US 11,938,943 B1
Slice-based dynamic neural networks
Khaled Refaat, Mountain View, CA (US); and Stéphane Ross, San Jose, CA (US)
Assigned to Waymo LLC, Mountain View, CA (US)
Filed by Waymo LLC, Mountain View, CA (US)
Filed on Sep. 28, 2021, as Appl. No. 17/487,903.
Int. Cl. B60W 40/00 (2006.01); G06F 18/211 (2023.01); G06F 18/23 (2023.01); G06N 3/08 (2023.01)
CPC B60W 40/00 (2013.01) [G06F 18/211 (2023.01); G06F 18/23 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method performed by one or more computers, the method comprising:
receiving a new input for processing by a neural network that has been trained on training data comprising a plurality of training inputs, wherein each training input belongs to one or more slices of a plurality of slices, and wherein the neural network comprises a first conditional neural network layer that has (i) a set of shared parameters and (ii) a respective set of slice parameters for each of the plurality of slices;
identifying one or more slices of the plurality of slices to which the new input belongs; and
processing the new input using the neural network to generate a network output for the new input, wherein the processing comprises:
receiving a layer input to the first conditional neural network layer; and
processing the layer input using (i) the set of shared parameters of the first conditional neural network layer, (ii) the respective one or more sets of slice parameters for the identified one or more slices in the plurality of slices, but not (iii) the respective sets of slice parameters for any other slices in the plurality of slices to which the new input does not belong.