17722858. METHOD AND APPARATUS WITH DYNAMIC CONVOLUTION simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

METHOD AND APPARATUS WITH DYNAMIC CONVOLUTION

Organization Name

SAMSUNG ELECTRONICS CO., LTD.

Inventor(s)

Seungin Park of Yongin-si (KR)

Sangil Jung of Yongin-si (KR)

Byung In Yoo of Seoul (KR)

METHOD AND APPARATUS WITH DYNAMIC CONVOLUTION - A simplified explanation of the abstract

This abstract first appeared for US patent application 17722858 titled 'METHOD AND APPARATUS WITH DYNAMIC CONVOLUTION

Simplified Explanation

The abstract of the patent application describes a method for dynamic convolution. Here is a simplified explanation:

• The method involves determining adaptation weights for weight matrices in a category set.
• These adaptation weights are represented by a plurality of predetermined discrete values.
• A unified kernel is determined based on the weight matrices and the corresponding adaptation weights.
• Finally, a convolution operation is performed using the unified kernel.

Potential Applications

• Image processing and computer vision algorithms.
• Deep learning and neural network models.
• Signal processing and audio analysis.
• Video compression and analysis.

Problems Solved

• Traditional convolution methods may not adapt well to different weight matrices.
• Limited flexibility in adjusting convolution kernels for specific tasks.
• Difficulty in optimizing convolution operations for different applications.

Benefits

• Improved adaptability and flexibility in convolution operations.
• Enhanced performance and accuracy in various applications.
• Simplified implementation of convolution algorithms.
• Potential for more efficient and optimized convolution operations.

Original Abstract Submitted

A method with dynamic convolution includes: determining kernel adaptation weights corresponding to weight matrices in a category set represented by a plurality of predetermined discrete values; determining a unified kernel based on the weight matrices and the kernel adaptation weights corresponding to the weight matrices; and performing a convolution operation based on the unified kernel.