Below is a list of scheduled presentations of bachelor's and master's theses by computer science students, sorted chronologically by the date of presentation.
Presentation of the Bachelor's Colloquiums by Mr. Gero Plettenberg:
"Experimental validation for kinematic control of a soft wearable exosuit"
Advisor: Prof. Dr. Lorenzo Masia
The presentation takes place on Thursday, December 1, 2022, at 3pm in room 443, 4. OG, ZITI INF 368. You can also participate in the presentation online via the link https://uni-heidelberg.webex.com/meet/lorenzo.masia.
Presentation of the master's thesis by Mr. Roman Erhardt:
"Engineering Algorithms for the Weighted Maximum Clique Problem"
Advisor: Prof. Dr. Christian Schulz
The presentation takes place on December 5, 2022, at 10am online via HeiCONF. In case you are interested in attending the presentation, please contact Christian Schulz at email@example.com.
The maximum weighted clique problem (MWC) is a well-known problem in graph theory with many applications. In this work, both exact and heuristic algorithms, which interleave successful techniques from related work with novel graph reduction rules are proposed. While graph reductions based on upper bounds have been used for MWC in the past, we present reduction rules, that make use of local graph structures in order to identify and remove vertices and edges without reducing the optimal solution. A set of exact reduction rules is employed in an exact algorithm called MWCRedu, while heuristic reduction techniques based on machine learning models such as MLP, Deepset and GNN are explored in the heuristic framework MWCPeel. Experiments on a broad range of graphs show, that MWCRedu outperforms the current state-of-the-art exact solver TSM-MWC  for most inputs. Specifically for naturally weighted, medium-sized street network graphs and random hyperbolic graphs, which are considered to model real-world graphs well, MWCRedu is faster by orders of magnitudes. The heuristic solver MWCPeel also outperforms its competitors FastWCLq  and SCCWalk4l  on these instances, but is slightly less effective on extremely dense or large instances. It is planned to publish the result.