Computing a sum of terms can be prohibitively demanding when the terms are numerous and specified only implicitly - the case in various applications in mathematics, physics, and statistics. The project "Adaptive Techniques for Approximate Counting" seeks novel methods to solve such counting problems well in practice. We allow the computation time vary in unpredictable ways from one problem instance to another. We also allow the solution deviate from the exact value, but require the method to yield an upper bound on the deviation.
Our methodology combines viewpoints of two research areas, algorithm theory and artificial intelligence. We focus on a few problems from machine learning and computational geometry, for which the state of the art is insufficient.
The four-year project will be conducted at the Department of Computer Science of University of Helsinki, by a team of a few researchers. The project includes significant collaboration with other teams in Finland and abroad.