Fig. 4: Training the stochastic Ising machine to behave in a deterministic way.
From: Training an Ising machine with equilibrium propagation
Distribution of the steady-state energy of the spin system modeled by the D-Wave Ising machine when the exact same inputs are applied 1000 times to the same neural network architecture. The light pink distribution corresponds to the energy of the system before training it on MNIST/100. The purple distribution corresponds to the energy of the system after training it with EP. The samples before training have a flat distribution with a large standard deviation - which is the signature of a highly stochastic system. The samples before training exhibit a flat distribution with a substantial standard deviation, indicative of a highly stochastic system. In contrast, the samples after training are concentrated within a narrower region with a considerably smaller standard deviation. This demonstrates that the training procedure causes the Ising machine to exhibit more deterministic behavior.
