Evolutionary Discovery of Multi-Agent Learning Algorithms with LLMs
Автор: Research Paper Review
Загружено: 2026-02-26
Просмотров: 37
Описание:
Google DeepMind researchers use AlphaEvolve, an evolutionary algorithm based on a large-scale language model (LLM), to automatically discover a new multi-agent enhanced learning algorithm. The research team has derived two innovative algorithm variants, VAD-CFR and SHOR-PSRO, which are highly effective by changing the code itself, away from the existing method that relied on human intuition. VAD-CFR dynamically adjusts the number of regrets according to the volatility of the learning process, and SHOR-PSRO has the ability to automatically optimize the balance of stability and exploration when establishing mixed strategies. As a result of the experiment, these automatically generated algorithms have proven to be faster and more efficient than conventional cutting-edge baselines in a variety of incomplete information games, including poker and dice games. As a result, this paper presents the possibility of automation of algorithm design that allows artificial intelligence to design its own advanced game theoretical learning rules.
https://arxiv.org/pdf/2602.16928
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
