Advanced Deep Learning for Physics

What it is about

In this module you will learn deep-learning methods and numerical simulation algorithms for physical materials such as fluids and deformable bodies. By the end you will be able to apply concepts such as generative models, time-series prediction, and numerical methods for partial differential equations, and select suitable network architectures and solvers for concrete tasks.

What you will be able to do

• Understanding fundamental deep-learning concepts (autoencoders, adversarial training, RNNs)
• Knowledge of specialized loss functions and choice of appropriate activations
• Understanding physical principles of elastic and plastic materials with emphasis on fluids
• Knowledge of conserved quantities: mass, momentum, divergence-free, vorticity
• Familiarity with discrete and continuous representations (phase fields, level sets, Cartesian/ tetrahedral meshes)
• Mastery of numerical algorithms for PDEs: finite differences, explicit/implicit integration, pressure projection
• Ability to compute derivatives of loss functions and construct training algorithms
• Evaluation of learning and simulation algorithms with respect to accuracy and complexity
• Practical implementation of central solver components in a high-level programming language

What the module consists of

• LectureConveying the theoretical foundations on deep learning, fluid and elasticity physics, and numerical methods
• Exercises/AssignmentsVoluntary group exercises to implement central algorithms; serve practical deepening and exam bonus
• Demos/ExperimentsIllustrative material through applications, videos and simulations to illustrate physical phenomena

Teaching method

• Powerpoint/SlidesPresentation of the lecture contents
• Board/BlackboardExplanation of mathematical derivations
• Demos and VideosVisualization of real phenomena and simulation results
• Experiments / Physics-Fact ChallengesActive engagement of students during the lecture
• Group work during exercisesPractical implementation and collaboration