Reminds me of one of my favourite ML papers, DreamCoder: https://dl.acm.org/doi/pdf/10.1145/3453483.3454080