Generalized Principal-Agent Problem with a Learning Agent

Authors: Tao Lin, Yiling Chen

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Our work defines a general model of principal-agent problems with a learning agent, answering all questions (1) (3). For (2) and (3), we provide a unified analytical framework to derive tight bounds on the principal s achievable utility against a no-regret or no-swap-regret learning agent in all generalized principalagent problems where the agent does not have private information. Specifically, we explicitly characterize the o(1) difference between the principal s utility and U in terms of the agent s regret. Result 1 (from Theorems 3.1, 4.1, 4.2). Against a no-regret learning agent with regret Reg(T) in T periods, the principal can obtain an average utility of at least U O q T , where U is the principal s optimal utility in the classic model with a best-responding agent.
Researcher Affiliation Academia Tao Lin, Yiling Chen John A. Paulson School of Engineering and Applied Sciences Harvard University EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Convert any MAB algorithm to a contextual MAB algorithm
Open Source Code No The paper does not contain any explicit statements about releasing source code or links to code repositories. The text is ambiguous and lacks a clear, affirmative statement of release.
Open Datasets No The paper focuses on theoretical models and their analysis, using theoretical examples such as a "Bayesian persuasion instance" in Example 4.1. It does not describe or use any publicly available or open datasets for empirical evaluation.
Dataset Splits No The paper is theoretical and does not perform experiments on datasets, thus no dataset splits are provided.
Hardware Specification No The paper focuses on theoretical analysis and does not report any experiments that would require specific hardware. No hardware specifications are mentioned.
Software Dependencies No The paper mentions general algorithms like "Exp3 (Auer et al., 2002)" and refers to concepts such as "linear program" or "Python" but does not specify any particular software, libraries, or solvers with version numbers that were used for computational implementation or experimental setup.
Experiment Setup No The paper focuses on theoretical contributions, including defining models, deriving theorems, and proving bounds. It does not describe any empirical experiments, and therefore no experimental setup details, hyperparameters, or training configurations are provided.