Inductive Logic Programming problem is to find set of formulas H (hypothesis), that satisfy logical equations:
Known-knowledge-base & H => Set-of-positive-and-negative-facts-and-observations.
Relational Reinforcement Learning is Reinforcement Learning problem that expresses its inferred state transition function
Q(s, a)->s' and reward function
R(s, a)->reward (from whose the optimal policy can be easily inferred if those functions are inferred and knowns).
My question is - can ILP problem be recast as RRL problem, i.e. can RRL approach be used to solve ILP problem?
Currently I am reading https://www.cs.ox.ac.uk/files/232/ox_working_papers2.pdf about use of ILP for (natural) grammar induction, that is just one example of wide application of ILP.
But from the phylosophical and epistemological view I will that there is nothing beyong RL, there is no more general algorithm as RL, whole Knowledge should be learnt by thus. So, ILP should be solvable by RL as well. So - maybe I can express ILP as RRL and then I can apply RRL to natural language induction?