US 12,307,218 B2
Information processing apparatus, control methods thereof, and recording medium for neural network learning models utilizing data minimization
Akihiro Tanabe, Tokyo (JP)
Assigned to CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed by CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed on Mar. 1, 2021, as Appl. No. 17/188,234.
Claims priority of application No. 2020-036041 (JP), filed on Mar. 3, 2020.
Prior Publication US 2021/0279572 A1, Sep. 9, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 7/485 (2006.01); G06F 7/50 (2006.01); G06F 18/22 (2023.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01)
CPC G06F 7/485 (2013.01) [G06F 7/50 (2013.01); G06F 18/22 (2023.01); G06N 3/08 (2013.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01)] 9 Claims
OG exemplary drawing
 
1. An information processing apparatus comprising:
one or more processors which execute computer instructions to perform the following:
inputting input data comprising digital image data;
determining if the input data has a size less than or equal to a predetermined threshold for the input data size;
if the input size is determined to be greater than the predetermined threshold for the input size data,
determining a predetermined value to be converted to 0 from among a plurality of values contained in the input data, wherein the predetermined value is a value that is calculated by the processor to have a highest frequency in the input data;
performing a processing operation of converting the values of the input data that match the predetermined value to 0, a processing operation of subtracting the predetermined value from the values of the input data other than the values converted to 0, and storing the corresponding subtraction results in computer memory; and
if a proportion of 0 values contained in processed data obtained as a result of the processing operations is less than a first threshold value, performing learning processing to generate a learned model using the processed input data that has undergone said conversion processing operation as input to the learned model;
if the input size is determined to be less than the predetermined threshold for the input size data, performing learning processing to generate a learned model using the processed input data that has undergone said conversion processing operation as input to the learned model,
wherein the learned model comprises learned coefficient parameters in at least one pre-created learned model, wherein in fully-connected layers of the pre-created learned model, the learned coefficient parameters correspond to a weight coefficient and a bias value of each edge connecting nodes of a plurality of different layers of the pre-created learned model, and in a convolutional neural network (CNN), the learned coefficient parameters correspond to a weight coefficient and a bias value of a kernel.