Within the seminar, an overview about important facets of practically relevant Deep Neural Networks (DNNs) will be given in the form of recent original publications from the literature. Topics which will be covered include the following:
- efficient training of DNNs
- automatic hyperparameter optimization
- adversarial examples for DNNs
- generative adversarial networks and their efficient training
- deep reinforcement learning
- deep recurrent and recursive networks
- neural Turing machines
- applications for language translation
- applications for tracking in vision
- applications for privacy preserving data storage
Teilnahmevoraussetzungen, notwendige Vorkenntnisse
Knowledge of basic math and computer science is required. Some knowledge about machine learning might be benefitial for some of the topics.
The articles covered in the seminar are available in the Lernraum / Dokumentenablage.
Further reading is available on the internet such as:
- the book: http://www.deeplearningbook.org/
- another very gentle introductory book: http://neuralnetworksanddeeplearning.com/
- collection of material: http://deeplearning.net/
- link to Andrew Ngs courses: https://www.deeplearning.ai/
- short introduction: https://machinelearningmastery.com/what-is-deep-learning/
Typically, one of the standard frameworks are used in practical applications, such as tensoflow, theano, pytorch, keras (on top of theano/tensorflow) (all in python), caffe (C++), deeplearning4you (java). Python seems the dominant language, currently. These frameworks come with an embedded technology to train the networks on suitable GPU.
Zu dieser Veranstaltung existiert ein Lernraum im E-Learning System. Lehrende können dort Materialien zu dieser Lehrveranstaltung bereitstellen: