| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.84 MB | Adobe PDF |
Authors
Abstract(s)
A motivação para a produção deste trabalho está essencialmente ligada com o interesse particular na área de machine learning e no potencial da sua aplicação no quotidiano. Trata-se de uma área extremamente complexa, com desafios interessantes tanto a nível de conceitos teóricos como nível de tecnologias e metodologias de implementação. Esta área mostra um potencial de evolução enorme para os próximos anos, tendo uma capacidade poderosa de processamento e suporte à tomada de decisão inigualável na humanidade, sendo um conceito aplicável transversalmente a qualquer indústria e que, se corretamente aplicada, poderá trazer benefícios até agora inalcançáveis para a qualidade de vida do nosso dia-a-dia. Este trabalho pretende explorar a aplicação de técnicas de machine learning à análise de padrões de aprendizagem em jogos (gaming), com o intuito principal de identificar e otimizar as capacidades de algoritmos atualmente existentes e ainda com vista a investigar a relevância da aplicação do conceito de curiosidade nesses algoritmos. Esta exploração passará pela aplicação dessas técnicas a vários jogos, através da implementação de algoritmos de machine learning que interajam diretamente com os ambientes dos jogos e que aprendam a conhecer todas as características inerentes, com vista a reforçar o conhecimento e daí obterem a capacidade de concluir os jogos com sucesso. Desta forma, torna-se possível identificar padrões específicos a cada jogo. Globalmente, através da aplicação de técnicas de machine learning, nomeadamente Q-learning, uma técnica de treino de modelos de aprendizagem com o intuito de ser agnóstica perante o ambiente onde opera e dotada da capacidade de aprender a conhecer um ambiente e os seus obstáculos por forma a superá-los, pretende-se obter resultados de benchmarking para a comparação de padrões de aprendizagem no que se refere à performance, facilidade de implementação e aplicabilidade em cenários reais. Estes resultados servirão ainda para retirar conclusões da mesma natureza sobre a aplicação do conceito de curiosidade num algoritmo de aprendizagem inteligente em gaming, nomeadamente nos jogos Snake e Tetris, que serão abordados no contexto deste documento.
ABSTRACT: The motivation to produce this project is intrinsically associated with the particular author's interest in the area of machine learning and its potential applications. Machine learning is an extremely complex area, with a wide variety of applications and exciting challenges, both in theoretical concepts as well as methodologies of implementation. This area shows great potential in the upcoming years since it has an excellent capacity for data processing and supports decision-making unmatched on human beings. It is a field with a wide variety of applications on all industries, and may, therefore, bring benefits until now unreachable to the quality of our day-to-day lives. This project intends to explore the application of machine learning techniques to the analysis of learning patterns in the practical context of gaming, with the primary purpose to identify and optimize the capacity of existing algorithms and to investigate the relevance of applying the concept of curiosity to these types of algorithms. The tasks to explore this concept will primarily be using the techniques to multiple games, through implementing machine learning algorithms that interact directly with the gaming environments. They can learn and understand the games’ characteristics, and to use this information to reinforce the knowledge and being capable of beating these games successfully. This way, it shall be possible to identify learning patterns for each game. Globally, through the application of machine learning techniques, namely Q-learning, a technique intended to train learning models agnostically in terms of what environment it operates on and with the capacity to learn from an environment what are the obstacles and how to surpass them, it is expected to obtain results of benchmarking and for the comparison of learning patterns in the aspects of performance, ease of implementation, and application in real-life scenarios. These results will provide conclusions of the same nature in the form of a concept of curiosity on an intelligent agent of Artificial Intelligence interacting with gaming, namely the games Snake and Tetris, which will be explored in this document.
ABSTRACT: The motivation to produce this project is intrinsically associated with the particular author's interest in the area of machine learning and its potential applications. Machine learning is an extremely complex area, with a wide variety of applications and exciting challenges, both in theoretical concepts as well as methodologies of implementation. This area shows great potential in the upcoming years since it has an excellent capacity for data processing and supports decision-making unmatched on human beings. It is a field with a wide variety of applications on all industries, and may, therefore, bring benefits until now unreachable to the quality of our day-to-day lives. This project intends to explore the application of machine learning techniques to the analysis of learning patterns in the practical context of gaming, with the primary purpose to identify and optimize the capacity of existing algorithms and to investigate the relevance of applying the concept of curiosity to these types of algorithms. The tasks to explore this concept will primarily be using the techniques to multiple games, through implementing machine learning algorithms that interact directly with the gaming environments. They can learn and understand the games’ characteristics, and to use this information to reinforce the knowledge and being capable of beating these games successfully. This way, it shall be possible to identify learning patterns for each game. Globally, through the application of machine learning techniques, namely Q-learning, a technique intended to train learning models agnostically in terms of what environment it operates on and with the capacity to learn from an environment what are the obstacles and how to surpass them, it is expected to obtain results of benchmarking and for the comparison of learning patterns in the aspects of performance, ease of implementation, and application in real-life scenarios. These results will provide conclusions of the same nature in the form of a concept of curiosity on an intelligent agent of Artificial Intelligence interacting with gaming, namely the games Snake and Tetris, which will be explored in this document.
Description
Keywords
Machine Learning Reinforcement Learning Q-Learning Redes Neurais Gaming