US 12,423,570 B2
Accelerated training for neural network models
Christopher Ian Schneider, Hillend (GB); Amy Leigh Rose, Chapel Hill, NC (US); Andrew James Woodard, Buckinghamshire (GB); and Benjemin Thomas Waine, Cheshunt (GB)
Assigned to NVIDIA Corporation, Santa Clara, CA (US)
Filed by NVIDIA Corporation, Santa Clara, CA (US)
Filed on Jun. 10, 2020, as Appl. No. 16/898,223.
Prior Publication US 2021/0390414 A1, Dec. 16, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06N 3/045 (2023.01); G06N 3/084 (2023.01); G06N 5/046 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 18/217 (2023.01); G06N 3/045 (2023.01); G06N 3/084 (2013.01); G06N 5/046 (2013.01)] 25 Claims
OG exemplary drawing
 
1. One or more processors, comprising:
circuitry to cause one or more first neural networks to generate a set of converged neural network weights and a set of bias values, wherein to generate the set of converged neural network weights the one or more first neural networks are to:
generate one or more first weights for one or more second neural networks based, at least in part, on a first training iteration of the one or more second neural networks;
infer a direction and intensity value of one or more weights of the one or more second neural networks based, at least in part, on the one or more first weights and a second training iteration of the one or more second neural networks; and
generate the set of converged neural network weights and the set of bias values based, at least in part, on the inferred direction and intensity values of the one or more weights of the one or more second neural networks.