Mirror Sinkhorn: Fast Online Optimization on Transport Polytopes

Authors: Marin Ballu, Quentin Berthet

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide theoretical guarantees for convex objectives and experimental results showcasing it effectiveness on both synthetic and real-world data.
Researcher Affiliation Collaboration 1University of Cambridge, UK 2Google Deep Mind, Paris, France. Correspondence to: Quentin Berthet <qberthet@google.com>.
Pseudocode Yes Algorithm 1 Mirror Sinkhorn; Algorithm 2 Rounding algorithm (Altschuler et al., 2017)
Open Source Code No The paper does not provide any explicit statement or link to open-source code for the described methodology.
Open Datasets Yes We also include an illustration of our method on two datasets used in (Altschuler et al., 2017), following their experimental setup: we use as instances of OT random pairs from MNIST (10 in total), and simulated SQUARES data... The SNARE-seq data (Chen et al., 2019) consists of 1047 vectors in dimension 10 and 19 respectively.
Dataset Splits No The paper mentions using datasets like MNIST and SNARE-seq but does not provide specific train/validation/test split percentages, absolute sample counts for splits, or references to predefined splits needed for reproduction.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes We minimize this functional by taking λ = 3, with a k-NN graph taken for k = 5. We recall that in this case, n = 1, 047. We are applying a step-size regime proportional to 1/(t + 1), for T = 105 steps.