School blog
Graduation project: Machine learning!
Coding your own neural network from scratch?
In today's interview, we introduce Matěj Fibiger and his outstanding graduation project.
His topic was: Machine Learning ITG — a neural network for recognising handwritten digits.
What did Matěj's project achieve?
No shortcuts: Matěj did not use existing specialised machine learning frameworks. He implemented the key parts of the system himself.
Advanced mathematics in practice: He thoroughly understood and applied the principles of neural networks, gradient descent and backpropagation.
A complete solution: It is more than a model. The project includes logically organised modules for training, input validation and hyperparameter optimisation.
A visual demonstration: The application also includes a simple graphical user interface (GUI) to demonstrate how recognition works in practice.
Matěj showed tremendous initiative and the ability to turn challenging theory into fully functioning code. We also appreciate his critical assessment of the project's limitations and his plans for further improvements, such as GPU acceleration and regularisation.
The interview was conducted by final-year student Robert. Thank you!
