Methods and systems for implementing a convolution transpose layer of a neural network

• Main Authors: Clifford Gibson, James Imber, Cagatay Dikici
• Format: Patent
• Language: English
• Published: 12.07.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

Performing a convolution transpose operation between an input tensor comprising a plurality of input elements and a filter comprising a plurality of filter weights, by: dividing the filter into a plurality of sub-filters (1302); performing a convolution operation between the input tensor and each of the plurality of sub-filters to generate a plurality of sub-output tensors, each sub-output tensor comprising a plurality of output elements (1304); and interleaving the output elements of the plurality of sub-output tensors to form a final output tensor for the convolution transpose (1306). The convolution transpose operations are performed over a plurality of hardware passes. The input tensor may be multi-dimensional. The convolution transpose may be performed in the first and second dimension of the input tensor at a first and second stride respectively; and the plurality of sub-filters may comprise a number of sub-filters equal to the product of the first and second strides.