The 7th Workshop on Artificial Intelligence for Robust Engineering & Science (AIRES 7), to be held on August 12-13, 2026 at Oak Ridge National Laboratory, Oak Ridge, Tennessee.

AIRES 7 will bring together students, researchers, engineers and thought leaders from the academia, industry and the laboratory complex to explore the frontiers of AI-driven scientific discovery and engineering innovation. This year’s workshop will be organized around three compelling themes:

1. Foundation Models for Science
Embedding physical laws as constraints — encoding conservation laws, symmetries, and governing equations directly into model architectures so outputs are physically consistent by construction

Surrogate and emulator models — training fast approximators on high-fidelity simulation data to replace expensive numerical solvers at inference time, enabling real-time feedback loops

Multi-scale and multi-physics coupling — bridging molecular, meso-, and continuum scales within unified models, avoiding the loss of information that occurs when scales are treated independently

Scientific foundation models and transfer learning — pre-training large models on broad scientific corpora (simulation outputs, experimental datasets) then fine-tuning for specific domains such as plasma dynamics, hydrology, or nuclear materials

Uncertainty-aware inference — propagating uncertainty through physics-constrained models so that predictions carry calibrated confidence bounds suitable for decision-making in engineering contexts

2. Agentic AI & Autonomous Scientific Workflows
Closed-loop experimental design — AI agents that propose, execute, observe, and iterate experiments autonomously, applying active learning to prioritize the most information-rich next steps

Robotic and automated laboratory integration — coupling AI planning layers with physical automation (liquid handling, synthesis platforms, characterization instruments) to run high-throughput experiments without human intervention

Hypothesis generation and scientific reasoning — going beyond parameter optimization to have agents formulate testable scientific hypotheses, reason over causal chains, and suggest mechanistic explanations for observed phenomena

Multi-agent orchestration across facilities — coordinating networks of specialized agents operating across geographically distributed labs, HPC clusters, and data repositories toward shared discovery objectives

Navigating combinatorially large design spaces — applying techniques such as Bayesian optimization, evolutionary search, and reinforcement learning to efficiently explore vast parameter spaces in materials, chemistry, and process engineering where exhaustive search is infeasible

3. Digital Twins & Real-Time Cyber-Physical Systems
Whole-facility and system-of-systems twins — constructing integrated virtual replicas that couple subsystems (thermal, mechanical, electrical, chemical) into a single coherent model rather than isolated component simulations

Real-time data assimilation and state estimation — continuously ingesting live sensor streams to update model state, distinguishing signal from noise and correcting model drift as operating conditions evolve

Predictive maintenance and anomaly detection — using twin-predicted baselines to identify early signatures of degradation, failure, or off-nominal behavior before physical consequences occur

Human-in-the-loop decision support — designing twin interfaces that surface actionable, interpretable recommendations to operators while preserving human authority over critical decisions

Simulation-to-reality fidelity and validation — developing rigorous protocols to quantify and reduce the gap between twin predictions and real-world behavior, particularly under novel or extreme operating conditions.

Full program details, registration information and logistics will be shared in the coming months. In the meantime, please mark your calendars and stay tuned for future announcements.