US 12,299,567 B2
Convolution size prediction to reduce calculations
Daren Croxford, Swaffham Prior (GB); Sharjeel Saeed, Cambridge (GB); and Rachel Jean Trimble, Grindleford (GB)
Assigned to Arm Limited, Cambridge (GB)
Filed by Arm Limited, Cambridge (GB)
Filed on Sep. 20, 2021, as Appl. No. 17/479,257.
Prior Publication US 2023/0089112 A1, Mar. 23, 2023
Int. Cl. G06N 3/04 (2023.01); G06F 18/10 (2023.01); G06F 18/211 (2023.01); G06N 3/0464 (2023.01); G06N 3/08 (2023.01); G06V 10/75 (2022.01)
CPC G06N 3/08 (2013.01) [G06F 18/10 (2023.01); G06F 18/211 (2023.01); G06N 3/0464 (2023.01); G06V 10/751 (2022.01)] 19 Claims
OG exemplary drawing
 
1. A data processing apparatus configured to perform machine learning comprising:
convolution circuitry configured to convolve a plurality of neighbouring regions of input data using a kernel to produce convolution outputs;
max-pooling circuitry configured to determine and select the largest of the convolution outputs as a pooled output; and
prediction circuitry configured to perform a size prediction of the convolution outputs based on the neighbouring regions, wherein the size prediction is performed prior to the max-pooling circuitry determining the largest of the convolution outputs and to adjust a behaviour of the convolution circuitry based on the size prediction, wherein
the prediction circuitry is configured to predict a size order of the convolution outputs; and
the prediction circuitry is configured to adjust the behaviour of the convolution circuitry based on the size prediction by causing the convolution circuitry to convolve the plurality of neighbouring regions based on the size order descending.