Machine Learning for the Quantified Self

2019-2020

Course Objective

The main aims of this course are to make the student familiar with
specific machine learning techniques for quantified self and sensory
data. Both the theory as well as the practical side of matters will be
addressed. More specifically, the following aims are distinguished:
1. Understanding the domain of the quantified self and sensory data
2. Knowing and being able to apply techniques for outlier detection on
sensory data.
3. Knowing and being able to apply techniques for feature engineering on
sensory data.
4. Knowing and being able to apply clustering techniques related to
sensory data.
5. Understanding the theory of supervised learning and its implications.
6. Learning how to apply non-temporal machine learning techniques to
sensory data.
7. Knowing and being able to apply temporal machine learning techniques
to sensory data.
8. Knowing and being able to apply reinforcement learning in the
application of the quantified self.

The course if focused on the Dublin descriptors Knowledge and
understanding (learning new relevant algorithm in this domain); Applying
Knowledge and Understanding (since the aim is also to apply the
algorithms in practice); Making judgements (which technique is
appropriate, how to best apply the techniques for a specific case);
Communication skills (how to report on your approach, choices and your
results), and Learning skills (able to find new relevant techniques,
assess their suitability, etc.).

Course Content

The quantified-self refers to large-scale data collection of a user's
behavior and context via a range of sensory devices, including smart
phones, smart watches, ambient sensors, etc. These measurements contain
a wealth of information that can be extracted by means of machine
learning techniques, for instance for the purpose of predictive
modeling. In addition, machine learning techniques can be a driver for
adaptive systems to support users in a personalized way based on the
aforementioned measurements. The type of data does however require
specialized machine learning techniques to fully exploit the information
contained in the data. Examples of challenges include the temporal
nature of the data, the variety in the type of data, the different
granularity of various sensors, noise, etcetera.

In this course specific techniques to handle quantified self (or broader
sensory data) will be treated. More in specific, it will address:
• Feature engineering (how do we come from raw data to usable features):
* Removing noise from data
* Handling missing data
* Identifying (temporal) features
• Learning of user patterns:
* Temporal machine learning approaches such as recurrent neural
networks, time series analysis
* Clustering approaches with dedicated distance metrics (including
dynamic time warping)
• Adaptive feedback and support
* Reinforcement learning
• Integration of the various components.

In addition, a number of real-life applications will be discussed. Next
to lectures, there will be an extensive practical part, where students
will learn to work with various algorithms and data sets. As a final
assignment, the students will work on a project they propose themselves.

Teaching Methods

The course will be taught in four weeks. Both regular lectures (l, first
two weeks) and practical sessions (p) will be organized. During the
first two weeks the emphasis will be on the lectures and assignments
related to their content. In the third week a larger practical
assignment (in the form a project) will be done which. The course ends
with an exam.

Method of Assessment

Written exam (E) (50%) and practical assignments (A) (50%). For both
parts the grade needs to be sufficient to obtain a final grade. For the
practical assignments the final assignment counts for 60% while
assignment 1 and 2 each count for 20%.

Literature

Hoogendoorn, M. and Funk, B. Machine Learning for the Quantified Self -
On the Art of Learning from Sensory Data, ISBN 978-3-319-66308-1,
Springer Verlag 2018. The book can be downloaded free of charge through
the UBVU, just follow the link to the book:
http://dx.doi.org/10.1007/978-3-319-66308-1

Target Audience

XM_AI, XM_BA, XM_CS

Additional Information

Lecturer:
Dr.M. Hoogendoorn

Recommended background knowledge

Programming experience. Knowledge of basic machine learning algorithms.

General Information

Course Code XM_40012
Credits 6 EC
Period P6
Course Level 400
Language of Tuition English
Faculty Faculty of Science
Course Coordinator dr. M. Hoogendoorn
Examiner dr. M. Hoogendoorn
Teaching Staff

Practical Information

You need to register for this course yourself

Last-minute registration is available for this course.

Teaching Methods Lecture, Practical
Target audiences

This course is also available as: