Modern precision experiments are complex adaptive systems: thousands of control loops, millions of sensor channels, and operating points that drift with temperature, alignment, and time. We build AI tools that act as collaborators—learning controllers, diagnostic systems, and design optimizers that make experiments more robust and more capable.
**Reinforcement learning** is a central theme. We train agents to acquire and maintain lock on optical cavities, adapting in real time to disturbances that would defeat hand-tuned controllers. These learned policies often discover strategies human operators never considered, and they scale to systems too complex for analytical design.
Beyond control, we use machine learning for **predictive diagnostics** (catching failures before they happen), **anomaly detection** (flagging data segments corrupted by transient noise), and **inverse design** (searching high-dimensional parameter spaces for optimal experiment configurations). A growing thrust is using AI to accelerate the design of quantum-enhanced measurements—finding non-Gaussian states and measurement protocols tailored to specific sensing tasks.
We emphasize **deployability**. Our tools run in real time on detector hardware, not just in offline analysis. Reliability matters: an AI controller that crashes during an observing run is worse than no AI at all. We invest heavily in validation, uncertainty quantification, and graceful degradation.
This pillar is for those who want to bring modern machine learning to bear on real physical systems, with immediate feedback on whether the algorithms actually work.
Representative topics
Reinforcement-learning controllers for interferometers
Bayesian / generative optical design tools
Digital-twin-based diagnostics and forecasting
Searching for unmodeled signals
Projects in this pillar
RL for classical feedback control in LIGO
— Apply reinforcement learning to design and tune classical feedback controllers that keep LIGO interferometers stably locked.
For an overview of all pillars and projects, see the
Research Projects page.
