Doubly Adversarial Federated Bandits

Authors: Jialin Yi, Milan Vojnovic

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

Reproducibility Variable Result LLM Response
Research Type Experimental We provide numerical evaluations of our algorithm to validate our theoretical results and demonstrate its effectiveness on synthetic and real-world datasets.
Researcher Affiliation Academia 1Department of Statistics, London School of Economics and Political Science, London, United Kingdom.
Pseudocode Yes Algorithm 1 FEDEXP3
Open Source Code Yes The code for producing our experimental results is available online in an anonymous Github repository: https://github.com/jialinyi94/doubly-stochasticfederataed-bandit.
Open Datasets Yes We evaluate the two algorithms using the Movie Lens Latest-full dataset which contains 58,000 movies, classified into 20 genres, with 27,000,000 ratings (rating scores in {0.5, 1, . . . , 5}) from 280,000 users.
Dataset Splits No The paper describes the construction of synthetic datasets and the use of the Movie Lens Latest-full dataset, but does not specify explicit training, validation, or test dataset splits (e.g., percentages or sample counts for each split).
Hardware Specification Yes All the experiments are run on a desktop with AMD Ryzen 5 2600 Six-Core Processor and 16GB memory.
Software Dependencies No The paper mentions 'Python', 'Numpy package', and 'Network X package' but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We set T = 3, 000 with N = 36 and K = 20. At each time step t, a subset Ut of N/2 agents are selected from V with replacement. For all activated agents Ut, the loss for arm i is sampled independently from Bernoulli distribution with mean µi = (i − 1)/(K − 1).