Extracting explanations about cooperation in a multi-agent reinforcement learning scenario

Reinforcement learning is a Machine Learning technique where an agent learns how to interact in an dynamic environment by performing actions and getting rewards or penalization as a feedback that will point how good or bad was the performed action. Many methods currently used in reinforcement learni...

ver descrição completa

Detalhes bibliográficos
Autor: Balaguera Lizcano, Miguel Angel
Tipo de documento: dissertação
Data de publicação:2023
País:España
Recursos:Universitat Politècnica de Catalunya (UPC)
Repositório:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglês
OAI Identifier:oai:upcommons.upc.edu:2117/396883
Acesso em linha:https://hdl.handle.net/2117/396883
Access Level:Acceso aberto
Palavra-chave:Reinforcement learning
reinforcement-learning
multi-agent
explainability
black-box
environment
actions
reward
Aprenentatge per reforç
Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic
Descrição
Resumo:Reinforcement learning is a Machine Learning technique where an agent learns how to interact in an dynamic environment by performing actions and getting rewards or penalization as a feedback that will point how good or bad was the performed action. Many methods currently used in reinforcement learning produce black-box models that are difficult to interpret, leading to unpredictable behavior and lack of trust on trained agents. Explainability methods can be used to tackle these issues, but they are usually applied on single-agent settings. We will explore the feasibility of applying a chosen method in a multi-agent setting with agents trained by us and we will explore how the explanations take into account the agent coordination aspects.