Time Series Analytics WS'15


Course Information

Type Seminar (7 ECTS)
Lecturer Dr. Hoang Vu Nguyen.
Examiner Dr. Jilles Vreeken.
Email hnguyen (plus) tsa15 (at) mpi-inf.mpg.de
Meetings Tuesdays, 10–12 o'clock in Room E1.7 323. First Lecture on 27th of October.
Max Capacity 10 students. See here how to sign up.
Summary In this seminar, we'll be investigating the following questions: What are time series, where can we find them, and why are they important at all? What are the challenges associated with data analysis on time series? As main theme, we'll look into prediction & forecasting, change detection, Granger causal analysis, pattern discovery, and graph stream processing.

Schedule

Month Day Type Topic Slides Reading Auxiliary
Oct 27 L Introduction PDF [1] Ch 1 &
[2] Ch 14
Nov 3 L Prediction PDF [1] Ch 2.3, 3.1, 3.3,
5.2, 5.4,
6.1, 6.4–6.6
data & code
Nov 10 D Prediction [3] OR [4] data & code
Nov 17 L Granger Causality PDF [5]
Nov 24 D Granger Causality [6] data & code
Dec 1 L Change Detection PDF, PPT [7] Ch 1.1, 1.2, 1.4
Dec 8 D Change Detection [8]
Dec 15 Winter break
Dec 22 Winter break
Dec 29 Winter break
Jan 5 L Motif Detection & Data Streams [2] Ch 14.4
Jan 12 Break
Jan 19 Break
Jan 26 S Maha, Shweta, Dang, Nurzat Student Presentations
Feb 2 S Prabal, Cristian, Amir, Cuong Student Presentations

Lecture type key:

Materials

All recommended reading will be made available here. You will need a username and password to access the papers outside the MPI network. Contact the lecturer if you don't know the username or password.

[1] Brockwell, P.J. & Davis, R.A. Introduction to Time Series and Forecasting. Springer, 2010.
[2] Aggarwal, C. Data Mining: The Textbook. Springer, 2015.
[3] Anava, O., Hazan, E., Mannor, S. & Shamir, O. Online Learning for Time Series Prediction. In COLT13, pages 172-184, 2013.
[4] Anava, O., Hazan, E. & Zeevi, A. Online Time Series Prediction with Missing Data. In ICML15, pages 2191-2199, 2015.
[5] Arnold, A., Liu, Y. & Abe, N. Temporal causal modeling with graphical granger methods. In Proceedings of the 13th ACM International Conference on Knowledge Discovery and Data Mining (SIGKDD), San Jose, CA, pages 66-75, 2007.
[6] Sun, Y., Li, J., Liu, J., Chow, C., Sun, B.Y. & Wang, R. Using causal discovery for feature selection in multivariate numerical time series. Machine Learning, 101(1-3):377-395, 2015.
[7] Basseville, M. & Nikiforov, I.V. Detection of Abrupt Changes: Theory and Application. Prentice Hall, 1993.
[8] Kuncheva, L.I. & Faithfull, W.J. PCA Feature Extraction for Change Detection in Multidimensional Unlabeled Data. IEEE Trans. Neural Netw. Learning Syst., 25(1):69-80, 2014.

Structure and Content

In general terms, the course will consist of

  1. introductory lectures,
  2. reading, discussing and presenting scientific articles, and
  3. a 'practical' assignment
At a high level, the topics we will cover will include
  1. Prediction and Forecasting
  2. Change Detection
  3. Granger Causality
  4. Motif Detection
  5. Graph Streams Processing
Loosely speaking, students will learn about the information theoretical optimal approach to learn from data, as well as learn about and how to apply the more practical approach of MDL for a variety of model selection problems.

Prerequisites

Students should have basic working knowledge of data analysis and statistics, e.g. by successfully having taken courses related to data mining, machine learning, and/or statistics, such as Topics in Algorithmic Data Analysis, Machine Learning, Probabilistic Graphical Models, Statistical Learning, Information Retrieval and Data Mining, etc.

Course format

The course has two hours of scheduled meetings per week. The first weeks will feature regular lectures covering the basic topics of the course. During the second phase the students will write essays based on the material covered in the lectures and scientific articles assigned by the lecturer. We will discuss materials in detail during the meeting. During the third phase the students will write an essay based on scientific articles assigned to them by the lecturer and will prepare a presentation to be held during the meeting.

There will be no weekly tutorial group meetings.

Signing Up

Discussions will be an essential element of this seminar, and hence there will a maximum of ten (10) participants.

Students that want to participate should send an email to the lecturer on or before Friday 23 October 16:00, in which they specify: a) name, b) matriculation number, c) level (bachelor, n-th year master, PhD student), d) relevant courses taken so far, e) short motivation for why they want to participate in the Time Series Analytics seminar.

When more than ten (10) students apply preference will given to highly motivated M.Sc. students. Students will be notified on or before Friday 23 October 18:00.