Resistive processing units and neural network training methods

A neural network comprising:

• an array of resistive processing units having:
• a plurality of rows of resistive processing units;
• and
• a plurality of columns of resistive processing units,
• wherein each resistive processing unit comprises:
• an AND gate configured to perform an AND operation of a first stochastic bit stream received from a first stochastic translator translating a number encoded from a neuron in a row and a second stochastic bit stream received from a second stochastic translator translating a number encoded from a neuron in a column;
• a first storage configured to store a weight value of the resistive processing unit;
• and
• a second storage configured to store an amount of change to the weight value of the resistive processing unit,
• wherein when the first stochastic bit stream and the second stochastic bit stream coincide, the amount of change to the weight value of the resistive processing unit is added to the weight value of the resistive processing unit.

A method of two step neural network training comprising:

• providing a first array of resistive processing units;
• performing a first step neural network training using the first array of resistive processing units;
• providing a high precision neural network;
• and
• performing a second step neural network training using the high precision neural network.