Congested Bandits: Optimal Routing via Short-term Resets

Authors: Pranjal Awasthi, Kush Bhatia, Sreenivas Gollapudi, Kostas Kollias

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental From an experimental standpoint, we corroborate the no-regret properties of our algorithms via a simulation study. ... Finally, using simulations, we perform an empirical evaluation of the effectiveness of our proposed algorithms. ... In this section, we evaluate both our proposed algorithms, CARMAB and CARCB, in the congested bandit framework and exhibit their no-regret properties. ... Figure 2. (Left) No-regret property of CARMAB. ... (Right) No-regret property of CARCB.
Researcher Affiliation Collaboration 1Google Research 2UC Berkeley. Correspondence to: Kush Bhatia <kushbhatia@berkeley.edu>.
Pseudocode Yes Algorithm 1: CARMAB: Congestion Aware Routing via Multi-Armed Bandits. ... Algorithm 2: CARCB: Congested linear contextual bandits with known contexts. ... Algorithm 3: CARMAB-ST: Congested multi-armed bandits.
Open Source Code No The paper does not provide any links to open-source code or explicitly state that the code for their methodology is publicly available.
Open Datasets No The paper describes how it generates synthetic data for its simulations ('We generate K arms and assign a base reward of ˆrj (0, 1) to each j [K]...'), but it does not use or provide access to a publicly available or open dataset.
Dataset Splits No The paper describes simulation experiments but does not specify any training, validation, or test dataset splits in terms of explicit percentages, sample counts, or predefined splits from existing datasets.
Hardware Specification No The paper mentions running 'simulations' but does not provide any specific details about the hardware (CPU, GPU models, memory, etc.) used for these simulations.
Software Dependencies No The paper mentions algorithms used (e.g., Karp's algorithm, UCRL2) but does not provide specific software names or version numbers for their implementation (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes We set parameter δ of algorithm CARMAB to 0.1. In terms of distributions, we draw ˆrj uniformly in (0, 1) and ϵt,j from N(0, 0.1). ... For each arm i [K] and each time step t [T], we draw a random context. A context xa,t is a vector of 10 numbers which are drawn uniformly in (0, 1) and then normalized so that the Euclidean norm of the vector is unit. We also draw the true parameter θ in the same way.