Mirror and Preconditioned Gradient Descent in Wasserstein Space

Authors: Clément Bonet, Théo Uscidda, Adam David, Pierre-Cyril Aubin-Frankowski, Anna Korba

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We illustrate the advantages of adapting the geometry induced by the regularizer on ill-conditioned optimization tasks, and showcase the improvement of choosing different discrepancies and geometries in a computational biology task of aligning single-cells.
Researcher Affiliation Academia Clément Bonet CREST, ENSAE, IP Paris clement.bonet@ensae.fr Théo Uscidda CREST, ENSAE, IP Paris theo.uscidda@ensae.fr Adam David Institute of Mathematics Technische Universität Berlin david@math.tu-berlin.de Pierre-Cyril Aubin-Frankowski TU Wien pierre-cyril.aubin@tuwien.ac.at Anna Korba CREST, ENSAE, IP Paris anna.korba@ensae.fr
Pseudocode No The paper describes algorithms using mathematical equations and text, but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes The code is available at https://github.com/clbonet/Mirror_and_Preconditioned_Gradient_ Descent_in_Wasserstein_Space.
Open Datasets Yes We focus on the datasets used in [21], consisting of cell lines analyzed using (i) 4i [58], and (ii) sc RNA sequencing [118].
Dataset Splits No The paper mentions '40% of unseen (test) target cells' for evaluation, but no specific training/validation splits (percentages or counts) are detailed for reproduction.
Hardware Specification Yes These experiments were run on a personal Laptop with a CPU Intel Core i5-9300H.
Software Dependencies No The paper mentions 'OTT-JAX' but does not specify a version number. No other specific software dependencies with version numbers are listed.
Experiment Setup Yes We set the step size τ = 1 for all the experiments. Then, we tune the parameter a very simply: for a given metric D and a profiling technology, we pick a random treatment and select a {1.25, 1.5, 1.75} by grid search...