Collective Robustness Certificates: Exploiting Interdependence in Graph Neural Networks

Authors: Jan Schuchardt, Aleksandar Bojchevski, Johannes Gasteiger, Stephan Günnemann

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on different datasets and with different base certificates. We show that incorporating locality alone is sufficient to obtain significantly better results. Our proposed certificate: ... 5 EXPERIMENTAL EVALUATION Experimental setup. We evaluate the proposed approach by certifiing node classifiers on multiple graphs and with different base certificates.
Researcher Affiliation Academia Jan Schuchardt, Aleksandar Bojchevski, Johannes Gasteiger & Stephan G unnemann Technical University of Munich, Germany {jan.schuchardt,bojchevs,j.gasteiger,guennemann}@in.tum.de
Pseudocode Yes Algorithm 1: Determining the pareto front of base certificates
Open Source Code Yes The code is publicly available under https://www.daml.in.tum.de/ collective-robustness/. We also uploaded the implementation as supplementary material.
Open Datasets Yes Datasets, models and base certificates. We train and certify models on the following datasets: Cora-ML (Mc Callum et al. (2000); Bojchevski & G unnemann (2018); N = 2810, 7981 edges, 7 classes), Citeseer (Sen et al. (2008); N = 2110, 3668 edges, 6 classes), Pub Med (Namata et al. (2012); N = 19717, 44324 edges, 3 classes), Reuters-21578 4 (N = 862, 2586 edges, 4 classes) and Web KB (Craven et al. (1998); N = 877, 2631 edges, 5 classes).
Dataset Splits Yes We use 20 nodes per class to construct a train and a validation set. We certify all remaining nodes. We repeat each experiment five times with different random initializations and data splits.
Hardware Specification No The paper mentions
Software Dependencies No The paper mentions specific components like
Experiment Setup Yes F HYPERPARAMETERS Training schedule for smoothed classifiers. Training is performed in a semi-supervised fashion with 20 nodes per class as a train set. Another 20 nodes per class serve as a validation set. Models are trained with Adam (learning rate = 0.001 [0.01 for SMA], β1 = 0.9, β2 = 0.999, ϵ = 10 8, weight decay = 0.001) for 3000 epochs, using the average cross-entropy loss across all training set nodes, with a batch size of 1. We employ early stopping, if the validation loss does not decrease for 50 epochs (300 epochs for SMA). In each epoch, a different graph is sampled from the smoothing distribution.