Automated Intelligent Decision Making in Domains With Large Action Spaces

点击翻译按钮获取中文摘要

Artificial Intelligence (AI) research involves the development of new techniques for making intelligent decisions in a wide variety of ever more complex real-world problems. This process of automated decision making requires modelling a given domain, and then reasoning about future actions which are possible from a given scenario, eventually deciding on which action to take with some goal in mind. One of the major challenges faced by AI researchers is how to choose an action to take in a domain where the number of possible future scenarios and actions is so large that there is no way to explore them all. These domains are said to have large action spaces and with many of them requiring real-time decisions to be made in fractions of a second, the problem quickly becomes intractable. The main objectives of this proposal are: to better understand the nature of domains with large action spaces, to use this knowledge to develop novel techniques for tackling domains which were previously too complex, and to apply these new techniques to solve real-world problems. My specific objective for the next five years is to conduct research with HQP to investigate initially promising techniques, resulting in the completed theses for 4-5 MSc students (2 years each), and 2 PhD students (4 years each). The completion of these short-term objectives will pave the way for long term goals which include developing world-class AI systems, and collaboration with the Canadian technology industry.In recent years, an increasing number of AI researchers have been using computer games and simulations as test-beds for developing and testing new AI techniques. With harsh real-time computational constraints, complex rules, and an ability to simulate real-life scenarios, computer games are able to model the properties of real-world problems while being simulated in a more convenient controlled environment. I plan to use computer games as a test-bed for state-of-the-art techniques in AI. Through rigorous experimentation and data analysis, I hope to develop novel techniques in the areas of heuristic search, state abstraction, hierarchical decompositions, and dimensionality reduction, all of which are key methods required for solving problems in these complex domains.AI is quickly becoming an increasing part of our daily lives, powering technologies which advance industries such as e-commerce, autonomous robotics, health care, and computer games. As the industrial revolution created machines to do our heavy lifting, the AI revolution is creating machines to do our heavy thinking. Artificial intelligence is predicted to become one of the most important technology sectors in the next few years, and success in this proposed research will lead to advances in AI which can significantly add value to the Canadian economy, and improve overall quality of life for all Canadians.