Rare events are events that occur in less than 5% of data points. These events are important for the users of the system, but they are often optimized away by the machine learning algorithms.
Two examples of such rare events are:
- disturbances in radio networks caused by meteorological events (e.g. snow, rain) or obstacles (e.g. construction cranes),
- changes in patient signals indicating specific conditions (e.g. stroke).
In this project, we design and develop methods and tools to identify these rare events and to ensure that machine learning algorithms focus on these events rather than optimize them away.
One of the presentations of our results is included here:
Funding agency: CHAIR (Chalmers AI Research Center)