Google llc (20240185047). ROTATING DATA FOR NEURAL NETWORK COMPUTATIONS simplified abstract
Contents
ROTATING DATA FOR NEURAL NETWORK COMPUTATIONS
Organization Name
Inventor(s)
Jonathan Ross of Mountain View CA (US)
Gregory Michael Thorson of Waunakee WI (US)
ROTATING DATA FOR NEURAL NETWORK COMPUTATIONS - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240185047 titled 'ROTATING DATA FOR NEURAL NETWORK COMPUTATIONS
Simplified Explanation
The patent application describes a method for computing a layer output for a convolutional neural network layer using a systolic array architecture.
- Receiving a plurality of activation inputs.
- Forming vector inputs from the activation inputs, with each vector input containing values from a distinct region within a multi-dimensional matrix.
- Sending the vector inputs to cells along a first dimension of the systolic array.
- Generating rotated kernel structures from each kernel.
- Sending the kernel structures and rotated kernel structures to cells along a second dimension of the systolic array.
- Generating an accumulated output based on the value inputs and kernels.
- Generating the layer output from the accumulated output.
Potential Applications: - Image recognition - Speech recognition - Autonomous vehicles
Problems Solved: - Efficient computation of convolutional neural network layers - Optimized processing of large datasets
Benefits: - Faster processing speeds - Reduced computational resources required - Improved accuracy in neural network predictions
Potential Commercial Applications: - AI software development - Cloud computing services - Data analytics platforms
Possible Prior Art: - Previous patents related to systolic array architectures for neural networks - Research papers on optimized convolutional neural network computations
Questions: 1. How does this method compare to traditional convolutional neural network layer computations? 2. Are there any limitations to the systolic array architecture described in the patent application?
Frequently Updated Research: - Stay up to date with advancements in systolic array architectures for neural networks by following conferences such as NeurIPS and CVPR.
Original Abstract Submitted
methods, systems, and apparatus, including computer programs encoded on computer storage media, for computing a layer output for a convolutional neural network layer, the method comprising: receiving a plurality of activation inputs; forming a plurality of vector inputs from the plurality of activation inputs, each vector input comprising values from a distinct region within the multi-dimensional matrix; sending the plurality of vector inputs to one or more cells along a first dimension of the systolic array; generating a plurality of rotated kernel structures from each of the plurality of kernel; sending each kernel structure and each rotated kernel structure to one or more cells along a second dimension of the systolic array; causing the systolic array to generate an accumulated output based on the plurality of value inputs and the plurality of kernels; and generating the layer output from the accumulated output.
