Course detail
Data Analysis and Artificial Intelligence Methods
FSI-TPX Acad. year: 2027/2028 Winter semester
Upon successful completion of the course, the student will be able to:
1. Explain the fundamental principles of probabilistic and statistical modeling of physical and experimental data.
2. Apply Bayes’ theorem, the maximum likelihood method, and the least squares method to parameter estimation problems.
3. Apply regression, classification, and clustering methods in data analysis, including the Naive Bayes classifier.
4. Perform correlation analysis and dimensionality reduction using Principal Component Analysis (PCA), and explain the significance of eigenvalues and eigenvectors in determining principal components.
5. Explain the principles of parameter optimization through the minimization of objective functions, understand gradient descent methods (Batch, Mini-batch, and Stochastic Gradient Descent) and the principles of the Metropolis Monte Carlo method, and assess their applications in parameter estimation and the search for suitable model parameters.
6. Explain the principles of neural networks, identify the fundamental building blocks of an artificial neuron, understand neural network architectures and training principles, recognize the significance and appropriate use of activation functions (ReLU, sigmoid, softmax) and loss functions, particularly categorical cross-entropy, and describe the fundamental principles of transformers, physics-informed neural networks, and reinforcement learning.
7. Use the Python programming language and appropriate libraries for numerical computations, data processing, visualization, and analysis.
8. Design and implement a solution to a selected data analysis problem, evaluate the obtained results, and interpret their physical significance.
9. Critically assess the outputs of AI assistants and verify the correctness of the computational procedures used.
10. Compare the fundamental principles and application areas of various statistical, optimization, and machine learning methods covered in the course, and assess their suitability for specific problems.
Language of instruction
Czech
Number of ECTS credits
2
Supervisor
Department
Learning outcomes of the course unit
Upon successful completion of the course, students will demonstrably be able to:
1. Import, preprocess, statistically analyse and appropriately visualise a selected set of experimental or physical data using Python.
2. Select a suitable data analysis method according to the characteristics of the data and the problem being addressed, and justify their choice.
3. Apply a selected method, such as parameter estimation, regression, PCA, classification or clustering, and assess whether the resulting findings are meaningful.
4. Produce and submit functional, understandable and reproducible Python code together with a concise report describing the problem, approach and results.
5. Interpret the results in the context of the problem, identify key limitations of the method used and potential sources of uncertainty.
6. Explain and defend their own solution while demonstrating basic familiarity with other methods covered in the course; compare their possible applications, advantages and limitations without being required to implement every method in detail.
7. Critically verify procedures and code developed with the assistance of AI tools and take responsibility for the correctness of the submitted solution.
Entry knowledge
Basic knowledge of mathematical analysis, linear algebra, probability and statistics is required, particularly working with functions, vectors, matrices and basic statistical characteristics of datasets.
Basic programming skills in Python are required, including variables, data structures, conditional statements, loops and functions. Students must be able to independently modify and execute simple programs.
Prior knowledge of specialised libraries for data analysis and machine learning is not required.
Rules for evaluation and completion of the course
The course is completed by a graded course credit (klasifikovaný zápočet) based on an individual seminar project and a professional discussion of its solution.
Successful completion requires meeting the attendance requirements for compulsory practical classes, submitting a seminar project on a topic approved by the instructor, and successfully completing a professional discussion. Attendance at lectures is optional. The instructor determines how missed practical classes must be made up, depending on the extent and content of the absence.
The seminar project consists of applying a selected statistical, data analysis or machine learning method to an individual problem. Students are encouraged to choose a topic related to their bachelor's thesis, experimental work or another area of interest. Students submit a short report of approximately two pages in PDF format describing the problem, methods used, results obtained and their interpretation, together with functional Python source code allowing the analysis to be reproduced.
During the project defence, students must demonstrate both an understanding of their own solution and basic familiarity with other methods covered in the course. They must be able to explain the principles of the methods used, justify their selection, interpret the results and discuss possible applications of other methods covered in the course, including their advantages and limitations. Detailed implementation knowledge of every method covered is not required.
The final grade (A–F) is based on an overall assessment of the technical correctness of the solution, suitability of the methods, implementation quality, interpretation of results, ability to defend the chosen approach and familiarity with other methods covered in the course. Assessment primarily emphasises the correctness and suitability of the chosen approach, understanding of the methods used and quality of the interpretation of results, rather than algorithmic complexity. The use of advanced methods is not required to achieve an excellent grade.
A (excellent): Independent and methodologically correct solution, high-quality interpretation of results, excellent understanding of the methods used and the ability to discuss alternative methods competently.
B (very good): High-quality solution with minor shortcomings, very good understanding and good familiarity with other methods.
C (good): Functional solution with some shortcomings, good understanding of the fundamental principles and adequate familiarity with other methods.
D (satisfactory): Mostly functional solution with more significant shortcomings, sufficient understanding of the student's own approach and basic familiarity with the methods covered.
E (sufficient): Minimally acceptable solution meeting the essential requirements, ability to explain the key aspects of the chosen approach and demonstration of minimum familiarity with other methods covered in the course.
F (fail): Failure to meet the essential requirements, serious methodological errors, inability to explain the procedures used or failure to demonstrate basic familiarity with the methods covered.
The use of AI assistants in the seminar project is permitted. However, students must be able to explain, verify and defend all essential parts of their submitted work.
An unsatisfactory seminar project may be revised and resubmitted once within a deadline specified by the instructor. The revised project will be reassessed and, if necessary, followed by another professional discussion. This procedure must comply with the applicable Brno University of Technology study regulations.
Aims
Upon successful completion of the course, students will be able to:
1. Explain the basic principles of probabilistic and statistical modelling of physical and experimental data.
2. Apply Bayes' theorem, maximum likelihood estimation and least-squares methods to parameter estimation problems.
3. Use regression, classification and clustering methods for data analysis.
4. Perform correlation analysis and dimensionality reduction using principal component analysis (PCA).
5. Explain the principles of artificial neural networks, their architectures, activation functions and training procedures.
6. Describe the fundamental principles of transformers, physics-informed neural networks and reinforcement learning.
7. Use Python and appropriate libraries for numerical computations, data processing, visualisation and analysis.
8. Design and implement a solution to a selected data analysis problem, evaluate the results and interpret their physical significance.
9. Critically assess outputs generated by AI assistants and verify the correctness of the computational procedures used.
10. Compare the basic principles and applications of different methods covered in the course and assess their suitability for a particular task.
Study aids
Lecture presentations, programming examples, datasets, exercises and information about the seminar project are available on the course website:
https://sites.google.com/view/adamui
Practical work is carried out using Python, particularly the NumPy, SciPy, Pandas, Matplotlib, scikit-learn and PyTorch libraries. Students may use the Spyder development environment or Google Colab.
The study programmes with the given course
Programme B-FIN-P: Physical Engineering and Nanotechnology, Bachelor's, compulsory
Type of course unit
Lecture
13 hours, optionally
Syllabus
I. Probabilistic Modelling and Optimisation
1. Introduction to AI in Physics. Basic concepts of data analysis, machine learning and artificial intelligence. Motivation, AI assistants, Python, Spyder and Google Colab.
2. Conditional Probability and Bayesian Inference. Conditional probability, Bayes' theorem, Bayesian classification and the Naive Bayes classifier.
3. Parameter Estimation and Optimisation. Maximum likelihood estimation (MLE), least-squares methods, numerical optimisation and stochastic gradient descent (SGD).
II. Classical Data Analysis and Machine Learning Methods
4. Covariance, Correlation and PCA. Random vectors, covariance and correlation matrices, principal component analysis and dimensionality reduction.
5. Experimental Spectral Analysis. Spectral data processing, examples from XPS and Raman spectroscopy, fitting, PCA and classification.
6. Regression and Fitting of Physical Data. Linear models, multiple regression, parameter estimation and Bayesian approaches.
7. Classification and Clustering. Support vector machines (SVM), k-means clustering and applications to physical data.
III. Neural Networks and Reinforcement Learning
8. Neural Networks and PyTorch. Perceptrons, neural network architectures, activation functions, training and backpropagation.
9. Transformer Architecture. The attention mechanism, basic transformer architecture and its connection to modern AI assistants.
10. Physics-Informed Neural Networks (PINNs). Incorporating physical equations and constraints into neural network training.
11. Reinforcement Learning I. Agents, environments, states, actions, rewards and the basic principles of reinforcement learning.
12. Reinforcement Learning II. Practical reinforcement learning examples, simulations and pendulum stabilisation.
13. Review and Practical Applications. Summary and integration of the methods covered, with discussion of their applications to physical data analysis.
Exercise
13 hours, compulsory
Syllabus
1. Python for Data Analysis. Working with a development environment, importing and visualising data, basic operations using NumPy, Pandas and Matplotlib.
2. Probability Computations. Simulation of random events, conditional probability, Bayes' theorem and the Naive Bayes classifier.
3. Parameter Estimation. Implementation of maximum likelihood estimation, least-squares methods and numerical optimisation.
4. Correlation Analysis and PCA. Computation of covariance and correlation matrices, principal component analysis and visualisation of reduced datasets.
5. Experimental Spectral Processing. Importing and preprocessing spectral data, fitting and analysis of results.
6. Regression Modelling. Linear and multiple regression, parameter estimation and assessment of model quality.
7. Classification and Clustering. Practical applications of SVM and k-means to selected datasets.
8. Neural Networks. Implementation of a simple neural network, selection of activation functions and model training using PyTorch.
9. Transformers. Demonstration of the attention mechanism and work with an example model.
10. Physics-Informed Neural Networks. Demonstration of a simple physical problem solved using a PINN.
11. Reinforcement Learning. Definition of environments, agents and reward functions, and simple simulation tasks.
12. Control of a Physical System. Demonstration of reinforcement learning using pendulum stabilisation.
13. Review and Practical Applications. Integration of the methods covered, selected practical exercises and consultations on their application to students' own datasets.