Leveraging Transfer Learning to Improve Convergence in All-Pay Auctions

Luis Eduardo Craizer, Edward Hermann, Moacyr Alvim Silva

2025

Abstract

In previous research on Multi-Agent Deep Deterministic Policy Gradient (MADDPG) in All-Pay Auctions, we identified a key limitation: as the number of agents increases, the tendency for some agents to bid 0.0 —resulting in local equilibrium — grows, leading to suboptimal bidding strategies. This issue diminishes the effectiveness of traditional reinforcement learning in large, complex auction environments. In this work, we propose a novel transfer learning approach to address this challenge. By training agents in smaller N auctions and transferring their learned policies to larger N settings, we significantly reduce the occurrence of local equilibrium. This method not only accelerates training but also enhances convergence toward optimal Nash equilibrium strategies in multi-agent settings. Our experimental results show that transfer learning successfully overcomes the limitations observed in previous research, yielding more robust and efficient bidding strategies in all-pay auctions.

Paper Citation

in Harvard Style

Craizer L., Hermann E. and Silva M. (2025). Leveraging Transfer Learning to Improve Convergence in All-Pay Auctions. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 534-543. DOI: 10.5220/0013294000003929

in Bibtex Style

@conference{iceis25,
author={Luis Craizer and Edward Hermann and Moacyr Silva},
title={Leveraging Transfer Learning to Improve Convergence in All-Pay Auctions},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={534-543},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013294000003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Leveraging Transfer Learning to Improve Convergence in All-Pay Auctions
SN - 978-989-758-749-8
AU - Craizer L.
AU - Hermann E.
AU - Silva M.
PY - 2025
SP - 534
EP - 543
DO - 10.5220/0013294000003929
PB - SciTePress