Name  Frau Prof. Dr. Christa Cuchiero 
Adresse  Universität Wien Kolingasse 1416 1090 Wien AUSTRIA 
Telefon  00431427738690 
cchrista@ethz.ch  
Departement  Management, Technologie und Ökonomie 
Beziehung  Dozentin 
Nummer  Titel  ECTS  Umfang  Dozierende  

365118300L  Reinforcement Learning: Insights and Applications Exclusively for MAS MTEC students (1st and 3rd semester).  1 KP  1S  C. Cuchiero, B. J. Bergmann, J. Teichmann  
Kurzbeschreibung  Reinforcement learning (RL) is a field of machine learning that focuses on developing algorithms that enable an agent by novel machine learning technologies to learn optimal strategies through interaction with its environment. In this course we shall understand the main building blocks of (deep) RL and we shall discuss recent applications from finance and robotics.  
Lernziel  After taking this course, students will  have an understanding of the fundamentals of reinforcement learning (RL), including the definition of an agent, environment, and rewards.  understand the idea of a Markov Decision Process, which is a mathematical framework used to model decisionmaking problems in RL.  understand the concept of a value function, which is used to measure the expected reward an agent can receive from a given state.  review various techniques for optimizing an agent's policy to maximize its expected reward  get an idea of Deep Reinforcement Learning: We will explore the use of deep neural networks in reinforcement learning and their advantages over traditional RL methods.  will understand the concept of Partially Observed Markov Decision Processes (POMDP) and its relation to MDPs.  gain handson experience with RL algorithms (optional) for MDPs and POMDPs.  see applications of DRL with a discussion of the realworld applications including finance and robotics.  
Inhalt  Reinforcement learning is a subfield of machine learning that focuses on developing algorithms that enable an agent to learn through trial and error by interacting with its environment. RL differs from other ML algorithms, e.g. supervised learning in not needing labelled input/output pairs to be presented, and in not needing suboptimal actions to be explicitly corrected. Instead, the focus is on finding a balance between exploration (of uncharted territory) and exploitation (of current knowledge). The environment is typically stated in the form of a Markov decision process (MDP). In this course we will go through the main architecture of reinforcement learning and review some of its applications. On day 1 of the 2day course the concept of a Markov Decision Process (MDP), its value function and the Bellmann equation are introduced and discussed. Several classical and ML powered algorithms are introduced and showcases presented. On Day 2 the concept of a partially observed Markov Decision Process is introduced. Aspects of Filtering and embedding partially observed Markov decision processes into the framework of MDPs are presented. Showcases from Robotics and Finance with an emphasis on the latter are presented in theory and applications. An understanding of basic machine learning concepts is welcomed but not mandatory (e.g. you took the class “Fundamentals on ML for Executives” or “AI for Executives”). In the beginning of the course, we will do a short primer on mathematics and statistics and some fundamental aspects of machine learning. We will provide coding examples for those you would like to follow the code. Grading (ungraded semester performance) is based on active participation in the class and a short written report (ungraded) after the course.  
Kompetenzen 
