Neural Network Pruning for Low-Power Accurate Vehicle Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network systems for autonomous driving face challenges in reducing power consumption and maintaining inference accuracy, particularly when pruning weighting factors, which can lower the accuracy of high-accuracy data inference.
Innovation Solution
An electronic control device and neural network update system that identify and prune connection information based on influence, creating a third connection information group from a first and second connection information group, reducing the amount of arithmetic operations and power consumption while maintaining inference accuracy for critical data sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If pruning processing is applied to reduce the amount of arithmetic operation and power consumption, then power consumption is reduced, but inference accuracy on critical data deteriorates
Solution Approach 1:
The patent applies different quality standards to different data types by maintaining complete connection information for high-accuracy required data (first data) while applying pruning to other data. This local differentiation ensures that critical inference tasks retain full accuracy while non-critical tasks benefit from reduced computational load and power consumption.
Solution Approach 2:
The patent segments the connection information into multiple groups based on the importance and accuracy requirements of different data types. By dividing the neural network parameters into segments with different pruning levels, the system can optimize power consumption for non-critical paths while preserving accuracy for critical inference tasks.
2Productivity
If the amount of connection information is reduced to decrease arithmetic operation amount, then processing speed increases, but inference quality deteriorates
Solution Approach 1:
Different quality levels are applied to different data processing paths based on their importance. High-accuracy required data maintains full connection information for optimal inference quality, while other data uses pruned connection information to improve processing speed and reduce computational overhead.
Solution Approach 2:
Instead of applying uniform pruning to all connection information, the patent applies partial pruning selectively. Complete connection information is preserved for critical data paths where inference quality is paramount, while pruning is applied to non-critical paths to achieve speed improvements without compromising overall system performance.
Data Source
AI summary
An electronic control device is an electronic control device that stores a neural network including plural neurons and a connection information group including plural pieces of connection information that associate the neurons with each other. The neural network is a neural network including a third connection information group created through deletion, by a pruning section, of at least one piece of connection information from a first connection information group for deciding an output value when an input value is given on the basis of the first connection information group and a second connection information group that is plural pieces of connection information having the degree of influence on predetermined data given as the input value, the degree of influence exceeding a predetermined value.