Table 1 Four types of classical neural network models implemented as algorithms in CPU or supercomputer (PEZY-Shobu) and the computational time to reach the 87.8% accuracy level provided by Goemans–Williamson semi-definite programing44

From: Coherent Ising machines—optical neural networks operating at the quantum limit

Coherent Ising machine

Experimental CIM 0.07 ms

Virtual CIM 100 s

Classical neural network

Deterministic

Stochastic

Discrete variable

Hopfield Network (HN) 0.924 (ms) by CPU

Simulated Annealing (SA) 2.10 (ms) by CPU

Continuous variable

Hopfield-Tank Neural Network (HTNN) 7.04 (ms) by SC

Langevin Dynamics (LD) 100 (ms) by SC

  1. CPU: Intel Xeon, E3–1225 v3, 3.2 GHz. Supercomputer: PEZY-SC, 733 MHz, 1024 cores. The computational time by experimental CIM23 and virtual CIM using the CSDE and replicator dynamics are also shown, where 16 PEZY-SC are used in parallel to implement the virtual CIM
Back to article page