Generalized Sobolev Transport for Probability Measures on a Graph

Authors: Tam Le, Truyen Nguyen, Kenji Fukumizu

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

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically illustrate that GST is several-order faster than the OW. Moreover, we provide preliminary evidences on the advantages of GST for document classification and for several tasks in topological data analysis.
Researcher Affiliation Academia 1Department of Advanced Data Science, The Institute of Statistical Mathematics (ISM), Tokyo, Japan 2RIKEN AIP, Tokyo, Japan 3The University of Akron, Ohio, US.
Pseudocode No The paper does not include pseudocode or a clearly labeled algorithm block for its own method.
Open Source Code Yes We have also released code for our proposed approach.2 The code repository is on https://github.com/ lttam/Generalized-Sobolev-Transport
Open Datasets Yes We consider 4 traditional document datasets: TWITTER, RECIPE, CLASSIC, AMAZON. We consider two tasks: orbit recognition on the synthesis Orbit dataset (Adams et al., 2017), and object shape classification on MPEG7 dataset (Latecki et al., 2000)
Dataset Splits Yes For each dataset, we randomly split it into 70%/30% for training and test with 10 repeats. We typically choose hyperparameters via cross validation. For validation, we further randomly split the training set into 70%/30% for validation-training and validation with 10 repeats to choose hyper-paramters in our simulations.
Hardware Specification No The paper does not explicitly describe the specific hardware used (e.g., GPU/CPU models, memory specifications) for running its experiments.
Software Dependencies No The paper mentions using "fmincon Trust Region Reflective solver in MATLAB" and "Libsvm" but does not provide specific version numbers for these software components.
Experiment Setup Yes For kernel hyperparameter, we choose 1/ t from {qs, 2qs, 5qs} with s = 10, 20, . . . , 90 where qs is the s% quantile of a subset of distances observed on a training set. For SVM regularization hyperparameter, we choose it from {0.01, 0.1, 1, 10}. For the root node z0 in graph G, we choose it from a random 10-root-node subset of V in G.