Course: Physics-Informed Machine Learning and Artificial Intelligence

» List of faculties » FBI » UAI
Course title Physics-Informed Machine Learning and Artificial Intelligence
Course code UAI/352
Organizational form of instruction Lecture
Level of course Doctoral
Year of study not specified
Frequency of the course In every academic year, in the summer semester
Semester Summer
Number of ECTS credits 4
Language of instruction Czech, English
Status of course Compulsory-optional
Form of instruction Face-to-face
Work placements This is not an internship
Recommended optional programme components None
Lecturer(s)
  • Bukovský Ivo, doc. Ing. Ph.D.
Course content
COMMON TOPICS T1 ? Introduction to Data-Driven Modelling in Physics Data-driven, model-driven, and hybrid approaches; role of machine learning; explainability, convergence, and stability of neural architectures; black-box vs. interpretable models; in-parameter-linear architectures. T2 ? LNU, HONU, and In-Parameter-Linear Neural Models LNU and HONU models; polynomial input expansion; incremental learning and backpropagation; weight interpretation; approximation of dynamical systems and state-space representation. T3 ? State-Space Models and Their Neural Representations Discrete-time state-space models; modelling systems with known states; derivation from HONU models; latent states; transition to attention-based architectures. T4 ? MLP, GRU, and LSTM Architectures MLP, GRU, and LSTM networks; Backpropagation Through Time; modelling long-term dependencies; interpretability; comparison with shallow neural models. T5 ? Optimization Algorithms for Neural Learning Gradient-based methods, conjugate gradient, and Levenberg?Marquardt optimization; convergence; loss minimization; regularization and hyperparameter tuning. T6 ? Dimensionality Reduction and Latent Space Projections PCA, t-SNE, and UMAP; visualization and compression of multivariate data; feature preparation for modelling and clustering. T7 ? Clustering and Self-Organizing Representations K-means, DBSCAN, and SOM; unsupervised state recognition; topological data representation; applications to physical signals. T8 ? Model Validation, Interpretability, and Generalization Model validation; sensitivity and collinearity analysis; Learning Entropy and AISLE metrics; robustness, novelty detection, interpretability; semester project. FOCUS TOPICS (At least four topics will be selected according to the students' specialization.) F1 ? Applied Fractal-Based Features in Prediction of Complex Dynamical Systems Fractal dimension, recurrence analysis, and entropy-based features; modelling turbulence, chaotic systems, and physiological signals. F2 ? Learning Entropy and Weight Dynamics Monitoring Learning Entropy and AISLE; monitoring weight updates; change detection and novelty assessment. F3 ? Deriving State-Space Representations from HONU Models State-space models derived from HONU; systems with known states; interpretability, convergence, and stability. F4 ? Attention Mechanisms in Systems with Unknown States Attention, self-attention, and multi-head attention; modelling latent states; anomaly detection and behavior segmentation. F5 ? Latent Space Analysis of Neural Networks Latent data representations; geometry and interpretation of latent spaces; clustering, visualization, and system identification. F6 ? Physics-Informed Neural Networks (PINNs) PINNs; PDE-constrained loss functions; automatic differentiation; collocation points; DeepONets and Fourier Neural Operators. F7 ? Autoencoders and Variational Models for Latent Feature Extraction Autoencoders and VAEs; data compression, denoising, and latent feature extraction; system identification. F8 ? Predictive Control with Neural Models Model predictive control; HONU-based predictors; optimization, feedback control, and stability analysis. F9 ? Modern State-Space Neural Architectures: Mamba and Beyond Mamba and state-space sequence models; linear-time memory; comparison with RNNs and Transformers; modelling physical processes. F10 ? Spiking Neural Networks and Event-Based Preprocessing Spiking neural networks; temporal spike coding; STDP and surrogate gradients; event-based data processing; neuromorphic hardware.

Learning activities and teaching methods
unspecified
Learning outcomes
The course focuses on data-driven modelling methods of physical systems using neural networks and advanced machine learning algorithms. The lectures consist of 8 common topics (T1-T8) and at least 4 specialized topics (F1?F10) according to the students? area of focus.

Prerequisites
linear algebra, ordinary differential equations, probability fundamentals, basic use of Python

Assessment methods and criteria
unspecified
Credit is given for active participation in classes and consultations and for semestral project , combined exam
Recommended literature
  • G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, ?Physics-informed machine learning,? Nat Rev Phys, vol. 3, no. 6, pp. 422?440, May 2021, doi: 10.1038/s42254-021-00314-5..
  • I. Bukovsky, G. Dohnal, et al, ?Letter on Convergence of In-Parameter-Linear Nonlinear Neural Architectures with Gradient Learnings,? IEEE Trans. Neural Networks and Learning Systems, vol. 34, no. 8, pp. 5189?5192, Aug. 2023, doi: 10.1109/TNNLS.2021.3123533..
  • I. Bukovsky, Interaktivní výukové podklady v Jupyter Notebook, Dpt. of Computer Science, Faculty of Science, Univ. of South Bohemia in Ceske Budejovice (TEAMs, MOODLE).
  • M. Raissi, P. Perdikaris, and G. E. Karniadakis, ?Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,? Journal of Computational Physics, vol. 378, pp. 686?707, Feb. 2019, doi: 10.1016/j.jcp.2018.10.045..
  • Physics-Informed Machine Learning,? in Fundamentals of Pattern Recognition and Machine Learning, Cham: Springer International Publishing, 2024, pp. 293?324. doi: 10.1007/978-3-031-60950-3_12..
  • S. L. Brunton and J. N. Kutz, Data-Driven Science and Engineering: Machine Learning, Dynamical Systems, and Control, 2nd ed. Cambridge: Cambridge University Press, 2022. https://doi.org/10.1017/9781009089517 (volně dostupné, free book: https://databookuw.com/databookV2.pdf).


Study plans that include the course
Faculty Study plan (Version) Category of Branch/Specialization Recommended year of study Recommended semester