Direkt zum Inhalt springen
login.png Login    |
de | en
MyTUM-Portal
Technical University of Munich

Technical University of Munich

Sitemap > Bulletin Board > Diplomarbeiten, Bachelor- und Masterarbeiten > MSc Thesis / Project Study (IN / MA): Improving Sample Efficiency in Multiagent Reinforcement Learning via Advanced Monte Carlo Methods
up   Back to  News Board    previous   Browse in News  next    

MSc Thesis / Project Study (IN / MA): Improving Sample Efficiency in Multiagent Reinforcement Learning via Advanced Monte Carlo Methods

16.10.2020, Diplomarbeiten, Bachelor- und Masterarbeiten

The chair for Decision Sciences and Systems (DSS) at the Department for Informatics is offering the following MSc Thesis or project work topic to students interested in Applied Mathematics, Machine Learning and Game Theory.

Current research at our institute explores the computation of market equilibria in game-theoretic settings via multi-agent machine learning techniques: To do so, market participants update their behavior via Machine Learning on data of past market outcomes. To achieve good precision, such algorithmic techniques require the simulation of millions of episodes and computationally intensive Monte-Carlo integration.
In this thesis, you will focus on advanced methods from applied mathematics (e.g. stratified sampling, quasirandom numbers, low-discrepency sequences, variance reduction techniques) to analyze and improve sample-efficiency in this machine-learning setting. The thesis should comprise both theoretical analysis and empirical work using implementations in python/pytorch.

This topic is best-suited for a Master's thesis or graded project (e.g. IDP / Guided Research, ...) in a degree program offered by the centre for Mathematics or the degree program Informatics, but we will consider students from other programs/departments as well, if the're a good match. (Note: We can supervise theses in the IN, MA and WI departments. Students from other departments will have to find an eligible co-supervisor.)


Your profile:
- Strong background in applied mathematics (e.g. statistics, probability theory, numerics ...)
- Previous experience with programming in python. Experience with efficient array-based programming (numpy / pytorch) is desirable but not required.
- Basic understanding of game theory and machine learning is desirable.

If you're interested or have any questions, please contact stefan.heidekrueger@in.tum.de

When applying, please attach a recent CV and transcript of records.

Kontakt: stefan.heidekrueger@in.tum.de

Todays events

00:00 - 23:59

Von der Erfindung zum Patent
(Lehrveranstaltung)

Calendar of events