URL study guide
https://tue.osiris-student.nl/onderwijscatalogus/extern/cursus?cursuscode=2AMS40&collegejaar=2026&taal=enDescription
This course is designed to help students understand the fundamental aspects of how to turn a wealth of data and information into good decisions. As such the course 1. capitalizes on the knowledge the students have previously gained within the DS&AI MSc in data analytics (descriptive and predictive analytics); 2. enhances the students' knowledge of prescriptive analytics; 3. helps the students to deeply understand and apply learning methods and data-driven optimization techniques to Stochastic Operations Research (SOR) applications; 4. sets the foundations for the students to develop further such learning methods and techniques.Week-by-week overview of the envisioned contents:
W1 Introduction to probabilities and stochastic models. Connection with data analytics (descriptive and predictive methods).
W2 Model based decision making; Infinite horizon; Discounted and average cost criteria; Bellman equations.
W3 Numerical algorithms; Value iteration and policy iteration; Approximate solutions.
W4 Project/Assignment 1.
W5 Extending model-based decision making to data-driven model-free; Q-learning; Double Q-learning. (`Deep' learning methods involving neural networks are covered in the Reinforcement Learning course, and will only be briefly mentioned but not treated in this course.)
W6 Multi-armed bandit problems.
W7 Probabilistic constraints; Beyond the discounted and average cost criteria; Risk-averse optimization.
W8 Project/Assignment 2.
Objectives
An exciting potential that we aim to address in this course is to explore how the presence of data can be used to verify and/or (partially) replace structural assumptions in existing mathematical models and how physical properties captured by the model can compensate for insufficient data.All in all, we expect after the successful completion of the course that
- Students should understand the distinction between descriptive, predictive and prescriptive analytics: Descriptive analytics parses historical data to better understand “what happened?”. Predictive analytics uses historical data to better predict “what will happen?”. Prescriptive analytics builds upon descriptive and predictive analytics. It is the final frontier of analytics as it goes beyond predicting future outcomes by also suggesting actions to benefit from the predictions and showing the implications of each decision option.
- Students should understand the relation between data, mathematical/physical models, and solution techniques. The presence of each of these entities is necessary for meaningful prescriptive analytics, and when insufficient data is available, a (partly) model-based approach may be crucial to compensate for the lack of data.
More concretely, we now list below several learning objectives that students should have reached after the successful completion of the course:
- Modeling: Students should be able to transform a real-life problem into a mathematical decision making / optimization model and understand how to incorporate stochastic characteristics and uncertainty in optimization problems.
- Solving: Students should be able to pick the most suitable solution algorithm to solve the decision problem (viz. understand the pros and cons of different solution methods and their underlying assumptions).
- Calculating: Students should be able to use basic software allowing them to solve data-rich problems (problems involving significant amounts of data).
- Critical thinking: Draw conclusions from the solutions and discuss issues that might have threatened their validity.
- Reporting: present results in writing.