Keynote 1
Reliable and Sustainable AI: From Mathematical Foundations to Next Generation AI Computing
Professor Gitta Kutyniok
Bavarian AI Chair for Mathematical Foundations of Artificial Intelligence
Ludwig Maximilian University of Munich
Abstract
The current wave of artificial intelligence is transforming industry, society, and the sciences at an unprecedented pace. Yet, despite its remarkable progress, today’s AI still suffers from two major limitations: a lack of reliability and excessive energy consumption.
This talk will begin with an overview of recent theoretical advances addressing reliability, including generalization and explainability, which are core aspects of trustworthy AI that are increasingly relevant in light of regulatory frameworks such as the EU AI Act. We will then discuss fundamental limitations of current AI systems, in particular from the perspective of computability and the energy inefficiency of existing digital hardware.
These challenges motivate the need to rethink the foundations of AI computing. In this context, we will introduce neuromorphic computing as a promising paradigm inspired by biological neural systems, focusing in particular on spiking neural networks. We will highlight recent mathematical results and outline how such approaches may enable a new generation of AI systems that are both provably reliable and significantly more energy-efficient.
Biography
Gitta Kutyniok is Bavarian AI Chair for Mathematical Foundations of Artificial Intelligence at Ludwig-Maximilians-Universität München (LMU), and is affiliated with the DLR German Aerospace Center and the University of Tromsø. She received her Ph.D. in mathematics and computer science from the University of Paderborn and completed her habilitation at Justus Liebig University Giessen. She previously held professorships at the University of Osnabrück and TU Berlin, where she was an Einstein Chair, and has held visiting positions at Princeton, Stanford, Yale, Georgia Tech, and Washington University in St. Louis.
Her research focuses on the mathematical foundations of reliable and sustainable artificial intelligence, next-generation AI computing, and automated scientific discovery, with applications in the life sciences, robotics, and telecommunications.
She is a SIAM Fellow (2019), IEEE Fellow (2024), and ELLIS Fellow (2026), and a member of the Berlin-Brandenburg Academy of Sciences and Humanities and the European Academy of Sciences. She has delivered invited and plenary lectures at major international conferences, including ICM (2022) and ICIAM (2023).
She currently serves as Chair of the SIAM Activity Group on Data Science, Vice President for Science & Research Relations at Venture AI Germany, LMU Director of the Konrad Zuse School of Excellence in Reliable AI (relAI), and Head of the Division of Computational and Information Sciences of the European Academy of Sciences.
Keynote 2
Scaling Down for Real-World Impact: Efficient and Robust Deep Learning from High-Resolution Perception to High-Frequency Finance
Tampere University
Abstract
As deep learning models continue to grow in parameter count and computational demand, a critical gap has emerged between laboratory benchmarks and real-world deployment. In domains such as autonomous robot perception, artificial intelligence of things, and high-frequency financial forecasting, the luxury of unlimited computational resources is non-existent. Instead, these fields demand models that are not only accurate but also lightweight, low-latency, and robust to the non-stationary nature of real-world data. This keynote will explore recent advancements in designing efficient deep learning architectures tailored for these three seemingly disparate yet structurally similar domains, and highlights the common methodological approach: the shift from "brute-force" learning to structured efficiency.
Biography
Alexandros Losifidis is a Professor of Machine Learning at Tampere University, Finland. He leads the Computational Intelligence group, and the Fundamental Machine Learning research theme at the Unit of Computing Sciences. He has contributed to more than thirty R&D projects financed by the EU, Finnish, and Danish funding agencies and companies. He has co-authored 100+ articles in international journals and 150+ papers in international conferences/workshops in topics of his expertise, and he co-edited the Deep Learning for Robot Perception and Cognition book (Academic Press, 2022). He served as Associate Editor in Chief (covering the Neural Networks research area) of the Neurocomputing journal between 2021 and 2025. He has been an Associate Editor for leading international journals, including the IEEE Transactions on Neural Networks & Learning Systems, the IEEE Transactions on Artificial Intelligence, and the IEEE Transactions on Circuits and Systems for Video Technology. He contributed to the organization of several international conferences, including as Area Chair for IEEE ICIP (2018-2026), IEEE ICASSP (2023-2025), as Senior Area Chair for IEEE ICASSP 2026.
Keynote 3
(How) Do LLMs Reason
Prof. Subbarao Kambhampati
Arizona State University
Short Abstract
Large Language Models, auto-regressively trained on the digital footprints of humanity, have shown impressive abilities in generating coherent text completions for a vast variety of prompts. While they excelled from the beginning in producing completions in appropriate style, factuality and reasoning/planning abilities remained their Achilles heel (premature claims notwithstanding). More recently a breed of approaches dubbed “reasoning models” (LRMs). These approaches leverage two broad and largely independent ideas: (i) test-time inference – which involves getting the base LLMs do more work than simply providing the most likely completion, including using them in generate and test approaches such as LLM-Modulo (that pair LLM generation with a bank of verifiers) and (ii) post-training methods–which go beyond simple auto-regressive training on web corpora by collecting, filtering and training on derivational traces (that are often anthropomorphically referred to as “chains of thought” and “reasoning traces”), and modifying the base LLM with it using supervised finetuning or reinforcement learning methods. Their success on benchmarks notwithstanding, there are significant questions and misunderstandings about these methods–including whether they can provide correctness guarantees, whether they do adaptive computation, whether the intermediate tokens they generate can be viewed as reasoning traces in any meaningful sense, and whether they are costly Rube Goldberg reasoning machines that incrementally compile verifier signal into the generator or truly the start of a golden era of general purpose System 1+2 AI systems. Drawing from our ongoing work in planning, I will present a broad perspective on these approaches and their promise and limitations.
Biography
Subbarao Kambhampati is a professor of computer science at Arizona State University. Kambhampati studies fundamental problems in planning and decision making, motivated in particular by the challenges of human-aware AI systems. He is a fellow of Association for the Advancement of Artificial Intelligence, American Association for the Advancement of Science, and Association for Computing machinery, and a recent recipient of the AAAI Patrick H. Winston Outstanding Educator award. He served as the president of the Association for the Advancement of Artificial Intelligence, a trustee of the International Joint Conference on Artificial Intelligence, the chair of AAAS Section T (Information, Communication and Computation), and a founding board member of Partnership on AI. He received his B.Tech. from IIT Madras, and MS and PhD from University of Maryland. He is a distinguished alumnus of both. Kambhampati’s research as well as his views on the progress and societal impacts of AI have been featured in multiple national and international media outlets. He can be followed on Twitter .