Written exam (in case of a small number of participants: oral tests)

Advanced ANN topics. First, revisiting backpropagation and backpropagation
through time; then: Advanced Recurrent Neural Networks (LSTM, GRU);
Very Deep Learning and Generative Adversarial Networks; Spatial and Temporal
Convolution; Reservoir Computing; Neuroevolution; Attention and Routing
Networks; Autoencoders and Restricted Boltzmann Machines; Gain Fields and
Switching Networks; Latent Space Visualization techniques; Generative Inference

Students know about and how to apply generative and typically recurrent artificial
neural networks in various domains including data classification, image
recognition, language processing, spatially-invariant recognition, spatial transformations,
and spatial mappings. They can apply complex, generative artificial
neural networks from scratch as well as with available tools. They know how to
optimize weights and network structures by means of gradient descent as well
as by alternative methods. They can use complex recurrent network structures
to selectively process aspects of the data. They know how to apply generative
networks as model-predictive neural controllers and as well as long-range
temporal predictors. They can combine retrospective latent state and motor
inference techniques with prospective motor control.

Literatur / Literature:
Wird in der Veranstaltung ausgegeben (Buchkapitel und Artikel in Englisch). / Will be supplied (book chapters and papers in English).

Voraussetzungen / Prerequisites:
Vorkenntnisse in maschinellem Lernen, Künstlichen Neuronalen Netzen, Deep Learning oder Künstlicher Intelligenz sind notwenig. / Knowledge about machine learning, artificial neural networks, deep learning, or artificial intelligence is required.