Imagine that we are not just given training samples (i.e. inputs and outputs) but specifically each of the samples is associated to a “cost” and might have a given “accuracy”.
This project investigates, how to find optimal machine learning models that do not only have a minimum loss / maximum accuracy but also a minimal overall cost. That is, we face a much more complex optimization task.
In the project, we touch topics in:
- various regression ML methods
- approximation theory
WARNING: Again a rather mathematical, but very beautiful topic. Could go from a very applied view to a pretty theoretical one.
This is highly research relevant and has very important applications in machine learning in simulation and other fields.
Some first links: