Controlling Graph Dynamics with Reinforcement Learning and Graph Neural Networks

Authors: Eli Meirom, Haggai Maron, Shie Mannor, Gal Chechik

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

Reproducibility Variable Result LLM Response
Research Type Experimental We test our approach on two very different problems, Influence Maximization and Epidemic Test Prioritization, and show that our approach outperforms state-of-the-art methods, often significantly. [...] 5. Experiments We evaluated our approach in two tasks: (1) Epidemic test prioritization, and (2) Dynamic influence maximization.
Researcher Affiliation Industry Eli A. Meirom 1 Haggai Maron 1 Shie Mannor 1 Gal Chechik 1 1NVIDIA Research, Israel. Correspondence to: Eli Meirom <emeirom@nvidia.com>.
Pseudocode No No pseudocode or clearly labeled algorithm block was found in the paper.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Real-World Datasets. We tested our algorithm and baselines on graphs of different sizes and sources, ranging from 5K to over 100K nodes. (1) CA-Gr Qc A A research collaboration network (Rossi & Ahmed, 2015). (2) Montreal, based on Wi Fi hotspot tracing(Hoen et al., 2015). (3) Portland: a compartment-based synthetic network (Wells et al., 2013; Eubank et al., 2004). (4) Email: An email network (Leskovec et al., 2007) (5) GEMSEC-RO: (Rozemberczki et al., 2019), friendship relations in the Deezer music service.
Dataset Splits No The paper states 'Algorithms were trained on randomly generated PA networks with 1000 nodes,' and evaluates on various real-world and synthetic datasets, but does not explicitly provide specific training/validation/test split percentages or sample counts for these datasets.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software components like Proximal Policy Optimization (PPO) and Graph Neural Networks (GNNs), and implicitly PyTorch through a citation, but does not provide specific version numbers for these software dependencies (e.g., 'PyTorch 1.9' or 'CUDA 11.1').
Experiment Setup No The paper mentions some aspects of the training procedure like 'Each experiment was performed with at least three random seeds' and the sampling mechanism including a parameter 'ϵ', but it does not provide comprehensive experimental setup details such as learning rate, batch size, number of epochs, or optimizer settings.