Loading
0000001145 20W 6SWS PR Master-Praktikum - Machine Learning in Crowd Modeling & Simulation (IN2106, IN4267)   Hilfe Logo

LV - Detailansicht

Wichtigste Meldungen anzeigenMeldungsfenster schließen
Allgemeine Angaben
Master-Praktikum - Machine Learning in Crowd Modeling & Simulation (IN2106, IN4267) 
0000001145
Praktikum
6
Wintersemester 2020/21
Informatik 5 - Lehrstuhl für Wissenschaftliches Rechnen (Prof. Bungartz)
(Kontakt)
Details
Zuordnungen: 1 
Angaben zur Abhaltung
Human crowd movement is comprised of highly complex dynamical systems, with hundreds or even thousands of acting and reacting participants that show several emergent behavior patterns. The scientific understanding of crowds involves topics ranging from mathematical modeling via algorithm development for machine learning (ML) from experimental and simulated data as well as the implementation of simulation software up to research in psychology and sociology.
Participants in this lab course will learn about the core mechanics in human movement and interactions in crowds. The current state of the art in mathematical modeling will be discussed, and students will implement models in several exercises. As a reference, the students will be introduced to the crowd simulation software VADERE (www.vadere.org).
After this introduction to the modeling of crowds, we will discuss current machine learning approaches to analyze the simulated results, as well as experimental data. Techniques from statistics, dynamical systems theory, manifold learning, and numerical analysis will be introduced in short lectures, implemented by the students, and then used by them to analyze their own simulation results from previous exercises.
In a final project, the participants can choose to focus on their own model of a crowd, a specific aspect of crowd simulation, or a particular technique in ML to analyze simulation or experimental data.
All exercises and the final project are designed for groups of three to four students. This results in a course size of about 30 students, which can be extended up to 40 by increasing the group size.
The lab course covers the following topics:
• Introduction to the modeling of crowds
• Introduction to dynamical systems theory
• Introduction to appropriate ML techniques
• Numerical analysis of complex systems
• Implementation of simulation software extensions and validation of models to data
• Implementing ML techniques with application to simulation software results
- Familiarity in either Java or Python
- Basic concepts of linear algebra (matrix/vector computations, eigendecomposition)
After the successful participation in the module, students are able to independently implement a code for the numerical simulation of human crowds and have a basic understanding of the current state of the art in crowd modelling. The participants know how to validate new models, verify their code and the resulting simulations. They are able to analyze the simulation results with appropriate tools used in ML and develop their own software. This includes the implementation of some of the main tools for the analysis of complex systems (numerical bifurcation software, state space embeddings through time delays, reconstruction of vector fields from observation data).
Englisch
Gruppenarbeit
Details
Für die Anmeldung zur Teilnahme müssen Sie sich in TUMonline als Studierende*r identifizieren.
Anmerkung: Matching System + Preliminary Meeting
Zusatzinformationen
- Book: Bungartz, H.-J.; Zimmer, S.; Buchholz, M. & Pflüger, D. “Modeling and Simulation: An Application-Oriented Introduction” Springer, 2014.
- Article: Adrian, J. et al. “A Glossary for Research on Human Crowd Dynamics” Collective Dynamics, Forschungszentrum Juelich, 2019, 4.
- Article: Budišic, M.; Mohr, R. & Mezic, I. “Applied Koopmanism” Chaos, 2012, 22, 047510.
- Article: Coifman, R. R. & Lafon, S. “Diffusion maps” Applied and Computational Harmonic Analysis, Elsevier BV, 2006, 21, 5-30.
Online Unterlagen
E-Learning Kurs (Moodle)