Shared Neural Network Representations for Vehicle Memory Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks in vehicles consume a significant amount of resources due to their large size and complexity, necessitating a reduction in memory consumption and processing requirements.
Innovation Solution
Implementing a group of neural networks with shared layer portions that are shared between different networks, determined based on input constraints and optimized for memory and performance, allowing for dynamic provisioning based on vehicle conditions and resource availability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks are made larger to improve accuracy and handling varying circumstances, then response accuracy is improved, but memory consumption and resource requirements increase
Solution Approach 1:
The patent merges common layers across multiple neural networks into a single shared layer. Instead of storing separate copies of common layers for each neural network, the system stores one shared layer that is referenced by multiple networks. This dramatically reduces memory consumption while maintaining the functionality of multiple neural networks for different driving scenarios.
Solution Approach 2:
The shared layer serves multiple neural networks simultaneously, enabling a single layer to perform multiple functions. The shared layer can be used by different neural networks for different tasks (e.g., object detection, lane detection, traffic sign recognition) without requiring separate copies, thus reducing overall resource requirements while maintaining accuracy.
2Adaptability or versatility
If multiple neural networks are stored to handle different driving scenarios, then adaptability is improved, but memory consumption increases
Solution Approach 1:
The patent combines multiple neural networks into a unified structure where common layers are shared. Instead of storing complete separate neural networks for different driving scenarios, the system stores a shared layer that can be configured for different scenarios, reducing memory usage while maintaining versatility.
Solution Approach 2:
The system dynamically configures which neural networks to activate based on current driving conditions. Rather than having all neural networks loaded in memory simultaneously, the system can dynamically load and unload different neural network configurations from the shared layer based on the driving scenario, optimizing memory usage while maintaining adaptability.
3Quantity of substance
If neural network resources are reduced to lower memory consumption, then resource utilization is improved, but processing capability may be compromised
Solution Approach 1:
By merging common layers across neural networks, the system reduces memory consumption while preserving the processing capability needed for accurate driving scenario handling. The shared layer maintains the necessary computational functionality through efficient resource sharing rather than duplication.
Solution Approach 2:
The system can dynamically adjust parameters such as the degree of sharing, the resolution of shared representations, and the configuration of neural network layers based on computational requirements. This allows optimization between memory consumption and processing capability by adjusting parameters rather than using fixed architectures.
Data Source
AI summary
A method of producing a group of neural networks, the method includes determining a first layer potion that is shared by a first neural network sub-group of the group; determining second layer portions that are sharable by second neural network sub-groups, such that different second layer portions are shared by different second neural network sub-groups of the group; and determining third layer portions that are sharable by third neural network sub-groups, such that different third layer portions are shared by different third neural network sub-groups of the group; wherein each neural network of the group further comprises a unique combination of layer portions.