Smart Tuning of Algorithm Parameters by Deep Reinforcement Learning

[EN] In the field of algorithm optimisation, finding the optimal configuration of parameters is essential for achieving peak performance. Traditional methods usually rely on manual tuning or exhaustive search techniques, which can be time-consuming and inefficient. This paper proposes a novel approa...

Descripción completa

Detalles Bibliográficos
Autores: Serrano-Ruiz, Julio Cesar|||0000-0002-6671-7077, Mula, Josefa|||0000-0002-8447-3387, Poler, R.|||0000-0003-4475-6371
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:dnet:riunet______::950c64e1f8ef17d701c90c1f92663560
Acceso en línea:https://riunet.upv.es/handle/10251/235158
Access Level:acceso abierto
Palabra clave:Parameters tuning
Deep reinforcement learning
Holt-Winters
09.- Desarrollar infraestructuras resilientes, promover la industrialización inclusiva y sostenible, y fomentar la innovación
Descripción
Sumario:[EN] In the field of algorithm optimisation, finding the optimal configuration of parameters is essential for achieving peak performance. Traditional methods usually rely on manual tuning or exhaustive search techniques, which can be time-consuming and inefficient. This paper proposes a novel approach utilising deep reinforcement learning (DRL) for automatically tuning the parameters of those algorithms that require it. By formulating the parameter tuning problem of the Holt-Winters algorithm as a reinforcement learning task, this research provides an example of the use of this method, which enables algorithms to autonomously tune their parameters based on feedback from the environment shaped by the problem. Leveraging an advantage actor-critic algorithm (A2C), our framework learns optimal parameter settings through exploration and exploitation strategies. The main findings highlight the versatility and efficiency of this DRL-based method for parameter tuning in optimising algorithms, paving the way for more adaptive and self-improving systems in various domains.