A course on deep learning covering athematical and statistical principles of deep learning and selecting suitable deep neural network layers and processing operations for specific tasks.
Course contents
- Deep neural networks layers: convolutional neural networks, recurrent neural networks, transformers, multilayer perceptrons
- Components of deep neural networks: nonlinearities, normalization, subsampling
- Task-specific loss functions
- Training deep neural networks: stochastic gradient descent, chain rule in gradient calculation, and flow of information
- DNN architectures: encoder-decoder structures, autoencoders, U-nets, handling the depth by residual and skip connections
- Supervised, self-supervised, adversarial learning
- Implementations in Python: Pytorch or Tensorflow
Learning outcomes
- Understanding the mathematical and statistical principles of common operations in common deep neural networks (e.g., CNNs, RNNs, normalization, attention mechanisms)
- Capability to select suitable deep neural network layers and processing operations for specific tasks
- Capability to optimise parameters and hyperparameters of deep neural networks for specific tasks using training datasets
- Understand in how to use different machine learning paradigms, like supervised, adversarial and self-supervised learning for training deep neural networks
- Capability to implement deep learning algorithms for specific tasks using deep learning software libraries such as Pytorch or Tensorflow
Course material
Course material will be available in Moodle.
Teaching schedule
Lectures and excercises are scheduled. Participation on campus or online. Lectures will be recorded.
Completion methods
Excercises. Exam on campus.
More information in the Tampere University study guide. You can get a digital badge after completing this course.