Distributed ANN Self-Learning for Unseen Vehicle Event Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks (ANNs) in autonomous vehicles struggle to accurately identify unexpected events or objects due to insufficient training, leading to unrecognized outputs and reduced precision in decision-making.
Innovation Solution
A distributed system for updating ANNs using a centralized server that performs supervised learning with sensor data collected from vehicles, combined with unsupervised learning at individual vehicles to enhance model accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If an ANN model is trained using only pre-deployment supervised learning, then the model achieves initial recognition capability for known events, but the model fails to accurately identify unexpected events or objects due to insufficient training data coverage
Solution Approach 1:
The system performs preliminary supervised learning training before deployment to establish baseline recognition capability. During operation, it continues to collect unexpected events and performs additional training, effectively doing preliminary preparation for future scenarios. This resolves the contradiction by ensuring the model is prepared for both known and unknown events through staged training approaches.
Solution Approach 2:
The ANN model performs self-updating by automatically collecting sensor data from unexpected events during operation, selecting relevant training data, and retraining itself without external intervention. This self-service mechanism enables the model to continuously improve its identification accuracy for unexpected events while maintaining its initial recognition capability, thereby resolving the technical contradiction.
2Productivity
If sensor data from all vehicles is continuously transmitted to a centralized server for supervised learning, then the model achieves comprehensive training coverage, but the system incurs high communication costs and data transmission delays
Solution Approach 1:
The system extracts only the essential and most valuable data for training purposes: sensor data from unexpected events. Rather than transmitting all sensor data continuously, it selectively extracts and transmits only those data points that contribute to improving model performance, thereby reducing communication overhead and transmission delays while maintaining comprehensive training coverage.
Solution Approach 2:
The system implements partial action by transmitting sensor data selectively rather than continuously. It performs supervised learning at the centralized server only when unexpected events occur, rather than maintaining constant communication. This partial approach achieves sufficient training coverage without the excessive communication costs and delays associated with continuous data transmission.
3Adaptability or versatility
If the ANN model is updated frequently with new training data, then the model maintains high adaptability to new events, but the system consumes excessive computational resources and energy
Solution Approach 1:
The system implements periodic supervised learning updates at the centralized server based on accumulated unexpected events rather than continuous updates. This periodic approach maintains high model adaptability by regularly incorporating new training data while significantly reducing computational energy consumption compared to frequent or continuous model retraining cycles.
Solution Approach 2:
The system performs preliminary data collection and selection at the edge devices before centralized processing. By pre-processing and selecting only relevant training data locally, it reduces the computational burden during the actual model training phase, thereby maintaining adaptability while lowering energy consumption during the computationally intensive training operations.
4Speed
If unsupervised learning is performed at individual vehicles, then the system achieves fast local adaptation to unexpected events, but the model lacks comprehensive training data from diverse vehicle environments
Solution Approach 1:
The system merges unsupervised learning at the individual vehicle level with centralized supervised learning. Local unsupervised learning provides fast adaptation to immediate unexpected events, while the centralized server aggregates data from multiple vehicles to perform comprehensive supervised learning that captures environmental diversity. This combination resolves the contradiction by integrating both speed and versatility.
Solution Approach 2:
The learning process is segmented into two distinct levels: local unsupervised learning at individual vehicles for rapid adaptation, and centralized supervised learning for comprehensive training. This segmentation allows each level to specialize - local learning handles speed requirements while centralized learning handles diversity requirements - thereby resolving the technical contradiction between adaptation speed and environmental coverage.
Data Source
AI summary
A vehicle having the first ANN model initially installed therein to generate outputs from inputs generated by one or more sensors of the vehicle. The vehicle selects an input based on an output generated from the input using the first ANN model. The vehicle has a module to incrementally train the first ANN model through unsupervised machine learning from sensor data that includes the input selected by the vehicle. Optionally, the sensor data used for the unsupervised learning may further include inputs selected by other vehicles in a population. Sensor inputs selected by vehicles are transmitted to a centralized computer server, which trains the first ANN model through supervised machine learning from sensor received inputs from the vehicles in the population and generates a second ANN model as replacement of the first ANN model previously incrementally improved via unsupervised machine learning in the population.