Adaptive Neural Network Pruning for Vehicle Onboard Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning models for vehicles are bulky and require expensive computational resources, making them inefficient for deployment within vehicles and often necessitating hosting on cloud servers rather than onboard systems.
Innovation Solution
The system prunes a trained deep neural network based on a pruning ratio and threshold values, selectively removing nodes with low activation function outputs or weighted input derivatives, resulting in a smaller, more computationally efficient model that can be deployed within vehicles for accurate predictions and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep neural network is used for vehicle operations, then accurate predictions and classification are achieved, but the model becomes bulky and requires expensive computational resources
Solution Approach 1:
The patent extracts and removes redundant or less important nodes from the neural network while preserving the essential functionality. This is achieved through adaptive pruning that identifies and eliminates nodes contributing minimally to the overall output, thereby reducing model size and computational requirements while maintaining prediction accuracy
Solution Approach 2:
The patent applies different pruning strategies to different parts of the neural network based on their specific characteristics. Nodes are evaluated individually or in groups, and pruning decisions are made locally based on each node's contribution to the network output, allowing selective retention of important nodes while removing less critical ones
2Ease of operation
If a deep neural network is deployed within vehicles, then onboard processing is enabled, but the computational resources and storage requirements are excessive
Solution Approach 1:
The patent extracts only the essential computational elements needed for accurate vehicle operations, removing redundant nodes and connections. This creates a streamlined network that fits within vehicle computational constraints while maintaining the capability for onboard processing of sensor data and decision-making
Solution Approach 2:
The patent changes the parameters of the neural network through adaptive pruning, dynamically adjusting the network structure based on performance requirements and resource constraints. This allows the model to be optimized for deployment in vehicles with limited computational resources while maintaining necessary accuracy
Data Source
AI summary
A system can include a computer including a processor and a memory. The memory includes a trained neural network with instructions such that the processor is programmed to receive a pruning ratio and prune at least one node of the trained deep neural network based on a pruning ratio.