward optimal trade-offs.
This talk explores the challenges, trends, and opportunities in reducing AI’s energy footprint—highlighting approximate, ML-driven design as a near-term solution, and pointing toward future possibilities in 3D circuits with new materials as we push the boundaries of sustainable intelligence at the edge.

Peter Schneider

Dresden University of Technology/Fraunhofer Institute for Integrated Circuits IIS, Germany

“Microelectronics as enabler for future application-oriented AI solutions – challenges, trends, and technologies”

The widespread use of artificial intelligence methods is driving technological progress in all major industries and shaping virtually all areas of society. Depending on the area of application, the requirements for solving specific tasks can vary greatly in terms of performance, cost, energy consumption, availability, etc.

The range of available hardware is diverse. It covers a broad spectrum: from powerful server infrastructures for processing large amounts of data or for training large language models on the one hand, to energy-efficient embedded systems (edge AI) with low latency for specific tasks, e.g., in production or logistics, on the other.
Microelectronics is the key technology. The presentation shows how the rapid implementation of solutions for different requirements is supported and how important technological foundations for future electronic systems are being developed. These include methods for benchmarking and rapidly transferring AI models to commercial hardware platforms, the further development of heterogeneous system integration technologies for implementing scalable, chiplet-based AI accelerators, and novel approaches in the field of quantum and neuromorphic computing.

István Szászi

Bosch Group, Hungary and Adriatic Region/Middle and Eastern European Engineering Cluster

“Boosting Bosch MEMS Sensors with Artificial Intelligence”

In my 15-minute talk, I will explore the transformative role of Artificial Intelligence in Bosch Micro-Electro-Mechanical Systems (MEMS).
During the last decades the size of sensors has been reduced by a factor of 50, on the other hand the accuracy and number of integrated functionalities have been increased rapidly. This growing complexity not only creates new opportunities for AI applications in these domains, but also makes artificial intelligence a fundamental, indispensable toolset for those who aim to compete and remain relevant in the field. Beyond the well-known use cases, like predictive maintenance, this talk will address the challenges faced in MEMS manufacturing and present solutions on how AI streamlines processes. A project in Budapest serves as a compelling example, illustrating how AI-driven analytics and predictive solutions support complete value streams, thereby reducing costs and improving product quality in Bosch’s MEMS sensor manufacturing processes.
Data-driven innovation is clearly essential for maintaining competitiveness in the rapidly evolving microelectronics landscape across all business sectors and ensuring strategic autonomy in it is a key priority for EU. Bosch fosters local ecosystems and cross-border collaboration, establishes strong partnerships with universities and research institutions which are essential for securing Europe’s competitiveness and resilience.

Pranciškus Vitta

Vilnius University, Faculty of Physics

“Beyond Silicon: III–V Compounds for High-Performance AI and Quantum Technologies”

Since the invention of the first silicon transistor in 1947, silicon has powered the growth of the global semiconductor industry and enabled Moore’s Law to transform computing. Yet today, as devices approach the fundamental limits of silicon, the industry faces major barriers in speed, energy efficiency, and thermal management. To meet the demands of high-performance computing, AI datacenters, advanced communications, and quantum technologies, new material platforms are required. Wide-bandgap semiconductors such as SiC, and especially III–V compound semiconductors, are emerging as key enablers of the next technological breakthrough.

Mindaugas Zilys

Kaunas University of Technology/Lithuanian Semiconductor Competence Centre, Lithuania

“Insights on Building Collaborative Ecosystems for AI and Microelectronics Technologies”

Kaunas University of Technology (KTU) demonstrates strong capacity to bridge cutting-edge research with industrial applications, leveraging interdisciplinary strengths in AI and microelectronics. As a member of the European Consortium of Innovative Universities (ECIU), KTU is committed to impactful, collaborative innovation with over 30 active AI-focused projects funded by Horizon Europe and other programs.
KTU is also at the heart of Lithuania’s evolving microelectronics ecosystem. The university consolidates knowledge and infrastructure to support reliable chip manufacturing tailored for industrial sectors including automotive, defence, and healthcare. Key infrastructure is based in Kaunas, where demand for such technologies is high. This demand is supported by major industry leaders such as Littelfuse, Rheinmetall, Continental, Hella, Kitron, Jotron, Carlo Gavazzi, Axioma Metering, Teltonika and so on.
Lithuania’s emerging AI Factory, along with KTU’s AI Excellence Centre, Excellence Centre of the Technology and Physical Sciences TiFEC, National Chip Competence Center Chip-C2 , Rapid design, prototyping and testing of low-volume micro- and nano-electronics products Protolab, and KTU Chip Academy, are forming a robust regional ecosystem. This ecosystem supports rapid development and prototyping of micro/nanoelectronics and fosters AI solutions addressing societal and industrial needs.