Variable Accuracy Computing System for Neural Network Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems for artificial neural networks face challenges in efficiently managing accuracy levels, leading to inefficient use of resources and potential low output accuracy due to fixed accuracy settings, which are not adaptable to specific application requirements.
Innovation Solution
A variable accuracy computing system that includes a data input, a computation unit, and a controller, where the controller monitors parameters of the input and output signals to adjust the number of bits used for processing, enabling dynamic control of accuracy based on monitored parameters, such as signal-to-noise ratio or environmental conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed accuracy settings are used in computing systems, then resource usage is simplified and device complexity is reduced, but output accuracy cannot be adapted to specific application requirements and resource efficiency deteriorates
Solution Approach 1:
The system dynamically adjusts the number of bits used for processing based on monitored parameters. The controller receives input data, monitors parameters of the input and/or desired output, and dynamically controls the computation unit to use different numbers of bits for different processing tasks, enabling adaptability without requiring complex fixed-precision configurations for all scenarios.
Solution Approach 2:
The invention changes the processing parameter (number of bits) based on monitored conditions. The controller modifies the bit depth used in computation units according to the monitored parameters of input data and desired output accuracy, allowing the system to optimize between accuracy and resource efficiency by adjusting this critical parameter dynamically rather than being fixed.
2Measurement precision
If higher accuracy processing is applied to all data, then output quality is improved, but power consumption and resource efficiency worsen
Solution Approach 1:
The system changes the processing precision parameter (number of bits) dynamically based on monitored input data characteristics and desired output accuracy. By adjusting this parameter rather than maintaining maximum precision consistently, the system reduces power consumption and resource usage when high accuracy is not required, while still maintaining high accuracy when needed.
Solution Approach 2:
The computation unit operates dynamically with variable precision levels. The controller adjusts the number of bits used in processing based on real-time monitoring of input parameters and desired output characteristics, enabling the system to consume less power during low-precision tasks while maintaining high precision capability when required.
3Use of energy by moving object
If lower accuracy processing is used, then power consumption and resource usage are reduced, but output quality and reliability deteriorate
Solution Approach 1:
The system adjusts the processing precision parameter (number of bits) based on monitored parameters and desired output accuracy requirements. By dynamically changing this parameter rather than using fixed low precision, the system ensures output reliability is maintained at appropriate levels while reducing power consumption when maximum reliability is not required.
Solution Approach 2:
The computation system operates with dynamically adjustable precision levels. The controller monitors input parameters and desired output characteristics to determine the appropriate number of bits to use, ensuring that output reliability matches the actual requirements of each processing task rather than being consistently high or low.
4Adaptability or versatility
If variable accuracy control is implemented, then adaptability to application requirements is improved and resource efficiency is enhanced, but device complexity and control mechanisms worsen
Solution Approach 1:
The system implements dynamic control where the controller adjusts the number of bits used in computation units based on monitored parameters. This dynamic approach provides high adaptability to different application requirements without requiring complex pre-configured systems for each scenario, as the control mechanism adjusts parameters in real-time based on actual processing needs.
Solution Approach 2:
The invention controls adaptability by changing the processing precision parameter (number of bits) based on monitored input data characteristics and desired output accuracy. This parameter-based control approach provides versatile adaptability across different applications while maintaining relatively simple control logic centered on monitoring and adjusting bit depth.
Data Source
AI summary
The present disclosure relates to a computing system. The computing system comprises a data input configured to receive an input data signal, a computation unit having an input coupled with the data input, the computation unit being operative to apply a weight to a signal received at its input to generate a weighted output signal, and a controller. The controller is configured to monitor a parameter of the input signal and/or a parameter of the output signal and to issue a control signal to the computation unit to control a level of accuracy of the weighted output signal based at least in part on the monitored parameter.