Distributed Multi-Player Bandits - a Game of Thrones Approach

Authors: Ilai Bistritz, Amir Leshem

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We simulated a multi-armed bandit game with {µn,i} that are chosen independently and uniformly at random in [0.05, 0.95]. The rewards are generated as rn,i (t) = µn,i + zn,i (t) where {zn,i (t)} are independent and uniformly distributed on [ 0.05, 0.05] for each n, i. In Fig. 2, we present the sample mean of the accumulated sum of utilities PN n=1 1 t Pt τ=1 un (a (τ)) as a function of time t, averaged over 100 experiments.
Researcher Affiliation Academia Ilai Bistritz Stanford University bistritz@stanford.edu Amir Leshem Bar Ilan University Amir.Leshem@biu.ac.il
Pseudocode Yes Algorithm 1 Game of Thrones Algorithm and Algorithm 2 Game of Thrones Dynamics
Open Source Code No The paper does not provide any explicit statements about the release of open-source code, nor does it include a link to a code repository.
Open Datasets No The paper describes a simulated environment where data is generated for experiments rather than using or providing a publicly available dataset. It states: 'We simulated a multi-armed bandit game with {µn,i} that are chosen independently and uniformly at random in [0.05, 0.95]. The rewards are generated as rn,i (t) = µn,i + zn,i (t) where {zn,i (t)} are independent and uniformly distributed on [ 0.05, 0.05] for each n, i.'
Dataset Splits No The paper does not mention train/validation/test dataset splits. It describes an online learning framework with 'exploration', 'Game of Thrones (Go T)', and 'exploitation' phases within its simulation.
Hardware Specification No The paper describes its simulations ('We simulated a multi-armed bandit game...') but does not provide any specific details about the hardware (e.g., CPU, GPU, memory) used to conduct these simulations.
Software Dependencies No The paper describes numerical simulations but does not specify any software dependencies or their version numbers (e.g., programming languages, libraries, or solvers).
Experiment Setup Yes Hence we choose c1 = 1000, c2 = c3 = 6000. We use ρ = 1/2 in the simulations we present, since the performance is very similar for ρ values not too close to zero or one. We use c = N, that gives the highest possible escape probability of εc from a content state.