CO-Optimal Transport

Authors: Vayer Titouan, Ievgen Redko, Rémi Flamary, Nicolas Courty

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide evidence of its versatility in machine learning by putting forward two applications in HDA and co-clustering where our approach achieves state-of-the-art results. The rest of this paper is organized as follows. ... In Section 4, we present an experimental study providing highly competitive results in HDA and co-clustering compared to several baselines.
Researcher Affiliation Academia Ievgen Redko Univ Lyon, UJM-Saint-Etienne, CNRS, UMR 5516 F-42023, Saint-Etienne ievgen.redko@univ-st-etienne.fr Titouan Vayer Univ. Bretagne-Sud, CNRS, IRISA F-56000 Vannes titouan.vayer@irisa.fr R emi Flamary Ecole Polytechnique, CMAP, UMR 7641 F-91120 Palaiseau remi.flamary@polytechnique.edu Nicolas Courty Univ. Bretagne-Sud, CNRS, IRISA F-56000 Vannes nicolas.courty@irisa.fr
Pseudocode Yes Algorithm 1 BCD for COOT 1: πs (0) ww T , πv (0) vv T , k 0 2: while k < max It and err > 0 do 3: πv (k) OT(v, v , L(X, X ) πs (k 1)) // OT problem on the samples 4: πs (k) OT(w, w , L(X, X ) πv (k 1)) // OT problem on the features 5: err ||πv (k 1) πv (k)||F 6: k k + 1
Open Source Code No The paper does not contain an explicit statement or link providing access to the authors' open-source code for the methodology described.
Open Datasets Yes We evaluate COOT on Amazon (A), Caltech-256 (C) and Webcam (W) domains from Caltech-Office dataset [43] ... benchmark MOVIELENS-100K3 data set that provides 100,000 user-movie ratings, on a scale of one to five, collected from 943 users on 1682 movies. (footnote 3: https://grouplens.org/datasets/movielens/100k/)
Dataset Splits No The paper discusses 'semi-supervised HDA, where one has access to a small number nt of labelled samples per class in the target domain' and mentions 'We selected 20 samples per class to form the learning sets'. However, it does not provide explicit training, validation, and test dataset splits with percentages, counts, or references to standard splits that would allow for direct reproduction of the data partitioning into these three subsets.
Hardware Specification Yes We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan X GPU used for this research.
Software Dependencies No The paper mentions software components like 'k-nn classifier' but does not specify any version numbers for libraries, frameworks, or other ancillary software dependencies, which are necessary for reproducible descriptions.
Experiment Setup Yes For EGW and SGW, the entropic regularization term was set to 0.1, and the two other regularization hyperparameters for the semi-supervised case to λ = 10−5 and γ = 10−2 as done in [16? ]. We use COOT with entropic regularization on the feature mapping, with parameter ϵ2 = 1 in all experiments.