Networked Restless Bandits with Positive Externalities

Authors: Christine Herlihy, John P. Dickerson

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results demonstrate that GRETA outperforms comparison policies across a range of hyperparameter values and graph topologies.
Researcher Affiliation Academia Christine Herlihy, John P. Dickerson Department of Computer Science University of Maryland, College Park College Park, MD, USA cherlihy@umd.edu, johnd@umd.edu
Pseudocode Yes Algorithm 1: Compute Whittle indices for V A \ {0} (...) Algorithm 2: GRETA: graph-aware, Whittle-based heuristic (...) Algorithm 3: Compute cost to pull u and message v (...) Algorithm 4: Cumulative subsidy of max pull-message set (...) Algorithm 5: Compute edge index values
Open Source Code Yes Code and appendices are available at https://github.com/crherlihy/networked_restless_bandits.
Open Datasets No The paper generates synthetic data: 'We consider a synthetic cohort of n = 100 restless arms whose transition matrices are randomly generated in such a way so as to satisfy the structural constraints introduced in Section 2. We use a stochastic block model (SBM) generator with pin = 0.2 and pout = 0.05, and consider both the random and by cluster options for φ.' It does not use a publicly available, pre-existing dataset with concrete access information.
Dataset Splits No The paper performs experiments on synthetic data but does not specify train/validation/test splits with percentages, sample counts, or references to predefined splits for reproducibility.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. It only mentions general experimental setup.
Software Dependencies No The paper mentions using a 'K-MEANS algorithm' and cites 'Scikit-learn: Machine Learning in Python' (Pedregosa et al. 2011), but it does not provide specific version numbers for these or any other software components used in the experiments.
Experiment Setup Yes Figure 1 reports results for a synthetic cohort of 8 arms embedded in a fully connected graph (i.e., pin = pout = 1.0). We let T = 120, ψ = 0.5, and report unnormalized Eπ[R], along with margins of error for 95% confidence intervals computed over 50 simulation seeds for values of B {1, 1.5, 2, 2.5, 3}. (...) We let T = 120, B = 10, and ψ = 0.5. (...) We hold message cost fixed at ψ = 0.5, let pin = 0.25, pout = 0.05, and consider values of B {5%, 10%, 15%} of n. (...) Here, we hold the budget fixed at 6, let pin = 0.25, pout = 0.05, and consider values of ψ {0.0, 0.25, 0.5, 0.75, 0.9}.