Multi-Resolution Neural Signal Monitoring for Faster Feature Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks for tracking features in time sequences of physical signals require extensive computations, leading to long execution times that negatively impact processing rates and performance.
Innovation Solution
Implementing a multi-resolution neural network with a single set of weights that switches between nominal and accelerated processing modes based on the probability of feature presence, using reduced resolution when the probability is high to increase processing rate and power efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a complex neural network is used to track features in physical signals, then measurement precision and reliability are improved, but execution time increases and processing rate decreases
Solution Approach 1:
The system dynamically switches between two neural network configurations: a full-precision neural network for initial feature detection and a simplified quantized neural network for continuous tracking. This dynamic adaptation allows the system to maintain high detection precision when needed while achieving high processing rates during tracking, effectively resolving the contradiction between precision and productivity
Solution Approach 2:
The invention changes the precision parameter of the neural network weights by using quantization (reducing weight precision) in the accelerated mode. This parameter change enables faster computation with acceptable performance, allowing the system to achieve high processing rates while maintaining sufficient feature tracking accuracy
2Productivity
If neural network execution time is reduced to increase processing rate, then productivity is improved, but measurement precision and reliability deteriorate
Solution Approach 1:
The system employs dynamic mode switching between nominal and accelerated processing based on the detection state. When a feature is first detected using the full-precision network, the system switches to the accelerated quantized network for tracking. This dynamic approach ensures high precision for detection while maintaining high processing rates for tracking, resolving the contradiction between productivity and measurement precision
Solution Approach 2:
The full-precision neural network performs preliminary feature detection before the accelerated network takes over for tracking. This preliminary action ensures that feature detection is performed with high precision using the complete network capabilities, while subsequent tracking operations can use the faster quantized network, thus maintaining both precision and productivity
3Productivity
If a simplified neural network is used to increase processing rate, then productivity is improved, but reliability and detection accuracy are lost
Solution Approach 1:
The system dynamically selects the appropriate neural network configuration based on the operational phase: full-precision network for initial detection requiring high reliability, and quantized network for continuous tracking where processing rate is more critical. This dynamic selection maintains overall system reliability while achieving high processing rates during tracking operations
Solution Approach 2:
The full-precision neural network performs preliminary detection to establish reliable feature identification before the quantized network begins tracking. This preliminary reliable detection ensures that the feature is correctly identified before switching to the faster but less reliable quantized network, maintaining overall tracking reliability while enabling high processing rates
Data Source
AI summary
According to one aspect, the disclosure proposes a method for detecting events or features in physical signals by implementing an artificial neural network. The method includes evaluating the probability of presence of the event or feature by implementing the artificial neural network. The method includes implementing the artificial neural network in a nominal mode and to which a physical signal having a first so-called nominal resolution is fed, as long as the probability of the presence of the event or feature is below a threshold. The method further includes implementing the artificial neural network in a reduced consumption mode with a reduced resolution, as long as the probability of the presence of the event or feature is above the threshold. The reduced resolution is lower than the first resolution.