Reinforcement learning method that recasts quantum T-count optimization as tensor decomposition, using gadgets to cut T gates in fault-tolerant circuits.
A method based on deep reinforcement learning that exploits the relationship between optimizing T-count and tensor decomposition.
It incorporates domain-specific knowledge about quantum computation and leverages gadgets, which significantly reduce the T-count of optimized circuits.
It discovered an efficient algorithm akin to Karatsuba's method for multiplication in finite fields, and found the best human-designed solutions for arithmetic used in Shor's algorithm and quantum chemistry simulation.