Unsupervised & Reinforcement Machine Learning
Course is scheduled in the period March 9 - April 20, 2020.
Course location: Erasmus University Rotterdam campus. Max. capacity: 25 students.
Registration for this course
Go the the registration page.
Classes are scheduled on Monday mornings starting on March 9 until April 6, and on Thursday mornings starting March 26 until April 9. Exam on April 20. Location: Campus Erasmus University.
Unsupervised & Reinforcement Machine Learning discusses unsupervised and reinforcement learning approaches often used to solve management science problems. The first part focusses on unsupervised machine learning techniques for finding meaningful relations between all variables in a data set simultaneously. In contrast to supervised machine learning, discussed in a previous course, in unsupervised techniques all variables play similar roles. Therefore, the relationships among all variables must be modelled, whereas in supervised learning only the relationships between the target variable and the features are of direct interest. An important application of unsupervised learning techniques in management is customer segmentation in targeted marketing.
The other main focus of this course is on reinforcement learning, where algorithms are used for sequential decision-making under uncertainty. Here, the focus will be on the design, solution, and implementation of reinforcement learning methods. Sequential decision problems involve a trade-off between exploitation (acting on the information already collected) and exploration (gathering more information). These problems arise in many important domains, ranging from online advertising, clinical trials, website optimization, marketing campaigning and revenue management.
An overview of techniques and ideas to be treated are:
- principal components analysis (PCA)
- cluster analysis (k-means, hierarchical)
- multidimensional scaling
- introduction to reinforcement learning and multi-armed bandits
- Examples, formulation and preliminary results
- multi-armed bandit methods
- Optimality of index-based policies
- Heuristics: one-step look ahead, regret policies, Thompson sampling
- multi-armed bandit modeling strategies and applications.
By the end of the course students will be able to:
- Understand the fundamental building blocks of deep learning methods;
- Understand the weaknesses and strengths of the different architectures;
- Know how to tackle weaknesses and tailor the model for a particular application;
- Program these methods, and
- Be able to describe the numerical computational steps applied by the machine.
Coordinator/Lecturer: prof. Gui Liberali (EUR),
Lecturer: dr. Pieter Schoonees (EUR)
Course fee for internal research master and PhD students: € 1.000,-
Course fee for external PhD students: € 1.500,-
Recommended knowledge: Business Foundations, Programming Basics (specifically programming in R, use of R Markdown or knitr), Mathematics, Statistics, Decision Theory for Business, Econometrics
Required knowledge: Linear Algebra, Linear and Logistic Regression
Link to course manual
Registration for this course
Go to the registration page