Miguel Angel Meza Martinez

Status: Closed

Problem Description

In the field of Physiological Computing research has been made to draw a mapping between annotations of affective experiences and measurable variables related to them. For this purpose, Ratings have been stablished as one of the most popular affect annotation tools varying from simple Likert scales to self-assessment manikins and the rating scales of the discrete states in the Geneva emotion wheel.

Machine Learning has been incorporated as a useful approach to be able to map the complex data generated when measuring physiological signals to the corresponding affective state. One of the most common approaches to achieve this mapping consist on using Classification algorithms which are applied after the rating annotations are transformed into nominal representation (classes) – e.g. low, medium and high. However, important disadvantages of treating ratings as nominal representation for a Classification have been discussed in the literature. Some of these disadvantages include the loss of information when transforming ratings from ordinal to nominal and the induce experimental biases generated via the selection of which ratings are assigned to each class.

A Preference Learning (PL) approach has been presented as an alternative to Classification algorithms to perform a better mapping of physiological signal to the corresponding affective state by addressing some of the disadvantages that Classification presents. Research has shown that PL methods could lead to more efficient, generic and robust models which could capture more information about the annotated affect.

Goal of Thesis

To implement a Preference Learning algorithm in the context of Physiological Computing in order to compare its performance against a Classification approach in the mapping of measured physiological variables to affective states by processing ratings of affect annotations.

• Investigate current state of the art in Preference Learning and its implementation to rank affective annotations.
• Design and develop (program) a Preference Learning model that ranks affective annotations.

Skills required

• Strong analytical skills
• Very good time management, organizational and communication skills
• Interest in Machine Learning and Data Analysis
• Development skills
• Programming skills for python and knowledge of common Machine Learning libraries such as sklearn.
• Good English skills (as the language of the thesis is English)

Contact

miguel.martinez@kit.edu