The article discusses the application of Inductive Logic Programming (ILP) for symbolic policy learning within text-based games (TBGs). EXPLORER, a system designed to collect data, extracts state descriptions and inventory information in order to manipulate entities as predicates for the ILP algorithm. It processes admissible actions and templates to identify entity types. To effectively learn rules, the ILP algorithm focuses on the goal, predicate list, and examples, showcasing the potential for dynamic rule generalization and incorporating feedback from experimental results.
To apply an ILP algorithm, first, EXPLORER needs to collect the State, Action, and Reward pairs while exploring the text-based environment. In a TBG, the two main components of the state are the state description and the inventory information of the agent.
By processing these templates over the admissible actions, EXPLORER can easily extract the type of each entity present in the environment and then convert them to predicates.
To learn the rules, an ILP algorithm requires three things - the goal, the predicate list, and the examples. The predicates give the explanation to a concept.
#artificial-intelligence #machine-learning #inductive-logic-programming #text-based-games #data-processing
Collection
[
|
...
]