Neural Network Weight Training via Parallel Evolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicle systems face challenges in adapting to varying operational environments due to the reliance on large training datasets and passive vision systems, leading to performance deterioration and increased computational load.
Innovation Solution
A computer-implemented method for training a low-resolution neural network using an artificial evolution algorithm that evaluates multiple sets of weights concurrently, allowing for efficient optimization and adaptation to specific tasks such as image classification and object detection, while reducing computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep neural networks with multiple layers are used to improve object identification accuracy, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent uses a simplified neural network architecture that copies only the essential functional elements needed for object identification, rather than implementing full deep neural networks. This allows achieving acceptable accuracy while significantly reducing computational complexity and hardware requirements.
Solution Approach 2:
The invention extracts and removes unnecessary layers and computational elements from traditional deep neural networks, retaining only the critical components needed for autonomous vehicle perception tasks. This extraction process reduces device complexity while maintaining core functionality.
2Manufacturing precision
If large training datasets are used to improve neural network performance, then manufacturing precision is improved, but loss of substance increases
Solution Approach 1:
The patent applies partial training actions by using smaller, strategically selected datasets rather than exhaustive large-scale training. The simplified architecture requires fewer training examples to achieve adequate performance, reducing the computational resources and data volume needed.
Solution Approach 2:
The invention uses lightweight, computationally inexpensive training approaches that consume fewer resources. The simplified neural network can be trained efficiently with limited resources, making the training process more sustainable and less resource-intensive.
3Device complexity
If passive vision systems are used to simplify image processing, then device complexity is reduced, but adaptability to varying environments deteriorates
Solution Approach 1:
The patent introduces dynamic elements to the vision processing system, allowing it to adapt its processing behavior based on environmental conditions. The system can dynamically adjust its operation to handle varying lighting, weather, and scene conditions while maintaining relatively simple hardware architecture.
Solution Approach 2:
The invention changes key operating parameters of the vision system to improve environmental adaptability. By adjusting parameters such as processing thresholds, sensitivity levels, and operational modes, the system achieves better adaptability without significantly increasing device complexity.
4Productivity
If specialized hardware is used to accelerate neural network inference, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent employs hardware that can perform multiple functions, including neural network inference, general-purpose computing, and various sensor processing tasks. This multi-functional approach accelerates inference without requiring dedicated specialized hardware for each function, thereby reducing overall device complexity.
Data Source
AI summary
A computer-implemented method and system for training a neural network are provided. The method includes generating a first plurality including a plurality of sets of weights for a neural network; evaluating each set of weights in the first plurality, wherein evaluating includes, for each set of weights: fitting the set of weights to a neural network; presenting training data to an input of the neural network; and calculating a fitness score for the set of weights based on a fitness function that is dependent on an output of the neural network, wherein evaluating each set of weights occurs at least partly concurrently, such that two or more sets of weights in the first plurality are evaluated at the same time; and generating a second plurality including a plurality of sets of weights for the neural network, wherein generating the second plurality includes applying a training algorithm to the sets of weights of the first plurality to generate the second plurality, the sets of weights of the second plurality being dependent on the sets of weights of the their respective fitness scores. The system comprises a primary module configured to implement the method apart from the evaluation, which is done at a secondary module which has parallel computing capabilities.