Content, progress and pedagogy of the module

Aim: This module aims at providing insights into the most foundational architectures of deep learning algorithms within both supervised and unsupervised learning, thus building a strong foundation for further exploration of more specific and cutting-edge techniques. Real-world problems that are approached with the techniques covered in this module include the development of advanced recommender systems (marketing), computer vision models (healthcare, economics), powerful unsupervised pattern recognition systems (fraud detection or credit default prediction in finance) and (attempts of) stock market index prediction.

Content:

This module focuses on the most recent developments in the field of data science that build on deep learning and different architectures of artificial neural networks. While conceptually, these techniques were already conceived in the 70s and 80s, it was only recently that Big Data created a need and modern computers allowed to use them in practice. Today, deep learning algorithms are behind a variety of online and offline applications. They are enabling massive recommender systems in online retail and entertainment and powering artificial intelligence applications in medical diagnostics. Vast interest and investment in R&D within this area spurred progress of these techniques and made them more accessible. Only a few years ago deep learning and AI were barely known outside computer science departments. Today, these approaches are widely used in medicine, natural sciences and increasingly seen in social science as well as humanities.

While many of these techniques constitute compelling approaches, especially for predictive modelling, yet they do not make more traditional modelling approaches (e.g. techniques learned in M1) obsolete, but offers many synergies. Therefore, the module is structured in a way that makes it easy for students to see, where the analysis can make use of deep learning approaches as an alternative to more established techniques (e.g. regression analysis). Emphasis will be put on outlining the cases in which traditional (often leaner) methods are more suited.

Upon completion, students will acquire theoretical and practical knowledge, enabling them to understand and explain central techniques and concepts of deep learning approaches as well as the fundamentals of artificial intelligence for analytics. They will be able to select and apply appropriate methods to real-world problems and critically reflect on them.

Learning objectives

Knowledge

Knowledge:

• Explain the central concepts within deep learning.
• Define key elements of artificial neural networks and depict their functionality.
• Describe main architectures of supervised deep learning algorithms.
• Describe main architectures of unsupervised deep learning algorithms.
• Show insight into recent developments in deep learning and artificial intelligence.
• Reflect on ethical and societal problems concerning the use of artificial intelligence.

Skills

Skills:

• Install and deploy relevant software packages and cloud services for deep learning approaches.
• Select and prepare various types of data for use in deep learning environments.
• Select and construct different kinds of deep learning architectures (e.g. Artificial Neural Networks, Convolutional Neural Networks, Recurrent Neural Networks, Self-Organizing Maps, Restricted Boltzmann Machines).
• Implement “correct” training of selected models.
• Tune and optimise models.
• Utilise trained models for prediction tasks.
• Evaluate model performance.

Competences

Competencies:

• Use deep learning techniques to solve social science problems in Big Data contexts.
• Make informed decisions about the selection of algorithms (also where it is better not to use deep learning/AI techniques at all).
• Identify cases that require particular attention concerning ethical and social consequences of deep learning and AI application.

Type of instruction

Lectures will be complemented by online resources and e-learning tools such as podcasting, online tutorials, and mini-assignments, as integral parts of the teaching methodology in order to enhance student engagement outside the classroom. Physical face-to-face time will be centred around the tacit and interactive components of the problem-solving processes.

Exam

Prerequisite for enrollment for the exam

• A prerequisite for participating in the exam is that the student has participated actively in developing written material during the module.

Exams