Convolutional neural networks using resistive processing unit array

This invention describes a way to make artificial intelligence, specifically convolutional neural networks (CNNs), run much faster and more efficiently by using a special type of electronic circuit called a resistive processing unit (RPU) array. Instead of traditional computer chips, this method uses these RPU arrays to directly perform the complex calculations needed for CNNs, including learning (training) and making predictions, by sending electrical pulses and reading the resulting currents. The core idea is a computer-implemented method that configures and uses these RPU arrays for all the main computational steps of a CNN.