DTE-2502 Neural Networks - 10 stp

This course can be taken as a single subject course.

However, the student is expected to have some basic knowledge of AL and ML, as well as experience in Python programming.

The course focuses on different approaches into the artificial intelligence domain focusing on neuralnet works.

The course introduces support vector machines (SVM) and compares the advantages of end-to-end training offered by Neural network models. It comprises of five modules:

• Multi-layer perceptron: Focusing on topics such as application / programming / understanding of concept of a perceptron, the multi-layer-perceptron,
• Convolution neural nets: Feed-forward neural nets, convolution neural nets (CNN) with emphasis placed on applications related to image analysis and related topics. VGG-Net, Res-NET.
• Recurrent neural nets: Recurrent Neural Networks (RNNs), Long Short-Term Memory networks(LSTMs), and Gated Recurrent Units (GRUs) for handling sequential data.
• Reinforcement learning: Basic introduction to reinforcement learning and deep reinforcement learning, emphasis will be placed on applications related to Q-learning for training (robotic /autonomous) agents and advance model architectures.
• Generative models: Introduction to Variational auto-encoders (VAEs) and Generative adversarial networks (GANs).

General study qualification with Mathematics R1+R2 and Physics FYS1. Application code: 9391

Recommended prerequisites:

• DTE-2602 Introduction to Machine Learning and Artificial Intelligence
• DTE-2510 Introduction to programming
• DTE-2511 Advanced programming
• Experience in Python programming

On completion of the course, the successful student is expected to have the following:

Knowledge

The student will have:

• An overview of history and numerous approaches within training deep neural networks.
• Understanding of "The curse of dimensionality" in AI.
• Basic understanding of back-propagation and complexity.

Skills

The student should be able to:

• Program, adapt and apply neural nets in different application domains.
• Identify and define features in a complex environment.
• Think critically with theoretical framework underpinning an deep learning architecture.

General Competence

• Can apply the knowledge and skills to solve problems and communicate about the results with other specialists in the field of computer science.

The subject uses so-called "Flipped classroom", i.e., lectures are posted online continuously during the semester in the form of short instructional videos and demonstrations. In addition, exercises and control questions related to each video are used.

The subject teaches in the autumn semester with teacher-led and assistant-led learning and / or exercises.