18093051. EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR simplified abstract (Apple Inc.)

EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR

Organization Name

Apple Inc.

Inventor(s)

Sayyed Karen Khatamifard of Bellevue WA (US)

Alon Yaakov of Raanana (IL)

Jeffrey D. Marker of Pleasant View UT (US)

EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR - A simplified explanation of the abstract

This abstract first appeared for US patent application 18093051 titled 'EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR

Simplified Explanation: The patent application describes a method for efficient zero-padding in convolution, where a set of values indicating a zero-padding pattern is used during computations associated with convolution.

• The method involves accessing a partition of an input tensor and dividing it into partitions in a raster-scan direction.
• For each row of a kernel used in convolution, a register is populated with a set of values representing a zero-padding pattern.
• Computations during convolution are performed based on the zero-padding pattern for each compute cycle in the row of the kernel.
• An updated zero-padding pattern for the next cycle is generated, and the values in the register are updated accordingly.

Key Features and Innovation:

• Efficient zero-padding method in convolution.
• Utilizes a zero-padding pattern for computations in convolution.
• Registers store and update the zero-padding pattern for each cycle in the kernel row.

Potential Applications:

• Image processing applications.
• Signal processing tasks.
• Machine learning algorithms.

Problems Solved:

• Improves efficiency in zero-padding during convolution.
• Enhances the accuracy of convolution operations.
• Streamlines the computation process in convolution tasks.

Benefits:

• Faster convolution operations.
• Reduced computational complexity.
• Improved accuracy in convolution results.

Commercial Applications: The technology can be applied in various industries such as:

• Computer vision systems.
• Medical imaging software.
• Autonomous vehicles for real-time image processing.

Prior Art: Readers can explore prior art related to efficient zero-padding methods in convolution in academic journals, patent databases, and research papers in the field of signal processing and machine learning.

Frequently Updated Research: Stay updated on advancements in efficient convolution techniques, signal processing algorithms, and machine learning models that incorporate zero-padding methods.

Questions about Efficient Zero-Padding in Convolution: 1. What are the potential implications of this technology in real-time image processing applications? 2. How does this method compare to traditional zero-padding techniques in terms of computational efficiency and accuracy?

Original Abstract Submitted

Embodiments relate to a method of efficient zero-padding in convolution. The method includes accessing a partition among a plurality of partitions of an input tensor. The input tensor is divided into the plurality of partitions in a raster-scan direction. For each row of a kernel for performing convolution on the input tensor, a register is populated with a set of values indicating a zero-padding pattern. For a compute cycle in the row of the kernel, computations associated with the convolution are performed based in part on the zero-padding pattern. After that, an updated zero-padding pattern representing a zero-padding pattern for a next cycle in the row of the kernel is generated. The set of values in the register is updated to the updated zero-padding pattern.