(Almost) Free Incentivized Exploration from Decentralized Learning Agents

Authors: Chengshuai Shi, Haifeng Xu, Wei Xiong, Cong Shen

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results are provided to complement the theoretical analysis. Numerical experiments have been carried out to evaluate OTI. All the results are averaged over 100 runs of horizon T = 105 and the agents perform the α-UCB algorithm specified in Section 5.1 with α = 2.
Researcher Affiliation Academia Chengshuai Shi University of Virginia cs7ync@virginia.edu Haifeng Xu University of Virginia hx4ad@virginia.edu Wei Xiong The Hong Kong University of Science and Technology wxiongae@connect.ust.hk Cong Shen University of Virginia cong@virginia.edu
Pseudocode Yes Algorithm 1 OTI: Principal
Open Source Code No The paper does not provide any links to open-source code or explicitly state that code for the methodology is released.
Open Datasets No The paper describes experiments run on simulated environments (e.g., 'toy example of M = 2 agents and K = 3 arms', 'random local instances with 30 arms are generated') rather than a publicly available dataset with concrete access information.
Dataset Splits No The paper discusses simulation parameters but does not specify explicit training, validation, or test dataset splits in the conventional sense for a supervised learning problem. The environment is simulated for multi-armed bandit scenarios.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions agents running the 'α-UCB algorithm', but does not specify any software names with version numbers used for implementation (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes All the results are averaged over 100 runs of horizon T = 105 and the agents perform the α-UCB algorithm specified in Section 5.1 with α = 2. First, with a toy example of M = 2 agents and K = 3 arms, the ineffectiveness of not incentivizing is illustrated. Under different M, random local instances with 30 arms are generated to compose global instances with min [4.5, 5.5] 10 3.