AI/ML techniques are being incorporated into all aspects of the scientific and engineering process. One early effort area has been to replace previous autonomic system components with AI models that can offer more nuanced and continuously updated automation behavior. Work has been done to try to predict IO behavior to enable more efficient machine throughput, log monitoring to detect patterns that may reveal either security concerns or faulty components that fail in consistent, but unusual ways, and to manage applications and caches to better address the system at a whole than by individual components. All of these and many more system-related tasks address a complex, sometimes intractable problem, and seeks to use AI tools to offer better solutions than either heuristics or scope limited solutions that have existed previously.

This workshop solicits novel work that explores how to effectively incorporate AI into system management and monitoring, particularly for complex systems that support scientific and engineering workloads (i.e., cloud and HPC).

Areas of interest and domains of work include, but are not limited to:

Submission

Submitted papers need to be formatted in the ACM conference format, with a page limit of no more than 5 pages long including everything except references. Accepted papers will be published in the ACM proceedings. Submissions will be peer-reviewed in a single blind way; author names and affiliations need to appear in the paper submission, but reviewer names will remain anonymous.

Submission Link: https://ai4sys25.hotcrp.com/

• Paper submission deadline: April 11 April 18, 2025 (AoE)
• Author notification: May 6, 2025
• Camera ready submissions: May 16, 2025
• Workshop: July 20, 2025

• 09.00 – 09.05 | Opening Remarks


• 09.05 – 09.50 | Keynote Talk by Anthony Kougkas, Associate Research Professor, Illinois Institute of Technology (IIT)
  Data Management Challenges in the Age of Agentic AI.


• 09.50 – 10.10 | Accepted Paper: XPF: Agentic AI System for Business Workflow Automation.
  Kunal Rao, Giuseppe Coviello, Gennaro Mellone, Ciro Giuseppe De Vita and Srimat Chakradhar
  NEC Laboratories America, Inc.
• 10.10 – 10.30 | Accepted Paper: BanditWare: A Contextual Bandit-based Framework for Hardware Prediction.
  Tainã Coleman, Hena Ahmed, Ravi Shende, Ismael Perez, İlkay Altintaş
  University of California San Diego, USA


• 10.30 – 11.00 | Coffee Break


• 11.00 – 11.45 | Keynote Talk by Jian Huang, Associate Professor, University of Illinois at Urbana-Champaign (UIUC)
  Building an AI-Driven Storage Ecosystem and Beyond


• 11.45 – 12.05 | Accepted Paper: Can Large Language Models Predict Parallel Code Performance?
  Gregory Bolet (Virginia Tech), Giorgis Georgakoudis (Lawrence Livermore National Laboratory), Harshitha Menon (Lawrence Livermore National Laboratory), Konstantinos Parasyris (Lawrence Livermore National Lab), Niranjan Hasabnis (Code Metal), Hayden Estes (Virginia Tech), Kirk Cameron (Virginia Tech), Gal Oren (Technion, Stanford University)
• 12.05 – 12.30 | Panel discussion with all presenters.

Keynote Talk by Anthony Kougkas, Associate Research Professor, Illinois Institute of Technology (IIT). [Website]
Title: Data Management Challenges in the Age of Agentic AI
Abstract: The era of data-driven science has arrived at a critical junction where simulation, analytics, and AI converge to create increasingly complex scientific workflows. As AI transforms research practices, it intensifies the already "data-starved" condition in scientific computing, with traditional storage systems buckling under unprecedented demands. If data is indeed the new “oil” of our digital economy, then data management represents the next-generation refinement and distribution infrastructure. But unlike petroleum, data offers a remarkable advantage—its value is inherently renewable. While oil evaporates with use, data's value can be continuously harvested, refined, and enhanced through proper management. Just as the oil industry requires sophisticated logistics—from extraction and refinement to distribution—scientific data demands a similar comprehensive approach. In this talk, we will examine how modern scientific workflows require comprehensive data management approaches that go beyond mere storage solutions. We will explore architectural frameworks that orchestrate the entire data lifecycle—from ingestion and optimization to movement across storage tiers and advanced analytics. We will demonstrate how integrating agentic AI capabilities into system management can create adaptive, intelligent environments where data-driven decision making becomes streamlined across complex multi-tiered infrastructures. The discussion will highlight emerging techniques for transforming diverse scientific data formats into unified representations that facilitate discovery, while addressing the practical challenges of hardware acceleration and framework integration. By examining these approaches in the context of current scientific computing bottlenecks, we aim to illuminate pathways toward systems that not only address today's performance demands but establish foundations for sustainable, AI-driven scientific discovery environments.
Speaker Bio: Dr. Anthony Kougkas is an Associate Research Professor at the Illinois Institute of Technology and Deputy Director of the Gnosis Research Center. He also serves as a Guest Research Faculty member at Argonne National Laboratory. His research brings together high-performance computing and AI-driven data management, emphasizing multi-tiered storage and I/O frameworks that accelerate discovery in data-intensive environments. By collaborating with national laboratories and other research teams, his work aims to create more agile and scalable solutions for modern scientific computing.

Keynote Talk by Jian Huang, Associate Professor, University of Illinois at Urbana-Champaign (UIUC).[Website]
Title: Building an AI-Driven Storage Ecosystem and Beyond
Abstract: Storage systems today have been built into a complicated ecosystem, which involves the development and deployment of storage devices, storage software, and application-level data stores. To rapidly meet the ever-increasing performance and efficiency requirements, the entire hardware and software stack need to adapt instantly in a coordinated fashion. However, it is challenging to achieve this with current human-driven systems-building approaches. In this talk, I will discuss how the storage ecosystem can be advanced to embrace AI techniques to facilitate its development, deployment, and optimizations across the full stack. I will also discuss how our experience of building an AI-driven storage ecosystem would shed light on the development of “machine learning for systems” in general.
Speaker Bio: Dr. Jian Huang is an Associate Professor and Y. T. Lo Faculty Fellow in the ECE department at the University of Illinois at Urbana-Champaign. He received his Ph.D. in Computer Science at Georgia Tech in 2017. His research interests include computer systems and architecture, sustainable AI infrastructure, memory/storage systems, data systems, systems security, and especially the intersections of them. His research contributions have been published at top-tier computer architecture and systems conferences such as ISCA, MICRO, ASPLOS, OSDI, and SOSP. His work received USENIX Best Paper Award, MICRO Best Paper Runner Up, and multiple IEEE Micro Top Picks (and Honorable Mentions). He also received the inaugural SIGMICRO Early Career Award, NSF CAREER Award, NSF CRII Award, Dean’s Award for Early Innovation, Illinois Proof-of-Concept Award, NetApp Faculty Fellowship Award, and Google Faculty Research Award. He is a co-founder of the Workshop on Hot Topics in System Infrastructure (HotInfra).

Organizing Committee

• Jay Lofstead (Sandia National Laboratories, USA)
• Jai Dayal (Cerebras Systems)
• Thaleia Dimitra Doudali (IMDEA Software Institute, Spain)

Program Committee

• Pamela Delgado (HES-SO, Switzerland)
• Gal Oren (Technion, Israel)
• Kevin Menear (National Renewable Energy Laboratory, USA)
• Jai Dayal (Cerebras Systems)
• Thaleia Dimitra Doudali (IMDEA Software Institute, Spain)

Past Editions