Keynotes ● Cluster 2026
Keynote 1: Wednesday, September 23rd
Abstract
We are in an AI boom with no bust in sight. Many of the challenges we now face in building efficient, large-scale AI systems are surprisingly familiar to the cluster and HPC communities. For more than two decades, I've been exploring how to measure where energy goes in large-scale computing systems, relate energy consumption to application behavior, and intelligently manage heterogeneous computing resources to maximize performance while minimizing energy. The foundation of this work, PowerPack, created an accurate, solution-agnostic measurement toolkit that helped enable energy-efficiency measurement for the Green500 and allowed us to develop and directly verify innovations in high-performance, power-aware computing, from algorithms to architectures and system designs. In this talk, we revisit those ideas in the age of AI. What still works? And what new measurement, modeling, benchmarking, and resource-management techniques will be needed to understand—and ultimately improve—the energy efficiency of emerging AI systems and applications?
Biography
Kirk W. Cameron is the Boeing Distinguished Professor of Computer Science and Managing Director of the Institute for Advanced Computing at Virginia Tech. His research focuses on high-performance computing, computer systems, and power and energy efficiency. For more than two decades, his research group has developed technologies for measuring, understanding, and improving the energy efficiency of large-scale computing systems, including the PowerPack Toolkit and foundational concepts for power-aware computing. Cameron is a co-founder of the Green500 and has contributed to widely adopted power and energy-efficiency standards, including SPECpower and SERT. He is an IEEE Fellow, an ACM Distinguished Scientist, and a recipient of the NSF CAREER Award. His current research applies this experience to measuring, modeling, and improving the energy efficiency of emerging AI systems and applications.
Keynote 2: Thursday, September 24th
Abstract
Scientific computing is increasingly a connected lifecycle in which experiments, data, high-performance computing, and AI work together to produce scientific results. Yet the final result is only the last page of the story: nondeterministic executions can diverge, AI models can reach similar answers through different learning trajectories, workflows can hide the origins of their outputs, and data, instruments, computing, and scientists increasingly operate across distributed facilities.
In this keynote, I will argue that trust in scientific computing requires a persistent “paper trail” that captures how a result came to be. Through four scenarios, I will show how graph-based analysis can expose and localize nondeterminism in HPC executions; how training histories and compact metadata can reveal how neural networks learn and enable earlier, more efficient decisions; how fine-grained provenance can connect scientific results to their data, software, models, parameters, and execution context; and how federated data ecosystems can extend this traceability across instruments, computing facilities, AI resources, and scientific communities. These examples point toward a future in which scientific results do not travel alone: they carry the evidence needed to reproduce, explain, evaluate, and reliably reuse them.
Biography
Michela Taufer, an AAAS Fellow and ACM Distinguished Scientist, is the MathWorks Professor at the University of Tennessee, Knoxville. She earned her Laurea in Computer Engineering from the University of Padova, Italy, and her doctoral degree in Computer Science from the Swiss Federal Institute of Technology in Zurich. Her postdoctoral work at the University of California, San Diego, and The Scripps Research Institute bridged computer systems and computational chemistry, underscoring her lifelong commitment to interdisciplinary research.
Throughout her career, Taufer has passionately combined computational and experimental sciences. Her cyberinfrastructure solutions leverage high-performance, cloud, and volunteer computing to advance scientific applications. An advocate for scientific rigor, she promotes reproducibility, replicability, and transparency across her projects. Taufer has held leadership roles in the HPC community, including serving as editor-in-chief of Future Generation Computer Systems and as a member of the Computing Community Consortium (CCC).