Machine learning approaches inspired by human reasoning often involve large ‘environments’ in which tasks are modelled. Existing computational methods do not scale well to deal with this, therefore this project is looking at developing ways of reducing the size of such environments. A first application and test case for these techniques is for playing games, like chess, go, or poker.
Explaining the science
Reinforcement learning is a very simple machine learning scenario: an agent put in an unknown environment performs actions; actions gives rewards; thus the agent wants to learn how to maximise rewards. This pervasive approach, inspired by human reasoning, is nowadays used in a huge number of fields.
The typical limitation is that the environment is often very large, and classical computational methods do not scale up. Hence the need for new paradigms.
This project is about developing sound ways of reducing the size of the environment reinforcement learning tasks. In other words, we want to first construct a smaller and hopefully equivalent environment of a task, to be able to then more easily perform the task. While there exists some statistical methods for approaching such reduction, the goal of this project is to approach model reduction in an algorithmic way.
The main challenge and long-term objective is to apply these techniques to program verification and synthesis.
A first application and test case for these techniques is for playing games, like chess, go, or poker. The environment for such games is beyond computer capabilities, hence model reduction may be a life changer, as recent progress has proved.