Predicate Invention

Predicate invention focuses on automatically generating new symbolic representations (predicates) to improve machine learning models, particularly in areas like reinforcement learning and inductive logic programming. Current research explores methods for learning predicates from data, using techniques like neural networks to extract predicates from pre-trained agents or employing inductive logic programming with gradient descent to synthesize large logic programs containing invented predicates. This capability enhances model explainability, improves generalization from limited data, and enables more efficient planning in complex environments, ultimately advancing both the theoretical understanding and practical applications of artificial intelligence.