Fachbereich 6 Mathematik/Informatik

Institut für Informatik

Navigation und Suche der Universität Osnabrück



Alle Veranstaltungen

Masterseminar Computational Intelligence
Dozent:Jan Hendrik Schoenke, M. Sc., Prof. Dr. rer. nat. Joachim Hertzberg
Veranstaltungstyp:Seminar (Offizielle Lehrveranstaltungen)
Beschreibung:Data stream analysis and learning from data streams are research areas of growing interest as in many application domains large volumes of data are continuously generated, e.g. multi-scale sensor networks, production and manufacturing lines, mobile robots etc., often with a high incoming rate. This results in the need for learning algorithms that can keep up with the incoming data and process them in (near) real-time using a single-pass learning procedure. Thus the desired algorithms need to be fast with bounded memory demands. The streaming nature of the data gives rise to challenges like the afore mentioned incoming rate, but more important are shifts and drifts in the data as the way the data is generated may change over time making former observations obsolete. These challenges are unique to data stream analysis. They relate to the stability-plasticity-dilemma and are accompanied by other data analysis related issues like noise, imbalanced data, curse of dimensionality, non-i.i.d. data distribution, bias-variance-dilemma, outliers, missing data and many more.

There are many approaches to tackle these challenges in data stream analysis which originate from classical automation related fields like System Identification, Filtering and Signal Processing as well as from data science fields like Statistics, Computational Intelligence and Machine Learning. This seminar will focus on recent developments on data stream analysis for tasks like modelling and prediction. The participants familiarize themselves with a chosen topic by implementing related algorithms and evaluating them on benchmark tests, present their results orally and document their work in a report.

Topics of interest are (but not limited to):

- Extreme Learning Machines
- Polynomial Networks
- Evolving Fuzzy Systems
- Non-stationary second order online learning
- Online support vector regression
- Instance based learning
- Locally weighted projection regression

- Incremental Principal Component Analysis
- Incremental Singular Value Decomposition
- Incremental Manifold Learning
- Incremental Tensor Analysis

Useful background information:
- https://www.youtube.com/watch?v=mbyG85GZ0PI&list=PLD63A284B7615313A
- https://www.youtube.com/watch?v=LnlW9gdjWfc&index=10&list=PLTPQEx-31JXgtDaC6-3HxWcp7fq4N8YGr
- http://homes.cs.washington.edu/~pedrod/papers/cacm12.pdf
- http://statweb.stanford.edu/~tibs/ElemStatLearn/
Semester:WS 2016/17
Zeiten:Mo. 10:00 - 12:00 (wöchentlich) - Sitzung
Erster Termin:Mo , 24.10.2016 12:00 - 14:00, Ort: (50/614)