> The fragment synthesis procedure uses a powerful, but computationally expensive, algebraic rewrite-based search algorithm to identify non-trivial compositions of blocks. The performance of the algorithm depends on the complexity of the target equation rather than the size of the overall system.
Could the construction of these blocks, as well as the composition step, benefit from training an RL agent on the state space of circuit configurations/the action space of connecting two blocks (or other operations)? Does any facet of the problem make the idea of using reinforcement learning intractable/otherwise a bad idea?
Could the construction of these blocks, as well as the composition step, benefit from training an RL agent on the state space of circuit configurations/the action space of connecting two blocks (or other operations)? Does any facet of the problem make the idea of using reinforcement learning intractable/otherwise a bad idea?