Public Thesis Defense of Pierre LAMBERT - ICTEAM
sst |
Reliable and interactive neighbour embedding for dimensionality reduction, visualisation, and machine learning
Thursday, February 26, 2026 - 4:15pm - Auditorium BARB94 -Place Sainte-Barbe, 1348 Louvain-la-Neuve
Dimensionality reduction enables the visualisation of high dimensional data by representing it in 2- or 3- dimensional spaces. Neighbour embeddings are a family of dimensionality reduction methods that excel at extracting fine grained details in potentially very high dimensional data.
This thesis addresses some of the shortcomings of neighbour embedding by accelerating them and making them suitable for continual learning, allowing the interactive visualisation of large datasets. The last stretch of the thesis attempts to isolate the properties that allow neighbour embeddings to so successfully discriminate between high dimensional structures, and a promising new algorithm for label propagation is introduced, building on these properties..
Jury members
Prof. John Lee (UCLouvain)(Promoteur)
Prof. Christophe Craeye (UCLouvain) (Président)
Prof. Michel Verleysen (UCLouvain) (Secrétaire)
Prof. Bruno Dumas (UNamur, Belgium)
Prof. Jaakko Peltonen (Tampere university, Finland)
Prof. Dmitry Kobak (Tübingen university, Germany)