| CPC G06N 3/088 (2013.01) [G06N 3/045 (2023.01)] | 34 Claims |
|
1. A processor-implemented method, the method comprising:
obtaining target data;
sampling a trained first neural network into a plurality of second neural networks,
wherein the sampling comprises randomly extracting architecture parameters for a plurality of sub-networks of the second neural network;
training each of the second neural networks based on a portion of the target data, wherein the training of each of the second neural networks comprises training an architecture parameter of each of the second neural networks using respective outputs obtained by inputting the portion of the target data to each of the second neural networks;
selecting a second neural network satisfying a predetermined condition among the trained second neural networks for performing an inference operation; and
performing the inference operation on the target data using the selected second neural network to output an operation result based on the target data.
|