Double Auctions with Two-sided Bandit Feedback

Authors: Soumya Basu, Abishek Sankararaman

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

Reproducibility Variable Result LLM Response
Research Type Experimental 6 Simulation Study We perform synthetic studies to augment our theoretical guarantees. For a fixed system of N buyers, M sellers, K participants, and gap, the rewards are Bernoulli, with means themselves chosen uniformly at random. We vary the confidence width of the buyers, b, and seller, s, in [ 1, 2]. Next we simulate the performance of the UCB( b) and LCB( s) over 100 independent sample paths with T = 50k. We report the mean, 25% and 75% value of the trajectories. We plot the cumulative regret of the buyers, Rb,i(t), and the sellers, Rs,j(t), the number of matches in the system K(t), and the price difference (p(t) p ). In Figure 2, we have a 8 8 system with K = 5. We see that K(t) converges to 5, where as (p(t) p ) converges to 0. The social regret grows as log(T). The participant and non-participant individual regret of this instance is presented in the appendix in Figure 3.
Researcher Affiliation Collaboration Soumya Basu Google Mountain View basusoumya@google.com Abishek Sankararaman AWS abishek.90@gmail.com 0Work done when AS was affiliated with UC Berkeley.
Pseudocode No The paper describes the bidding strategy and mechanism in text and equations (Equation 2) but does not provide a separate, structured pseudocode block or algorithm listing.
Open Source Code No The paper does not provide any explicit statements about open-sourcing code or links to a code repository.
Open Datasets No The paper uses synthetic data ("rewards are Bernoulli, with means themselves chosen uniformly at random") and does not provide access information for a publicly available or open dataset.
Dataset Splits No The paper describes simulation parameters like "100 independent sample paths with T = 50k" but does not specify training, validation, or test dataset splits.
Hardware Specification No The paper describes the simulation setup (e.g., "100 independent sample paths with T = 50k") but does not provide any specific hardware details like GPU/CPU models or memory used for running the experiments.
Software Dependencies No The paper does not mention any specific software dependencies or libraries with version numbers (e.g., Python, PyTorch, etc.) that were used to implement the algorithms or run the simulations.
Experiment Setup Yes Next we simulate the performance of the UCB( b) and LCB( s) over 100 independent sample paths with T = 50k. We report the mean, 25% and 75% value of the trajectories. We plot the cumulative regret of the buyers, Rb,i(t), and the sellers, Rs,j(t), the number of matches in the system K(t), and the price difference (p(t) p ). In Figure 2, we have a 8 8 system with K = 5, = 0.2, 1 = 4, and 2 = 8. We vary the confidence width of the buyers, b, and seller, s, in [ 1, 2].