Distributed Contextual Linear Bandits with Minimax Optimal Communication Cost

Authors: Sanae Amani, Tor Lattimore, András György, Lin Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We complement our theoretical results with numerical simulations in Section 5. 5. Experiments In this section, we present numerical simulations to confirm our theoretical findings.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of California, Los Angeles. 2Deep Mind, London.
Pseudocode Yes Algorithm 1 Dis BE-LUCB for agent i, Algorithm 2 Dec BE-LUCB for agent i, Algorithm 3 Comm for Agent i, Algorithm 4 Exp Pol, Algorithm 5 Core Identification (Algorithm 4 in (Ruan et al., 2021)), Algorithm 6 Mixed Soft Max.
Open Source Code No No explicit statement or link providing concrete access to the source code for the methodology described in this paper was found.
Open Datasets No The decision set distribution D is chosen to be uniform over { X1, X2, . . . , X100}, where each Xi is a set of K vectors drawn from N(0, Id) and then normalized to unit norm. This is a description of synthetic data generation, not a publicly available dataset with concrete access information.
Dataset Splits No No specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning was found.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment.
Experiment Setup Yes In this section, we present numerical simulations to confirm our theoretical findings. We evaluate the performance of Dis BE-LUCB on synthetic data and compare it to that of Dis Lin UCB proposed by Wang et al. (2019). The results shown in Figure 1 depict averages over 20 realizations, for which we have chosen K = 20, δ = 0.01 and T = 100000. For each realization, the parameter θ is drawn from N(0, Id) and then normalized to unit norm and noise variables are zero-mean Gaussian random variables with variance 0.01. The decision set distribution D is chosen to be uniform over { X1, X2, . . . , X100}, where each Xi is a set of K vectors drawn from N(0, Id) and then normalized to unit norm.