DISA

• When? October 4th 12-13
• Where? Online – the link will be sent to those who sign up
• Registration? Sign up via this link no later than October 3rd.

This seminar will be presented by the DISTA research group within DISA, you will meet and listen to Welf Löwe, Maria Ulan, Morgan Ericsson, Anna Wingkvist

Aggregation combines several independent variables to a dependent variable. The independent variables are different, possible mutually dependent observations of a real world. The dependent variable should preserve properties of the independent variables, e.g., the ranking or relative distance of the independent variable tuples, and ultimately the properties of the real world. However, while there usually exist large amounts independent variable tuples, there is no ground truth data available mapping these tuples to the corresponding dependent variable values. This makes aggregation an unsupervised machine learning problem, as opposed to, e.g., regression where data comprises independent variable tuples and the corresponding dependent variable values.

Instances of the problem frequently occur in software engineering, e.g., when trying to assess the quality of software by metrics. Metrics (independent variables) can easily be measured for a lot of software artifacts, but it is hard to measure quality (dependent variable). Instances also occur in many other assessment situations including, but not limited to the assessment of project proposals, financial investments, and human movements.

In our talk, we present
1) aggregation as unsupervised learning including unweighted and weighted approaches
2) ways to evaluate and compare different aggregation approaches including an evaluation of the approaches introduced in 1)
3) applications to software engineering problems applying the evaluation introduce in 2)

The recording of this session and previous recordings will be available at the following link