The project Online Algorithms with Machine Learning Predictors is funded by the DKK 770,400 from the Independent Research Fund Denmark, Natural Sciences, running from January 1, 2021 through June 30, 2023.
The project is carried out at the Department of Mathematics and Computer Science (IMADA) at the University of Southern Denmark.

Participants

Description

In online problems requests arrive online and irrevocable decisions must be made without knowledge of future requests. For resource optimization problems, online algorithms provide guarantees by focusing on worst-case scenarios with less attention for normal cases. In contrast, machine learning focuses on the frequently occurring cases, but usually with no guarantees, which is problematic for planning and safety. We aim at combining the best features of the two, using machine learning as an untrusted advisory component in an online algorithmic solution, following the advice of the component when this does not jeopardize necessary guarantees. Thus, the relevant techniques will be developed in the area of online algorithms, where SDU's online algorithms group is a leading research group with respect to performance measures. Our recent focus on advice complexity, which is a related component-based online algorithms approach, makes us highly qualified for undertaking this endeavor.

Environment

The project is carried out at the Department of Mathematics and Computer Science (IMADA) at the University of Southern Denmark. The participants are member of the Online Algorithms group and the Algorithms group, and are all associated with the Research Training Program in Computer Science as Ph.D. advisors.

Activities

Almost all activities on this grant are research related traveling to conferences, meetings by invitation, research collaboration, and hosting guests.

Publications

Here we will list project publications when the project starts. Slightly older publications can be found on the page for our previous project. Complete lists for each participant can be found via our individual home pages or via dblp, the standard search engine for Computer Science publications. We link to the official site for published papers using the doi (digital object identifier) of the papers. For open access versions, we refer to each author's own home page.

 


Data protection at SDUDatabeskyttelse på SDU