Dueling Bandits with Adversarial Sleeping

Authors: Aadirupa Saha, Pierre Gaillard

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our results are corroborated empirically.
Researcher Affiliation Collaboration Microsoft Research, New York, US; aasa@microsoft.com. Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, 38000 Grenoble, France. pierre.gaillard@inria.fr
Pseudocode Yes Algorithm 1 Sl DB-UCB
Open Source Code No The paper does not provide any explicit statements about releasing source code, nor does it include a link to a code repository.
Open Datasets No We use the following three different utility based Plackett Luce(θ) preference models (see Sec. 2) that ensures a total-ordering. We now construct three types of problem instances 1. Easy 2. Medium 3. Hard, for any given K, such that items with their respective θ parameters are assigned as follows: 1. Easy: θ(1 : K/2 ) = 1, θ( K/2 + 1 : K) = 0.5. 2. Medium: θ(1 : K/3 ) = 1, θ( K/3 + 1 : 2K/3 ) = 0.7, θ( 2K/3 + 1 : K) = 0.4. 3. Hard: θ(i) = 1 (i 1)/K, i [K].
Dataset Splits No The paper describes an online learning framework where data is generated sequentially, and does not discuss traditional training, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments.
Software Dependencies No The paper does not provide specific software dependencies or version numbers for the experimental setup.
Experiment Setup Yes In every experiment, we set the learning parameters α = 0.51, δ = 1/T for Sl DB-UCB (Alg. 1) and as per Thm. 6 for Sl DB-ED (Alg. 2). All results are averaged over 50 runs.