US 11,836,610 B2
Concurrent training of functional subnetworks of a neural network
Dmitri Yudanov, Austin, TX (US); and Nicholas Penha Malaya, Austin, TX (US)
Assigned to Advanced Micro Devices, Inc., Santa Clara, CA (US)
Filed by ADVANCED MICRO DEVICES, INC., Santa Clara, CA (US)
Filed on Dec. 13, 2017, as Appl. No. 15/841,030.
Prior Publication US 2019/0180176 A1, Jun. 13, 2019
Int. Cl. G06N 3/08 (2023.01); G06N 3/045 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/045 (2023.01)] 19 Claims
OG exemplary drawing
 
1. A method of training an artificial neural network that comprises first subnetworks to implement known functions and second subnetworks to implement unknown functions, the method comprising:
training the first subnetworks separately and in parallel on corresponding known training data sets by executing separate instances of the first subnetworks concurrently on processing elements of a processor to determine first parameter values that define the first subnetworks;
providing input values from a network training data set to the artificial neural network including the trained first subnetworks;
generating error values by comparing output values produced by the artificial neural network to labeled output values of the network training data set;
using the error values to modify second parameter values that define the second subnetworks without modifying the first parameter values; and
storing the first and second parameter values,
wherein, after training of the artificial neural network is completed, the second subnetworks are substantially encompassed by the first subnetworks such that inputs to the second subnetworks are provided by the first subnetworks and outputs from the second subnetworks are provided to the first subnetworks.