Convolutional Neural Network Segmentation for Processing Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks require significant computing power, which is expensive in terms of operation per Watt, and is hindered by scheduling overheads and data movement between on-die caches and off-die memory.
Innovation Solution
Divide convolutional neural networks into two parts, with weight factors stored locally in the first part to reduce the need for cache or external memory updates, optimizing the division point based on hardware availability and performance requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If convolutional neural networks use traditional centralized processing architecture, then data processing can be performed, but computing power consumption is high and processing speed is limited due to scheduling overheads and data movement between on-die caches and off-die memory
Solution Approach 1:
The patent divides the convolutional neural network processing into two distinct parts: a first part that processes input data and generates intermediate results, and a second part that processes the intermediate results to generate final output. This segmentation allows each part to be optimized independently, with the first part focusing on speed and the second part on accuracy, thereby improving overall processing efficiency while reducing energy consumption.
Solution Approach 2:
The patent introduces an intermediary mechanism where the first part of the neural network generates intermediate results that are then processed by the second part. This intermediary structure enables decoupling of computation-intensive operations from memory access operations, reducing scheduling overheads and data movement between on-die caches and off-die memory, thus improving processing speed while lowering energy consumption.
2Measurement precision
If convolutional neural networks require large computing power, then accurate data processing can be achieved, but operational expenses increase
Solution Approach 1:
The patent segments the neural network into two parts with different optimization goals: the first part is optimized for processing speed to handle computation-intensive operations efficiently, while the second part is optimized for processing accuracy to ensure high-quality output. This segmentation allows the system to achieve accurate data processing without requiring excessive computing power throughout the entire network, thereby reducing operational expenses.
Data Source
AI summary
A convolutional neural network system includes a first part of the convolutional neural network comprising an initial processor configured to process an input data set and store a weight factor set in the first part of the convolutional neural network; and a second part of the convolutional neural network comprising a main computing system configured to process an export data set provided from the first part of the convolutional neural network.