Apple inc. (20240220764). EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR simplified abstract
EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR
Organization Name
Inventor(s)
Sayyed Karen Khatamifard of Bellevue WA (US)
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 20240220764 titled 'EFFICIENT ZERO PADDING IN CONVOLUTION AT NEURAL PROCESSOR
The method described in the patent application involves efficient zero-padding in convolution by accessing partitions of an input tensor and populating a register with values indicating a zero-padding pattern for each row of a kernel.
- The input tensor is divided into partitions in a raster-scan direction.
- Computation associated with convolution is performed based on the zero-padding pattern in the register.
- An updated zero-padding pattern for the next cycle in the row of the kernel is generated.
- The values in the register are updated to reflect the updated zero-padding pattern.
Potential Applications: - Image processing - Signal processing - Machine learning algorithms
Problems Solved: - Improving efficiency in convolution operations - Optimizing zero-padding techniques
Benefits: - Faster computation in convolution tasks - Enhanced performance in deep learning models
Commercial Applications: Title: "Efficient Zero-Padding Method for Convolution in Image Processing" This technology can be utilized in industries such as: - Computer vision - Autonomous vehicles - Medical imaging
Questions about the technology: 1. How does this method compare to traditional zero-padding techniques in terms of computational efficiency? 2. What impact could this innovation have on the development of advanced AI algorithms?
Frequently Updated Research: Stay updated on advancements in convolution techniques and optimization strategies for deep learning models.
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.
