Mutli-Armed Bandits with Network Interference

Authors: Abhineet Agarwal, Anish Agarwal, Lorenzo Masoero, Justin Whitehouse

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we corroborate our theoretical findings via numerical simulations. [...] 6 Simulations
Researcher Affiliation Collaboration Abhineet Agarwal Department of Statistics UC Berkeley aa3797@berkeley.edu Anish Agarwal Department of IEOR Columbia University aa5194@columbia.edu Lorenzo Masoero Amazon masoerl@amazon.com Justin Whitehouse Computer Science Department Carnegie Mellon University jwhiteho@andrew.cmu.edu
Pseudocode Yes Algorithm 1 Network Explore-Then-Commit with Known Interference [...] Algorithm 2 Network Explore-Then-Commit with Unknown Interference
Open Source Code Yes Code for our methods and experiments can be found at https://github.com/aagarwal1996/Network MAB.
Open Datasets No The paper describes a 'Data Generating Process' for simulations but does not provide public access to the generated dataset. They create the data dynamically for their experiments.
Dataset Splits Yes For our Algorithms, we choose all hyper-parameters via 3-fold CV
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., CPU/GPU models, memory).
Software Dependencies No The paper mentions using 'the scikit-learn implementation of the Lasso' but does not specify its version number or other software dependencies with versions.
Experiment Setup Yes Data Generating Process. We generate interference patterns with varying number of units N {5, . . . , 10}, and A = 2. For each N, we use s = 4. We generate rewards rn = θn, χ(a) , where the non-zero elements of θn (i.e., θn,S for S Bn) are drawn uniform from [0, 1]. We normalize rewards so that they are contained in [0, 1], and add 1 sub-gaussian noise to sampled rewards. [...] For our Algorithms, we choose all hyper-parameters via 3-fold CV, and use the scikit-learn implementation of the Lasso. [...] Algorithm 2 run with λ = 4 p E 1 log(2AN) + δ where E := (TAs)2/3 log Nδ + N log(A) 1/3