Osion kuvaus

  • During the course, we will discuss advanced deep learning models. Each student will present one topic and implement the presented model in PyTorch. The topics of the seminars will be selected according to the students' interests. The list of possible topics (note that we will not cover deep reinforcement learning in the course):

    1. Optimization and regularization
    • Adversarial deep learning, adversarial training
    • Bayesian neural networks (reserved)
    2. Image processing
    • Capsule networks (reserved)
    • Learning image metrics
    3. Processing of sequential data
    • Attention is all you need, universal transformers (reserved)
    • Neural Turing machine (reserved)
    4. Unsupervised learning
    • Deep autoregressive models (reserved)
    • Variational auto-encoders (reserved)
    • Advances in GANs (reserved)
    • BERT (reserved)
    5. Semi-supervised learning and meta-learning
    • Semi-supervised learning: mean teacher (reserved)
    • Semi-supervised learning: virtual adversarial training (reserved)
    • Few-shot learning: Prototypical networks, MAML
    • Conditional neural processes
    6. Relational models
    • Graph convolution networks (reserved)
    • Interaction networks
    Assessment: a presentation and a project work.

    Registration is closed.