Course: Surrogate Model Learning

» List of faculties » FEL » KEV
Course title Surrogate Model Learning
Course code KEV/ATNM
Organizational form of instruction Lecture + Tutorial
Level of course Master
Year of study not specified
Semester Winter
Number of ECTS credits 4
Language of instruction Czech, English
Status of course unspecified
Form of instruction Face-to-face
Work placements This is not an internship
Recommended optional programme components None
Lecturer(s)
  • Šmídl Václav, Prof. Ing. Ph.D.
  • Ševčík Jakub, Ing.
Course content
1 Introduction and aims of the course. Julia language primer: syntax, control and data structures, primitive types. 2 Type system of Julia, multiple-dispatch, macros and generated code 3 Code organization in the Julia language, use and creation of the packages, visualization 4 Elements of machine learning demonstrated on lineární regression: optimization, train/test data, model selection and comparison 5 Optimization techniques, gradient descent and its variants, constraint optimization, local minima. Package Optim. 6 Automatic differentiation, forward and backward regime. Source-to-source differentiation. Packages ForwardDiff, ReverseDiff, Zygote 7 Neural networks as a universal approximator. Basic architectures and examples of their use. Package Flux. 8 Learning of dynamical systems. Linear systems, nonlinear systems, nelineární (Sindy method) and recurrent neural networks. 9 Neural differential equations, ordinary and partial. Physics-informed neural networks. Packages DataDrivenDiffEq a SciML. 10 Analytical probabilistic methods. Gaussian least squares methods, Gaussian processes. Package AbstractGP. 11 Monte Carlo methods with a focus on Hamiltonovské MC, probabilistic programming. Package Turing. 12 Design of experiments, development of a solution for linear models, Bayesian optimization.

Learning activities and teaching methods
Laboratory work, Lecture
  • Contact hours - 52 hours per semester
  • Preparation for laboratory testing; outcome analysis (1-8) - 10 hours per semester
  • Individual project (40) - 40 hours per semester
prerequisite
Knowledge
independently use elementary knowledge of mathematical analysis, linear algebra, probability and statistics
develop simple mathematical models using the laws of physics
Skills
program in higher-level computer languges, such as C++, C#, MATLAB or python
studovat odbornou literaturu v anglickém jazyce
Competences
N/A
learning outcomes
Knowledge
using selected methods of machine learning for developing mathematical models from data
use the mathematical model for optimization of the physical system
Skills
elementary use of the Julia language
use existing tools for practical applications
apply machine learning methods for faster innovation and research in the electrical engineering
Competences
N/A
N/A
teaching methods
Knowledge
Lecture
Interactive lecture
Skills
Practicum
Project-based instruction
Competences
Task-based study method
assessment methods
Knowledge
Project
Continuous assessment
Skills
Project
Competences
Continuous assessment
Recommended literature
  • Bishop, Christopher M. Pattern recognition and machine learning. New York : Springer, 2006. ISBN 0-387-31073-8.


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