Keynote Speakers

Kyle Chard

The University of Chicago and Argonne National Laboratory, USA

Scientific discovery increasingly relies on integrating large-scale experiments, simulations, and AI models across diverse computing environments. However, existing workflows and middleware often struggle with the complexity of scaling across federated resources and lack the autonomy needed to overcome human bottlenecks. In this talk, I will present a vision for autonomous, agent-driven science enabled by federated serverless systems. I will show how agentic middleware can orchestrate stateful, collaborative agents across heterogeneous resources, relying on serverless computing to provide elastic, on-demand execution at extreme scale. Drawing on our prior work with Parsl, funcX, Globus, and Academy, I will present examples that demonstrate how federated agentic systems can deliver high-throughput, resilient, and scalable scientific computing, enabling more autonomous and efficient discovery.

Kyle Chard is a Research Associate Professor in the Department of Computer Science at the University of Chicago, with a joint appointment at Argonne National Laboratory. His research focuses on systems for scientific computing and AI, particularly the infrastructure for executing tasks, managing data, and deploying AI models and agents across federated scientific resources. He leads the Parsl parallel programming library, Globus Compute (previously funcX), and the Academy agent middleware, which together are used by thousands of researchers. He received the IEEE TCHPC Award for Excellence for Early Career Researchers in High Performance Computing.

Antonio Loureiro

Universidade Federal de Minas Gerais, Brazil

The Internet of Things has become one of the largest sources of data. Industry analysts counted about 18.5 billion connected IoT devices in 2024 and expect roughly 39 billion by 2030, spanning factories, hospitals, homes, farms, vehicles and, increasingly, satellites. Yet data is not understanding. Streams arrive in real time, incomplete, noisy and sometimes simply wrong, and their meaning depends on context that lives outside the sensors themselves, such as weather, energy prices or human routines. This keynote argues for a different path from raw IoT data to useful action. A semantic construction process distributed across the edge–cloud–space computing continuum turns heterogeneous streams into structured, quality-annotated descriptions of what is happening. A semantic learning process uses these descriptions to train an IoT foundation model that captures the regularities of the physical world rather than those of text. On top of this foundation, a reasoning model answers natural-language questions with physically grounded, verifiable explanations, and powers multi-agent applications that monitor, diagnose, act, and keep humans informed. I will argue that such a reasoning model should not simply be an off-the-shelf LLM, and that the physical world itself, which grades every prediction minutes later, is the ideal verifier for training it. I will also discuss how external datasets such as weather must be fused into reasoning, and how missing, faulty, delayed, and drifting data can be handled at each tier of the continuum. The talk closes by examining whether this vision qualifies as next-generation computing and by outlining a research agenda on semantics, trustworthy reasoning, autonomy, and sustainability across edge, cloud, and space.

Antonio A.F. Loureiro received the B.Sc. and M.Sc. degrees in computer science from Universidade Federal de Minas Gerais (UFMG), Brazil, and the Ph.D. degree in computer science from The University of British Columbia, Canada. He is currently a Full Professor with UFMG, where he leads the research group on ubiquitous and mobile computing. He was the recipient of the 2015 IEEE Ad Hoc and Sensor (AHSN) Technical Achievement Award and the Computer Networks and Distributed Systems Interest Group Technical Achievement Award of the Brazilian Computer Society. In the last 25 years, he has published regularly in international conferences and journals related to those areas, presented keynotes and tutorials at international conferences, and consulted with national and multinational corporations. Several of his studies were awarded best papers, and he is currently one of the most cited Brazilian researchers in Computer Science. Currently, he is a CNPq A researcher.

Miriam Capretz

Western University, Canada

For two decades, many of us believed learning from data meant bringing data together into cloud data stores, big data platforms, and machine-learning services. This keynote revisits that journey through five beliefs I once held, and what our data taught me instead: that the cloud is where learning happens, that more data means better models, that a trained model is a finished model, that keeping data local keeps it safe, and that bigger models will solve it.

Drawing on our work with real building sensor data, I will show why learning now spans the edge–fog–cloud continuum, why context, and sometimes synthetic data, matters more than volume, and how living digital twins stay calibrated by combining data-driven and physics-based models. I will also share results on how vulnerable federated learning is to adversarial attacks, and on what large and foundation models cost in both accuracy and carbon. The talk closes with open questions for the utility, cloud, and big data communities.

Miriam A. M. Capretz is a Professor of Software Engineering, Associate Dean (Research) of Western Engineering and Director of the newly established Western Autonomous and Intelligent Systems Hub at Western University, Canada. She earned her BSc in Computer Science and her master’s in electrical engineering at UNICAMP in Brazil, and her PhD from the University of Durham, UK. She spent six years at the University of Aizu, Japan, before joining Western in 2000. Her research covers big data and cloud computing, machine learning for sensor and energy data, and digital twins across edge, fog, and cloud, including federated learning and synthetic data. She is a Fellow of the Canadian Academy of Engineering and has served as Program Chair and Chief of the IEEE COMPSAC Conference (2025-2027). She has held international fellowships from RWTH Aachen (Germany), INESC-TEC (Portugal), and CNPq. She is delighted to return to Brazil, where she began her career.