|
Lecturer(s)
|
-
Šmídl Václav, prof. Ing. Ph.D.
|
|
Course content
|
1. Definition of standard tasks in time series (prediction, fault detection, smoothing). 2. Classical approaches (fft, filtration). 3. Modern methods of time series prediction - recurrent neural networks. 4. Structured state models (SSM) transformers and their connections. 5. Processing data from multiple modalities (sensor fusion). 6. Applications in fault and anomaly detection. 7. Graph theory, graph neural networks. 8. Use of graph neural networks in electrical circuits. 9. Use of neural networks for solving and identifying differential equations (PINN). 10. Estimation of model prediction uncertainty in general (MC dropout, ensembles) and for individual special cases. 11. Use in active learning methods and device design. 12. Methods of model explainability (attention, attribution, abductive explanation).
|
|
Learning activities and teaching methods
|
One-to-One tutorial, Group discussion, Individual study
- Individual project (40)
- 40 hours per semester
- Graduate study programme term essay (40-50)
- 20 hours per semester
- Contact hours
- 52 hours per semester
|
| prerequisite |
|---|
| Knowledge |
|---|
| to have basic knowledge of programming concepts in Python |
| to have basic knowledge of machine learning concepts (cross-validation, optimization, plausibility) |
| Skills |
|---|
| to operate commonly available computer technology |
| to use basic algorithms, data and control structures |
| Competences |
|---|
| N/A |
| N/A |
| N/A |
| learning outcomes |
|---|
| Knowledge |
|---|
| to explain modern machine learning algorithms and apply them to practical problems |
| to have knowledge of various types of data including time series, graphs, and differential equations |
| Skills |
|---|
| to apply modern machine learning algorithms to practical problems |
| to work with different types of data including time series, graphs, and differential equations |
| to apply methods to data from real-world problems |
| Competences |
|---|
| N/A |
| N/A |
| N/A |
| teaching methods |
|---|
| Knowledge |
|---|
| Lecture |
| Lecture supplemented with a discussion |
| Multimedia supported teaching |
| Skills |
|---|
| Practicum |
| Project-based instruction |
| Students' portfolio |
| Competences |
|---|
| Lecture supplemented with a discussion |
| Students' portfolio |
| assessment methods |
|---|
| Knowledge |
|---|
| Project |
| Continuous assessment |
| Skills |
|---|
| Individual presentation at a seminar |
| Skills demonstration during practicum |
| Competences |
|---|
| Individual presentation at a seminar |
| Continuous assessment |
|
Recommended literature
|
-
Albert Gu, Karan Goel, Christopher Ré. Efficiently Modeling Long Sequences with Structured State Spaces. .
-
Goodfellow, I., Bengio, Y., Courville, A., & Bengio, Y. Deep learning. Cambridge. 2016.
-
Raissi, Maziar, Paris Perdikaris, and George E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. .
-
Zhang, Aston; Lipton, Zachary C.,; Li, Mu; Smola, Alexander J. Dive into deep learning. First published. 2024. ISBN 978-1-009-38943-3.
|