February 02, 2023
3.00 - 5.00 pm
Auditoire DOYEN 21
Doctoral Commission in Economics and Management Sciences
The Rector of UCLouvain and the Doctoral Commission in Economics and Management Sciences announce that Mr. Pierre Leleux will publicly support his dissertation, for the title of Doctor in Economics and Management Sciences of UCLouvain.
Entropy-regularized biased random walks on a graph: Applications to network data analysis
Network (or graph) data analysis finds applications in many contexts, including biology, finance, marketing, and physics, to name a few. With the recent thriving of internet and social networks, these graph mining techniques have grown in popularity as an important tool to process, analyze, and make predictions from the quickly increasing amount of network data. This thesis mostly focuses on the randomized shortest path (RSP) framework, which defines a family of dissimilarities between nodes of a network that interpolates between the least cost path distance (optimal behavior) and the commute-cost distance (purely random behavior). This thesis therefore builds upon and extends this framework by proposing, among others, new dissimilarity measures, a design of biased random walk, as well as fast solutions to approximate the RSP for large graphs. It also explores various applications like node clustering, stochastic Markov decision processes, connectivity modeling in ecological landscapes, and collaborative filtering.
For the good organisation of this event, please inform Pierre Leleux of your presence.