LFU:online

The subject of this course is an approach to Machine Learning that achieves great power and flexibiity through a representation that follows a hierarchical approach achieving abstracticity from a starting point of simple concepts. These simple concepts are combined in order to represent higher concepts a different layers of the network. Students will acquire the knowledge on different network architectures, parameter choices, learning algorithms and applications.

When programmable computers were first conceived, people wondered whether such machines might become intelligent. Artificial Intelligence (AI) is a thriving field with many practical applications and active research topics. AI systems need the ability to acquire their own knowledge by extracting patterns of raw data. This capability is known as Machine Learning. It can be extremely difficult to extract high-level abstract representations and knowledge just from raw data. Deep Learning tackles this problem by representing that abstract knowledge in terms of simpler representations. Those representations are organized into a hierarchy of layers on top of each other, reason why it is called Deep Learning. The course will cover Deep Feedforward netwoks, regularization, optimization, Convolutional Neural Networks, Recurrent Networks, Sparse Coding, Autoencoders and Deep Generative Models.

Motivated by practical applications, the lecture covers theoretical material that the complementary proseminar exercises in discussions, written problems, and programming projects.

written exam

Deep Learning by Ian Goodfellow, Yoshua Bengio and Aaron Courville. MIT Press 2016.

Elective Module Introduction to Machine Learning

Remark for students of the Bachelor's Programme Computer Science according to the Curriculum 2007W: The course Deep Learning VU 3 can be used as elective module. Allocation of confirmation is required.

Some material builds on concepts taught in the Advanced Machine Learning (703642) course.