I enjoyed the paper but one thing they didn't seem to consider was the human-difficulty of the move e.g. sometimes long horizontal moves are harder to see (sorry, no reference, just anecdotal based on experience), so one could see later works that fold that in.

Also, they used endgames with IIRC < 7 pieces, so again, some derivative work might use middlegame positions and use engine analysis as an oracle, similar to Regan's work (http://www.buffalo.edu/news/experts/ken-regan-faculty-expert...).