用于神经网络的数据处理方法和装置
By separating the sparse processing of tensors and weights in systems with different data processing rates, the problems of wasted hardware resources and increased power consumption are solved, and efficient utilization of hardware resources and rational allocation of computing resources are achieved.
CN122174900BActive Publication Date: 2026-07-17MOXIN ARTIFICIAL INTELLIGENCE TECH (SHENZHEN) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MOXIN ARTIFICIAL INTELLIGENCE TECH (SHENZHEN) CO LTD
- Filing Date
- 2026-04-22
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
In existing neural network sparse computing techniques, the sparse processing of weights and tensors in the same high-speed computing system leads to a waste of hardware resources and an increase in power consumption, resulting in low hardware utilization.
Method used
The sparse processing of tensors and the sparse processing of weights are performed separately in two systems with different data processing rates. The processing rate of the tensor sparse system is lower than that of the weight sparse system, and reasonable computing resources are allocated to each system.
Benefits of technology
It improved the utilization rate of hardware resources, optimized the allocation of hardware and computing resources, and reduced device power consumption.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122174900B_ABST
Abstract
本公开提供了一种用于神经网络的数据处理方法和装置。该方法包括：利用张量稀疏系统对从共享数据缓存区取出的一个或多个稠密张量进行稀疏，以获得对应的稀疏张量，稀疏张量被缓存在共享数据缓存区中；利用权重稀疏系统对从共享数据缓存区取出的一个或多个稠密权重进行稀疏，以获得对应的稀疏权重，稀疏权重被缓存在权重稀疏系统中，其中，张量稀疏系统的数据处理速率比权重稀疏系统的数据处理速率低；将来自权重稀疏系统的稀疏权重和来自共享数据缓存区的稀疏张量送往乘法累加阵列进行计算，以获得神经网络的输出结果。
Need to check novelty before this filing date? Find Prior Art