COPT: Coordinated Optimal Transport on Graphs

Authors: Yihe Dong, Will Sawin

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, COPT outperforms state of the art methods in graph classification on both synthetic and real datasets.
Researcher Affiliation Collaboration Yihe Dong Microsoft Will Sawin Department of Mathematics Columbia University
Pseudocode Yes Algorithm 1 COPT graph sketching and graph distance
Open Source Code No The paper does not provide concrete access to source code (e.g., a specific repository link or an explicit code release statement) for the methodology described.
Open Datasets Yes This is done on four benchmark datasets over diverse domains: Proteins [6], BZR_MD [26], MSRC_9 [38], and Enzymes [45].
Dataset Splits Yes The SVM is trained with parameters found using 3-fold cross validation on the training set, using a fast approximation of the multiscale Laplacian kernel (using the Nyström Method [53]).
Hardware Specification Yes Our implemention uses Py Torch and one P100 GPU, on a 2.60GHz six-core Intel CPU machine.
Software Dependencies No The paper mentions "Py Torch" but does not specify its version number or any other software dependencies with version numbers.
Experiment Setup Yes Gradient descent is used to minimize the analytic formulation Eq (5), where L Y and P are updated at each step, with the Adam optimizer [23] with a multistep learning rate scheduler that reduces the learning rate multiplicatively at regular intervals.