Mitigating Emergent Robustness Degradation while Scaling Graph Learning

Authors: Xiangchi Yuan, Chunhui Zhang, Yijun Tian, Yanfang Ye, Chuxu Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments have been conducted to demonstrate the anti-degraded robustness and scalability of our method, as compared to popular graph adversarial learning methods, under diverse attack intensities and various datasets of different sizes.
Researcher Affiliation Academia 1Brandeis University, {xiangchiyuan,chuxuzhang}@brandeis.edu 2Dartmouth College, {chunhui.zhang.gr}@dartmouth.edu 3University of Notre Dame, {yijun.tian,yye7}@nd.edu
Pseudocode No The paper describes the method using figures and text descriptions, but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes The code can be accessed through https://github.com/chunhuizng/emergent-degradation.
Open Datasets Yes We evaluate the robustness and scalability of our proposed DRAGON framework using the Graph Robustness Benchmark (GRB) dataset (Zheng et al., 2021), which includes graphs of varying scales, such as grb-cora (small-scale), grb-citeseer (small-scale), grb-flickr (medium-scale), grb-reddit (large-scale), and grb-aminer (large-scale).
Dataset Splits Yes Additionally, we adhere to the GRB benchmark s data splitting protocol, with 60% of the graph data as the training set, 10% as the validation set, and 30% as the test set for each benchmark dataset.
Hardware Specification Yes All experiments are performed on an NVIDIA V100 GPU with 32 GB of memory.
Software Dependencies No The paper refers to using GCN as a surrogate model and adhering to GRB benchmark configurations for baselines, but it does not specify software dependencies with version numbers (e.g., Python, PyTorch/TensorFlow versions, or specific library versions).
Experiment Setup Yes The hyperparameters of DPMo E are given in Table 5, including hyperparameters of adversarial training listed in Table 3 Specifically, for Cora, we set the total number of experts to N = 10 and the number of activated experts to k = 2. For other datasets, by default, we set the total number of experts to N = 4 and the number of activated experts to k = 1. Additionally, hyperparameters of DMGAN are shown in Table 10. The mask rate is 0.7, the walks per node is 1, and the walk length is 3.