Bot, Daniël M.Huo, PeiyangArleo, AlessioPaulovich, FernandoAerts, JanKucher, KostiantynDiehl, AlexandraGillmann, Christina2024-05-212024-05-212024978-3-03868-258-5https://doi.org/10.2312/evp.20241088https://diglib.eg.org/handle/10.2312/evp20241088Recent dimensionality reduction algorithms operate on a manifold assumption and expect data to be uniformly sampled from that underlying manifold. While some algorithms attempt to be robust for non-uniform sampling, their reliance on k-nearest neighbours to approximate manifolds limits how well they can span sampling gaps without introducing shortcuts. We present a minimum-spanning-tree-based manifold approximation approach that overcomes this problem and demonstrate it crosses sampling-gaps without introducing shortcuts while creating networks with few edges. A python package implementing our algorithm is available at https://github.com/vda-lab/multi_mst.Attribution 4.0 International LicenseCCS Concepts: Computing methodologies → Dimensionality reduction and manifold learningComputing methodologies → Dimensionality reduction and manifold learning10.2312/evp.202410883 pages