Video Science and Discovery
AI methods (supervised, unsupervised, reinforcement-based) are seeing a progressive integration in traditional sciences,unlocking unprecedented capabilities in terms of accurate measurements, efficient simulation, and effective control of non-linear physical systems. Additionally, there is clear evidence that machine learning models can significantly outperform our best models and theories in the analysis of physical systems, e.g., in connection with non-linear systems and turbulence. This opens new possibilities for fundamental understanding.
What you will learn
This track of the master program AI&ES focuses on the interplay between fundamental physical sciences and applications of AI. Conversely, physical sciences, advanced materials and electronics can be used in dedicated hardware to optimize AI systems. Emergent methodologies aiming at rendering machine learning models more effective thanks to the integration of known physical properties (symmetries) will also be considered. Examples of science applications include the following:
- Surrogate modelling of physics models for accelerated multi-physics simulation
- Data-driven surrogate model development
- Measurement and reinforcement-based control of flowing systems
- AI-based physics simulation acceleration
- Development of real-timediagnostics through ML-accelerated tomographic inversion + analysis chains
- Reinforcement Learning for tokamak trajectory optimization based on simulators
- Explore the link between deep learning and the glass transition.
- Integration of structural knowledge(symmetries) into machine learning models
- Data-driven modeling and prediction of material deformationdue to applied forces
- Hardware-based (neuromorphic) systems for efficient A.I.
- Heat and flow inlow pressure systems: physics of interfaces
- Material discovery for heat storage applications
To integrate basic knowledge, courses on modeling and simulation of soft and flowing matter (fluids, plasmas) will also be offered.