Adaptive ANN Inference Modes for Edge Power Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a challenge in balancing the requirements for high performance and low power consumption in artificial neural network (ANN) systems used in edge devices, particularly in always-on applications like mobile phones, due to the significant processing and power burden imposed by continuous operation.
Innovation Solution
An ANN computing system with a buffer, inference engine, and controller that adjusts operation modes based on an interest metric, transitioning to higher power and performance when necessary, using a subset of neural network layers and varying clock frequencies to optimize power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the inference engine operates continuously in high-performance mode, then detection accuracy and response speed are improved, but power consumption increases significantly
Solution Approach 1:
The inference engine dynamically switches between low-power mode and high-performance mode based on the interest metric of incoming data. The controller monitors the interest metric and adjusts the operating mode accordingly, making the system adaptive rather than static. This resolves the contradiction by allowing high detection accuracy only when necessary (high interest metric) while maintaining low power consumption during normal operation.
Solution Approach 2:
The system changes operational parameters (clock frequency, active layers) based on the interest metric. When the interest metric exceeds a threshold, the controller increases clock frequency and activates more neural network layers, thereby improving detection accuracy. When the interest metric is low, the system reduces parameters to minimize power consumption. This parameter adaptation resolves the contradiction between accuracy and power usage.
2Measurement precision
If the inference engine processes all input data with full neural network layers, then detection accuracy is improved, but processing speed and power efficiency deteriorate
Solution Approach 1:
The neural network is segmented into multiple layers that can be selectively activated. The controller can choose to process data through all layers (for high accuracy) or only through a subset of layers (for faster processing). This segmentation allows the system to balance detection accuracy with processing speed by activating only the necessary portion of the network based on the interest metric.
Solution Approach 2:
The system applies partial action by processing data through only the necessary number of neural network layers. When the interest metric is high, more layers are activated for thorough analysis. When the interest metric is low, fewer layers are sufficient, allowing faster processing. This partial action resolves the contradiction by avoiding unnecessary processing overhead while maintaining adequate detection accuracy.
3Use of energy by moving object
If the inference engine operates in low-power mode, then power consumption is reduced, but detection accuracy and performance decrease
Solution Approach 1:
The system uses feedback from the interest metric to control the operating mode. The interest metric provides information about the significance of incoming data, and the controller uses this feedback to switch between low-power and high-performance modes. This feedback mechanism ensures that detection accuracy is maintained when it matters (high interest metric) while minimizing power consumption during routine operation, resolving the contradiction between power efficiency and detection accuracy.
4Measurement precision
If the system transitions to high-performance mode frequently, then detection accuracy is maintained, but power consumption increases
Solution Approach 1:
The system serves itself by automatically adjusting its performance level based on the intrinsic quality of the input data (interest metric). The controller monitors the interest metric and autonomously decides when to switch modes without external intervention. This self-service mechanism ensures that high-performance mode is activated only when the data warrants it, preventing unnecessary power consumption while maintaining detection accuracy when needed.
Data Source
AI summary
The present disclosure relates to an artificial neural network (ANN) computing system comprising: a buffer configured to store data indicative of input data received from an input device; an inference engine operative to process data from the buffer to generate an interest metric for the input data; and a controller. The controller is operative to control a mode of operation of the inference engine according to the interest metric for the input data.