Shared DNN Feature Extraction With EWC for Multi-Task Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous and assisted driving systems face high computing loads due to the need to process diverse scenarios, limiting the functions and features they can provide, as existing methods struggle to retain learned tasks while training for multiple tasks simultaneously.
Innovation Solution
The application of elastic weight consolidation (EWC) on a base neural network for feature extraction, allowing it to be sequentially connected to multiple networks for different tasks, thereby reducing computational load and preventing 'forgetting' of previously learned tasks, enabling efficient training and deployment across various tasks like object detection and lane finding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple separate neural networks are trained for different tasks, then each task can be performed accurately, but the computational load increases and training time extends
Solution Approach 1:
The patent combines multiple separate neural networks into a single shared neural network that performs multiple tasks simultaneously. The network is trained to perform object detection, lane finding, and other autonomous driving tasks within a unified architecture, reducing the number of separate models needed while maintaining task performance accuracy.
Solution Approach 2:
The neural network is designed with multi-functionality to handle various autonomous driving tasks through a single model. The network processes different input types (images, sensor data) and produces multiple output types (object detections, lane markings) simultaneously, eliminating the need for separate specialized networks for each task.
2Productivity
If a single neural network is trained for multiple tasks simultaneously, then computational load is reduced, but the network forgets previously learned tasks
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network on one task before introducing additional tasks. The network is first trained on object detection, then gradually extended to include lane finding and other tasks. This sequential task introduction allows the network to retain previously learned capabilities while adapting to new functions.
Solution Approach 2:
The training process incorporates feedback mechanisms where the network's performance on previously learned tasks is monitored during the addition of new tasks. If performance degradation is detected on earlier tasks, the training process adjusts by reinforcing those capabilities or adjusting the learning rate, ensuring that previously learned tasks are not forgotten.
3Reliability
If comprehensive training data is collected for all tasks, then model performance is improved, but data processing time and storage requirements increase
Solution Approach 1:
The patent segments the training data processing by handling different task-specific data streams separately within the unified network architecture. Object detection data, lane finding data, and other task data are processed through dedicated processing paths that feed into the shared network, allowing parallel processing and reducing overall data processing time while maintaining comprehensive training coverage.
Data Source
AI summary
A system includes a processor for performing one or more autonomous driving or assisted driving tasks based on a neural network. The neural network includes a base portion for performing feature extraction simultaneously for a plurality of tasks on a single set of input data. The neural network includes a plurality of subtask portions for performing the plurality of tasks based on feature extraction output from the base portion. Each of the plurality of subtask portions comprise nodes or layers of a neutral network trained on different sets of training data, and the base portion comprises nodes or layers of a neural network trained using each of the different sets of training data constrained by elastic weight consolidation to limit the base portion from forgetting a previously learned task.