IEEE eScience 2026 Workshop

AGENT4SC
1st Workshop on Agentic AI
for Large-scale Science

Advancing autonomous, trustworthy, and reproducible AI agents across the edge–cloud–HPC continuum for scientific discovery at scale.

Format Half-Day Workshop
Papers Due July 13, 2026
Submission EasyChair

Why Agentic AI for Large-scale Science?

Agentic AI is rapidly moving from "chat with tools" prototypes to autonomous systems that can reason, plan, coordinate, and act across complex digital research ecosystems. For the eScience community, this shift represents the emergence of a new control plane for computational and data-driven research.

Agents can request allocations, launch ensembles, steer workflows, move PB-scale datasets, trigger experiment/compute co-scheduling, and generate decisions that impact scientific validity — spanning the computing continuum from instruments and autonomous laboratories to simulations, data centers, cloud systems, and leadership-class HPC.

Recent closed-loop campaigns have revealed gaps in hallucination detection and mitigation, scheduling visibility, energy accounting, and reproducibility guarantees. At scale, even minor errors or blind spots can escalate into megawatt-hour waste, irreproducibility, and compromised scientific validity.

AGENT4SC provides a dedicated forum to advance scalable architectures, cross-continuum coordination, evaluation frameworks, verification and mitigation strategies, provenance, and observability mechanisms before agentic systems become embedded in production research workflows.

New Control Plane for Science

Agents reshaping how researchers generate, validate, share, and reuse data and knowledge across distributed infrastructures.

Operational Urgency

How to bound agent actions, ensure auditability, define safety controls, and support human-in-the-loop oversight for long autonomous campaigns.

Integrated Data Infrastructures

Scalable databases, vector stores, caching layers, provenance services, and knowledge graphs — performant and auditable under high concurrency.

Recurring Venue

AGENT4SC is intended to evolve into a recurring forum aligned with eScience's mission and the rapidly growing agentic AI community.

Key Topics

🏗️

System Architectures & Design Principles

Data and execution models and design principles for agentic AI beyond LLMs at scale.

🌐

Edge–Cloud–HPC Continuum

Agentic AI for cross-facility science spanning instruments, simulations, and learning across distributed infrastructure.

🧠

Agent Memory & Long-Horizon Reasoning

Planning and reasoning under extreme compute and data constraints for sustained scientific campaigns.

🗄️

Data & Metadata Architectures

Databases, vector stores, caching layers, and knowledge graphs designed for agentic systems at scale.

👤

Human-in-the-Loop & Oversight

Approval policies and escalation paths for long agentic workflow runs requiring steering and oversight.

🔍

Provenance & Observability

Auditability, reproducibility, monitoring, tracking, debugging, and observability of agentic systems at scale.

🛡️

Safety, Security & Accountability

Reliability and accountability mechanisms for agentic workflows operating in production scientific environments.

⚠️

Hallucination Detection & Recovery

Verification, failure detection, mitigation, and recovery techniques for large-scale agent-driven systems.

📊

Performance Modeling

Performance analysis and modeling for agentic systems under realistic scientific workloads.

🔗

Interoperability & Standardization

Interfaces, schemas, protocols, frameworks, and execution models for cross-platform agent coordination.

📅

Cross-Platform Scheduling

Allocation and resource awareness under agent-driven control across heterogeneous systems.

🚀

Real-World Experience Reports

Lessons from large-scale scientific deployments of agentic systems in production environments.

Call for Papers

AGENT4SC invites original research papers, position papers, and experience reports on the systems foundations required to operationalize agentic AI in large-scale scientific environments. We encourage submissions from academia, national laboratories, industry, and operational HPC centers.

Paper Formats

4
Short Paper
Work-in-progress
not including references
8
Full Paper
Complete research
not including references

Topics of Interest

We welcome contributions on any of the key topics listed above, including but not limited to:

  • Novel architectures for autonomous agents operating across distributed scientific infrastructure
  • Systems integration of agents with HPC schedulers, data services, and experiment instruments
  • Provenance, observability, and auditability for agent-driven scientific campaigns
  • Safety, hallucination mitigation, and failure recovery mechanisms at scale
  • Human-in-the-loop and oversight frameworks for long autonomous workflows
  • Experience reports from real deployments, including lessons learned and operational gaps

Submission Guidelines

  • Papers must be written in English and formatted according to IEEE eScience 2026 author guidelines
  • All papers will receive at least three peer reviews from the Program Committee
  • Accepted papers will be published in the official eScience 2026 workshop proceedings
  • Authors are encouraged to include reproducibility information about relevant software, data artifacts, and AI assistants used

Submission System

Papers must be submitted through EasyChair. The submission link will be available closer to the submission deadline.

Important Dates

Papers Due July 13, 2026
Notification of Acceptance August 3, 2026
Camera-Ready Papers August 10, 2026
Workshop IEEE eScience 2026
Submit via EasyChair →

Tentative Schedule

Time Activity
9:00 – 9:15 Opening & Welcome Remarks Opening
9:15 – 9:30 Paper Presentation 1 Paper
9:30 – 9:45 Paper Presentation 2 Paper
9:45 – 10:00 Paper Presentation 3 Paper
10:00 – 10:15 Paper Presentation 4 Paper
10:15 – 10:30 Paper Presentation 5 Paper
10:30 – 11:00 Coffee Break Break
11:00 – 12:30 Interactive Panel Discussion Panel
Structured discussion from concrete scenarios (allocation requests, data movement spikes, unsafe tool invocation, hallucination risks) to actionable patterns (approval policies, escalation paths, safe shutdown, observability, audit trails). Audience participation encouraged.

Panelists

* Panelist list is tentative and subject to confirmation.

Organizers & Committee

AG
Co-Organizer

Amal Gueroudji

Argonne National Laboratory, USA

Assistant Computer Scientist at ANL (MCS), focusing on data management for HPC+AI workflows and agentic AI systems, with emphasis on vector databases, provenance-aware architectures, and trustworthy, scalable data infrastructures across heterogeneous environments. Ph.D. from Université Grenoble Alpes.

RS
Co-Organizer

Renan Souza

Oak Ridge National Laboratory, USA

Tech lead, senior software engineer and research scientist of intelligent data and AI platforms to accelerate scientific discovery. 15+ years of experience at IBM Research, ORNL, SLAC, and UFRJ. Focus on scalable, low-latency, observable, provenance- and metadata-first architectures. Author of 50+ papers and holder of 10+ USPTO patents.

Steering Committee

Rosa Filgueira

University of St Andrews, UK

Rafael Ferreira da Silva

Oak Ridge National Laboratory, USA

Kyle Chard

Argonne National Laboratory, USA

Program Committee

Alok Kamatar — University of Chicago, USA
Paula Olaya — NVIDIA, USA
Amal Gueroudji — Argonne National Laboratory, USA
Raül Sirvent — Barcelona Supercomputing Center, Spain
Andrew E. Shao — Hewlett Packard Enterprise, USA
Renan Souza — Oak Ridge National Laboratory, USA
Daniel Balouek — Inria, France
Rosa Filgueira — Edinburgh Parallel Computing Centre, UK
Daniel de Oliveira — Fluminense Federal University, Brazil
Sandro Fiore — University of Trento, Italy
Daniel Garijo — Universidad Politécnica de Madrid, Spain
Seth Ockerman — University of Wisconsin, USA
Daniel Rosendo — Oak Ridge National Laboratory, USA
Silvina Caíno-Lores — Inria, France
Frédéric Suter — Oak Ridge National Laboratory, USA
Soumyendu Sarkar — Hewlett Packard Enterprise, USA
Haytham Fayek — RMIT University, Australia
Tainã Coleman — San Diego Supercomputing Center, USA
Jack Marquez — University of Tennessee, USA
Wes Brewer — Oak Ridge National Laboratory, USA
Logan Ward — NVIDIA, USA
William Godoy — Oak Ridge National Laboratory, USA
Loïc Pottier — Lawrence Livermore National Laboratory, USA
Woong Shin — Oak Ridge National Laboratory, USA
Matthieu Dorier — Argonne National Laboratory, USA
Yogesh Simmhan — Indian Institute of Science, India

Workshop Venue

Co-located with IEEE eScience 2026

AGENT4SC is a workshop at the IEEE International Conference on e-Science, the premier forum for data-intensive and compute-intensive research across the full scientific lifecycle.

Duration Half Day
Expected Attendance 20 – 40
Submission System EasyChair