1

I want to try a hierarchical reinforcement learning (HRL) approach to hard logical problems with combinatorial complexity, i.e. games like chess or Rubik's cube. The majority of HRL papers I have found so far focus either on training a control policy or they tackle quite simple games.

By HRL I mean all methods that (among others):

  • split hard and complex problem into a series of simpler ones
  • create desired intermediate goals (or spaces of such goals)
  • somehow think in terms of 'what to achieve' rather than 'how to achieve'

Do you know any examples of solving logically hard problems with HRL or maybe just any promising approaches to such problems?

Tomasz Odrzygozdz
  • 11
  • 2
  • I am not sure if "hard logical problems with combinatorial complexity" is the right term/expression to describe games like chess or Rubik's cube, though I think I understand what you mean. Maybe the term "combinatorial optimization problems" is what you mean? – nbro Nov 23 '20 at 20:09
  • Check this paper [Combinatorial Optimization by Graph Pointer Networks and Hierarchical Reinforcement Learning](https://arxiv.org/pdf/1911.04936.pdf). – nbro Nov 30 '20 at 14:15
  • Thanks a lot. It may be something I was looking for. – Tomasz Odrzygozdz Dec 01 '20 at 21:22

0 Answers0