Deep learning

  • Course teacher Marko Subašić
  • Associate teachers -
  • ECTS credits 5
  • Number of hours: Lectures + Seminars + Exercises 30 0 6

Course objectives

Acquire the basics of fully connected, convolutional, recurrent and generative deep models and master their application in practice.

Through numerous successful applications, the mentioned techniques have demonstrated a successful ability to learn and understand input data (eg. understanding scenes in image analysis). Such learning models and algorithms are to a greater or lesser extent inspired by biological neural networks and their functioning, but in the end they deviate significantly from them. Nevertheless, any successful deep learning model represents a potential explanation of the unknown functioning of biological neural networks, their learning and their understanding of input information.

Enrolment requirements and/or entry competences required for the course

-

Learning outcomes at the level of the programme to which the course contributes

  • Apply theoretical knowledge of the fundamentals of the six core disciplines and their relationship within cognitive science.
  • Critically evaluate cognitive science findings and synthesize information to be employed in a collaborative professional environment.
  • Participate in data-driven innovation projects and apply appropriate data science tools.
  • Apply AI tools in concrete tasks and practical contexts.

Course content (syllabus)

  1. Neural networks introduction
  2. Artificial neural network basic concepts and building blocks
  3. Artificial neural network basic architectures
  4. Artificial neural network basic learning algorithms
  5. Deep neural network strengths and weaknesses. Applications.
  6. Deep neural network specifics: activation functions, regularization, momentum, adaptive learning
  7. Deep neural network specifics: activation functions, regularization, momentum, adaptive learning
  8. Deep convolutional networks: layers, architectures, visualization, fine tuning, applications, implementation
  9. Deep convolutional networks: layers, architectures, visualization, fine tuning, applications, implementation
  10. Fully convolutional networks: architectures, applications
  11. Interpretability of neutral network results
  12. Deep recurrent neural networks: RNN, bidirectional RNN, deep RNN, long short-term memory, sequence modelling, applications
  13. Deep recurrent neural networks: RNN, bidirectional RNN, deep RNN, long short-term memory, sequence modelling, applications
  14. Deep generative models: stacked RBMs, convolutional autoencoders, variational autoencoders, adversarial models
  15. Deep generative models: stacked RBMs, convolutional autoencoders, variational autoencoders, adversarial models

Student responsibilities

Class attendance. Engagement in classroom activities.

Required literature

  • -

Optional literature

  • Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT Press. 2017.
  • Eli Stevens, Luca Antiga, Thomas Viehmann. Deep Learning with PyTorch. Manning Publications 2020.