Graph Convolution for Semi-Supervised Classification: Improved Linear Separability and Out-of-Distribution Generalization

Authors: Aseem Baranwal, Kimon Fountoulakis, Aukosh Jagannath

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we provide experiments to demonstrate our theoretical results in Section 4. To solve problem (2) we used CVX, a package for specifying and solving convex programs (Grant & Boyd, 2013; Blondel et al., 2008). Throughout the section we set R = d in (2) for all our experiments.
Researcher Affiliation Academia 1David R. Cheriton School of Computer Science, University of Waterloo, Waterloo, Canada 2Department of Statistics and Actuarial Science, Department of Applied Mathematics, University of Waterloo, Waterloo, Canada.
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating the release of open-source code for the described methodology.
Open Datasets Yes We use the popular real data Cora, Pub Med and Wikipedia Network. These data are publicly available and can be downloaded from (Fey & Lenssen, 2019).
Dataset Splits No The paper mentions training and testing but does not explicitly specify a validation set or its split percentages. It refers to a 'semi-supervised setting where only a fraction of the labels are available' but not a validation split.
Hardware Specification No The paper does not specify any particular hardware (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No To solve problem (2) we used CVX, a package for specifying and solving convex programs (Grant & Boyd, 2013; Blondel et al., 2008). While a software package is mentioned, a specific version number for CVX is not provided.
Experiment Setup Yes Throughout the section we set R = d in (2) for all our experiments. For this experiment we train and test on a CSBM with p = 0.5, q = 0.1, d = 60, and n = 400 which is roughly equal to 0.85 d3/2, and each class has 200 nodes. We present results averaged over 10 trials for the training data and 10 trials for the test data.