Visiting Scientist at CERN · Leading SARL at the University of Salzburg

Simon Hirländer

Vice Director, IDA Lab · Head of Smart Analytics & Reinforcement Learning
Department of Artificial Intelligence and Human Interfaces · University of Salzburg

Theoretical Physicist and Decision Scientist bridging rigorous mathematical physics (Hamiltonian mechanics, symplectic geometry, potential theory) with modern Deep Reinforcement Learning. In 2018/19 I designed and carried out the first deep reinforcement learning experiment at CERN, on the Low Energy Ion Ring. In 2020 I brought reinforcement learning to the University of Salzburg: what started as a single seminar topic is now a regular part of the curriculum, with a Reinforcement Learning lecture and exercise in the BSc Artificial Intelligence and advanced RL courses in the MSc Data Science. Today, my research team develops safe, uncertainty-aware, and physics-informed AI systems for critical physical infrastructures and industrial processes.

Simon Hirländer
Role
Vice Director, IDA Lab · Head of Smart Analytics and Reinforcement Learning
Funding
€1.89M across 20 funded projects, principal investigator on 19
Teaching
Introduced RL at Salzburg (2020) · now in the BSc AI curriculum (VO/UE 536.111/112) · MSc Data Science · courses
Supervision
10 doctoral, 7 master and 11 bachelor theses
Collaboration
CERN · GSI/FAIR · DLR · COPA-DATA · Salzburg Research
Recent
Invited talk, NeurIPS 2026 workshop, Paris · four IPAC’26 papers · EPJ Research Infrastructures, Critical Care

Core Research Pillars

Integrating physical invariants and control guarantees into modern decision-making architectures.

Reinforcement Learning Theory
Sample-efficient offline-to-online RL, Model-Based Policy Optimization (MBPO/MOPO), and Hamiltonian neural networks preserving symplectic phase-space invariants.
Explore Theory →
Autonomous Accelerators
Data-driven control and Bayesian optimization for the world's most complex machines: CERN LHC/SPS, AWAKE, DESY, KIT, and GSI/FAIR SIS18.
Explore Accelerators →
Industrial Systems & Energy
Physics-informed world models and predictive control for heavy industry, including roughing mill optimization (Danieli) and energy management (COPA-DATA FOCUS).
Explore Industrial AI →
Safe Control & GP-MPC
Gaussian Process Model Predictive Control (Causal GP-MPC), constrained exploration, and uncertainty quantification for safety-critical operations.
Explore Safe Control →

Recent News & Highlights

December 12, 2026

Invited Talk at NeurIPS 2026 Workshop (Paris)

Invited talk on "Koopman Stabilised Symplectic World Models for Safe Offline-to-Online RL in Particle Accelerators" at the NeurIPS 2026 Workshop RL for Experimental Sciences: Bridging the Simulation-to-Reality Gap in Paris.

Invited Talk NeurIPS 2026
2026

Podcast: S2 #17 | Reinforcement Learning in Business

Featured episode on DIH West's EASY cheesy DIGITAL exploring decision-making under uncertainty, agentic AI, and business transformation through Reinforcement Learning.

Spotify Podcast
June 2026

Four IPAC'26 Papers Accepted

Contributions on Koopman-stabilised world models, Causal GP-MPC, delayed-consequence RL beyond greedy optimization, and SIS18 injection optimization accepted for presentation.

IPAC'26 Peer Reviewed
2026

KI-MINTFIT Doctoral Program Launches (€1.08M)

Launch of the cooperative doctoral program funded by the Austrian BMBWF in collaboration with PH Salzburg, advancing AI-driven STEM subject instruction and student models.

€1.08M BMBWF Doctoral Program
View All News & Archive →

Real-World Impact & Results

Demonstrated breakthroughs across extreme, high-dimensional experimental environments.

Particle Accelerators
SIS18 Injection Losses: 45% → 12%

Automated tuning at the GSI/FAIR SIS18 heavy-ion synchrotron (Geoff framework, EPJ Research Infrastructures 2026); my team contributed the data-driven Gaussian Process MPC.

Pioneering Experiment
First Deep RL Experiment at CERN

A Deep Q-Network agent learned, directly on the machine, to steer the beam at injection into the Low Energy Ion Ring (LEIR). Presented at the 2nd ICFA ML Workshop, PSI, Feb 2019 (slides); followed by the AWAKE and LINAC4 experiments (PRAB 23, 124801, 2020).

Industrial Automation
Heavy Steel Roughing Mill (R² > 0.93)

Trained Stable Koopman State-Space Models on 58M+ operational industrial time-series points with Danieli Automation, achieving high-accuracy torque and tension prediction.

Institutional Partners & Collaborators

University of Salzburg CERN Accelerators GSI Helmholtzzentrum / FAIR COPA-DATA Danieli Automation B&R Industrial Automation (ABB) DIH West PH Salzburg Montanuniversität Leoben

Interested in Research Collaboration or Theses?

We regularly supervise motivated MSc and PhD students and collaborate with industrial partners and research labs on reinforcement learning and autonomous systems.

Get in Touch Meet the SARL Team