Vehicle Neural Network Tainted-Result Handling for Data Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vehicle neural network processors face unique and frequent errors due to their application-specific nature, which can lead to safety concerns during autonomous operation, as they are difficult to test thoroughly and may result in incorrect or unusable output data, potentially causing accidents.
Innovation Solution
A system and method for handling errors in neural network processors, including an error detector to identify data errors and a neural network controller that signals tainted results without terminating the neural network's execution, allowing the vehicle to continue operating safely by distinguishing between different types of errors and taking appropriate remedial actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the neural network processor terminates execution upon detecting a data error, then the reliability of the output data is improved, but the productivity of the vehicle operation is reduced
Solution Approach 1:
The patent segments the error handling process by distinguishing between different types of errors (data errors vs. other errors) and applying different responses. When a data error is detected in a specific computation, only that particular result is marked as invalid, while the neural network processor continues executing other computations without termination. This selective segmentation allows the system to maintain high reliability for valid outputs while preserving overall productivity.
2Productivity
If the neural network processor continues execution after detecting a data error, then the productivity is maintained, but the reliability of the output data deteriorates
Solution Approach 1:
The patent introduces an intermediary mechanism in the form of an error detector and controller that mediates between the computational units and the output system. When a data error is detected, the controller intervenes to mark the specific erroneous result as invalid without stopping the entire processing system. This intermediary layer enables the system to continue productive operations while protecting the reliability of output data by filtering out corrupted results.
3Reliability
If thorough testing is performed on the neural network processor, then the reliability is improved, but the time required for deployment is increased
Solution Approach 1:
The patent implements beforehand cushioning by incorporating error detection and handling capabilities directly into the neural network processor architecture before deployment. The error detector and controller are built-in components that prepare the system to handle errors gracefully during operation. This prior preparation allows the processor to be deployed with confidence in its reliability without requiring exhaustive testing of all possible error scenarios, thereby reducing deployment time while maintaining high reliability.
Data Source
AI summary
A system for handling errors in a neural network includes a neural network processor for executing a neural network associated with use of a vehicle. The neural network processor includes an error detector configured to detect a data error associated with execution of the neural network and a neural network controller configured to receive a report of the data error from the error detector. In response to receiving the report, the neural network controller is further configured to signal that a pending result of the neural network is tainted without terminating execution of the neural network.