Neural Network Controller Boundary Condition Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks lack the ability to effectively recognize and correct incorrect output signals, failing to adaptively reduce mistakes with new information, leading to ineffective operation in autonomous systems.
Innovation Solution
A controller system that includes a first neural network trained with initial data and a manager to re-train it incrementally using local data, with boundary conditions to filter and store input-output pairs for later training, and a security architecture to prevent malfunctions and attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a conventional neural network is trained using training data, then it can generate output signals for given inputs, but it becomes incapable of recognizing or determining when incorrect output is generated
Solution Approach 1:
The patent implements a feedback mechanism where the neural network's output is fed back into the system through a boundary condition checker. This creates a closed-loop system that continuously monitors whether outputs satisfy predefined boundary conditions, enabling the system to recognize incorrect outputs without requiring additional complexity in the neural network itself.
Solution Approach 2:
The patent introduces a boundary condition checker as an intermediary component between the neural network and the final output. This mediator evaluates whether the neural network's output satisfies predefined boundary conditions, allowing the system to detect incorrect outputs while keeping the neural network structure relatively simple.
2Reliability
If a conventional neural network operates without corrective mechanisms, then it maintains simple operation, but it cannot take corrective measures when incorrect output is generated
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically detects boundary condition violations and triggers retraining processes without external intervention. The neural network manager autonomously identifies when corrective action is needed and initiates the appropriate correction procedures, maintaining ease of operation while enabling mistake correction.
Solution Approach 2:
The patent establishes predefined boundary conditions before the neural network begins operation. These boundary conditions serve as preliminary criteria that the output must satisfy, allowing the system to quickly determine whether corrective action is needed without complex real-time analysis, thus maintaining operational simplicity.
3Adaptability or versatility
If the neural network operates without adaptive retraining, then it maintains stable performance, but it cannot reduce mistakes with new information
Solution Approach 1:
The patent implements a dynamic system where the neural network's training state can change from static (initial training) to dynamic (incremental retraining) based on boundary condition violations. The system adapts its behavior by transitioning between different operational modes, allowing it to learn from new information while maintaining stability during normal operation.
Solution Approach 2:
The patent enables continuous learning through incremental retraining that occurs whenever boundary conditions are violated. Rather than discrete periodic retraining, the system continuously adapts by processing new information as it becomes available, maintaining both adaptability and stability through this continuous improvement process.
Data Source
AI summary
Systems and methods for automatically self-correcting or correcting in real-time one or more neural networks after detecting a triggering event, or breaching boundary conditions are provided. Such a triggering event may indicate incorrect output signal or data being generated by the one or more neural networks. In particular, machine controllers of the invention limit the operations of neural networks to be within boundary conditions. Autonomous machines of the invention can be self-corrected after a breach of a boundary condition is detected. Autonomous land vehicles of the invention are capable of determining the timing of automatic transition to the manual control from automated driving mode. The controller of the invention filters and saves input-output data sets that fall within boundary conditions for later training of neural networks. The controllers of the invention include security architectures to prevent damages from virus attacks or system malfunctions.