No-Regret and Incentive-Compatible Online Learning

Authors: Rupert Freeman, David Pennock, Chara Podimata, Jennifer Wortman Vaughan

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In experiments on datasets from Five Thirty Eight, our algorithms have regret comparable to classic no-regret algorithms, which are not incentive-compatible.
Researcher Affiliation Collaboration 1Darden School of Business, University of Virginia. 2DIMACS, Rutgers University. 3Harvard University. 4Microsoft Research NYC.
Pseudocode Yes Algorithm 1 WSU-UX with parameters η and γ such that 0 < η, γ < 1/2 and ηK/γ 1/2.
Open Source Code Yes Our code and the datasets we use are publicly available online. Code: https://github.com/charapod/noregr-and-ic.
Open Datasets Yes Our code and the datasets we use are publicly available online. Datasets: https://github.com/fivethirtyeight/nfl-elo-game
Dataset Splits No The paper describes the number of games and how forecasters were sampled for different values of K, but it does not specify explicit training, validation, or test dataset splits in terms of percentages or counts.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as GPU or CPU models.
Software Dependencies No The paper states that code is available but does not specify any software dependencies or their version numbers, such as programming languages, libraries, or frameworks.
Experiment Setup Yes For WSU, 'for step size η = p ln(K)/T yields regret R 2 T ln K'. For WSU-UX, 'For T K ln K and parameters η = ln K 4K1/2T 2/3 and γ = K ln K / 4T 1/3, WSU-UX is incentive compatible and yields regret R 2(4T)2/3(K ln K)1/3.'