Calibration of matrix-vector operations on resistive processing unit hardware

A computer program product, comprising:

• one or more non-transitory computer readable storage media, and program instructions collectively stored on the one or more non-transitory computer readable storage media, the program instructions comprising:
• program instructions to control a resistive processing unit (RPU) system to store a matrix of weight values in a crossbar array of cells of a resistive processing unit (RPU) system, the cells comprising respective resistive memory devices having respective conductances that are programmable, wherein the matrix of weight values is stored by programming the conductances of at least portion of the resistive memory devices to store the matrix of weight values in the crossbar array of cells;
• program instructions to perform a calibration process to determine calibration factors that are configured to compensate for hardware errors of the RPU system when the digital processing system utilizes the RPU system to perform analog matrix-vector multiplication operations using the stored matrix in the crossbar array of cells, wherein the program instructions to perform the calibration process comprise:
• program instruction to generate a first set of calibration parameters for calibrating forward pass matrix-vector multiplication operations performed on the stored matrix in the crossbar array of cells of the RPU system;
• and
• program instruction to generate a second set of calibration parameters for calibrating backward pass matrix-vector multiplication operations performed on a transpose of the stored matrix in the crossbar array of cells;
• wherein the first set and the second set of calibration parameters are configured to calibrate respective forward pass and backward pass matrix-vector multiplication operations so that effective weight values of the stored matrix for a given forward pass matrix-vector multiplication operation correspond to effective weight values of the transpose of the stored matrix for a given backward pass matrix-vector multiplication operation;
• program instructions to calibrate forward pass matrix-vector multiplication operations by applying the first set of calibration parameters to input vectors that are applied to the crossbar array of cells to perform forward pass matrix-vector multiplication operations, or to output vectors that are output from the crossbar array of cells after performing forward pass matrix-vector multiplication operations;
• and
• program instructions to calibrate backward pass matrix-vector multiplication operations by applying the second set of calibration parameters to input vectors that are applied to the crossbar array of cells to perform backward pass matrix-vector multiplication operations, or to output vectors that are output from the crossbar array of cells after performing backward pass matrix-vector multiplication operations.