Students will be able to:

(i) select the appropriate neural network architecture for the selected data type

(ii) design the neural network model and select the appropriate optimization algorithm for training

Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning. MIT Press, 2016. http://www.deeplearningbook.org

https://moodle.vscht.cz/enrol/index.php?id=55

Feed-forward neural networks

• basic architectures and activation functions

• optimization algorithms for training

• selection of hyperparameters

Regularization of neural network models

• commonly used regularization techniques - dropout, label-smoothing

Convolutional neural networks

• convolution layers, normalization

• architectures suitable for deep convolutional neural networks

• pre-training and fine-tuning of deep neural networks

Recurrent neural networks

• basic recurrent networks and problems of their training

• LSTM, GRU

• bidirectional and deep recurrent networks

Transformer architecture

Design and optimization of neural networks in various environments - Python, MATLAB

basic programming skills in Python, MATLAB (at least one of them) are advisable

The student passes the practicals by submission of sufficient number of assignments (obtaining the appropriate number of points, including bonus points). The assignments are announced regularly during the whole semester. The student can choose which of the assignments to work on in order to obtain the necessary number of points. The written exam test consists of randomly selected questions from a set of previously announced exam questions. Classification in the exam can be improved or replaced by submission of an extended number of assignments (obtaining the extended number of points).