Bayesian Optimization of Functions over Node Subsets in Graphs

Authors: Huidong Liang, Xingchen Wan, Xiaowen Dong

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments under both synthetic and real-world setups demonstrate the effectiveness of the proposed BO framework on various types of graphs and optimization tasks, where its behavior is analyzed in detail with ablation studies.
Researcher Affiliation Collaboration Huidong Liang Xingchen Wan Xiaowen Dong Department of Engineering Science, University of Oxford {huidong.liang,xiaowen.dong}@eng.ox.ac.uk, xingchenw@google.com Now at Google.
Pseudocode Yes Algorithm 1 Recursive combo-subgraph sampling
Open Source Code Yes The experiment code can be found at github.com/Leon Research/Graph Com BO.
Open Datasets Yes We conduct comprehensive experiments on four synthetic problems and five real-world tasks to validate our proposed framework...Table 1: Summary statistics of the underlying graphs used in our experiments...The contact network [47] used in this experiment is collected by proximity sensors from a primary school in France, which contains 236 nodes and 5, 899 edges.
Dataset Splits Yes After standardization, we use 25% combo-nodes as the training set to fit the models and validate their performance on the rest 75% combo-nodes with Spearman s rank-based correlation coefficient ρ.
Hardware Specification Yes All the experiments are conducted on a computing cluster of 96 Intel-Xeon@2.30GHz CPU cores with 250 GB working memory.
Software Dependencies No The paper mentions software components like 'diffusion kernel' and 'Expected Improvement' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Specifically, we query 300 times and repeat 20 times with different random seeds for each task, in which the mean and standard error of the cumulative optima are reported for all methods. For simplicity, we use a diffusion kernel [40] with automatic relevance determination and adopt Expected Improvement [26] as the acquisition function to investigate subset sizes of k r4, 8, 16, 32s, where we also fix Q 4, 000 and failtol 30 across all experiments. In addition, we also initialize the algorithm with 10 queries using simple random walks when k ě 16.