TU Berlin

Methods of Geoinformation ScienceDeep Learning for Geographical Data

Page Content

to Navigation

Deep Learning for Geographical Data

Course details
Term
SPW
Type
Lecturer
Person in charge
3rd. (WS)
4
Lecture
Exercises
Prof. Dr.-Ing. Martin Kada
Prof. Dr.-Ing. Martin Kada

Course content

  • Introduction to artificial intelligence
  • Artificial neural networks, linear classifier, the perceptron
  • Connected neural networks, training by backpropagation, loss and activation functions
  • Weight initialization and regularization, optimizers, hyperparameter tuning
  • Multi-task learning
  • Convolutional neural networks (CNN)
  • Common network architectures and custom models
  • Deep Learning on 3D data and aerial laser scanning point clouds
  • Time series geographical data
  • Recurrent neural networks (RNN)
  • Attention mechanism
  • Generative learning with Autoencoders and generative adversarial networks (GAN)
  • Deep Learning libraries (e.g. TensorFlow, PyTorch)

Didactic concept

  • Lecture (45%)
  • Hands-on exercises (45%)
  • Independent reading (10%)

Navigation

Quick Access

Schnellnavigation zur Seite über Nummerneingabe