Neural Network Flag Layer Processor for Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Convolutional Neural Networks (CNN) models in image recognition systems consume excessive power due to continuous operation across all layers, even when input images show minimal or no change, leading to inefficient power usage in portable devices.
Innovation Solution
An artificial neural network apparatus and method that includes a flag layer processor to compare pooling output values of current and previous frames, outputting a flag to stop the operation of subsequent layer processors when no change is detected, thereby minimizing unnecessary power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all layers from input layer to output layer are activated to generate recognition results, then object recognition accuracy is improved, but power consumption increases excessively
Solution Approach 1:
The neural network is segmented into multiple layers with a flag layer processor and subsequent layer processors. The system processes only necessary layers based on frame comparison results, dividing the computation into conditional segments rather than always processing all layers.
Solution Approach 2:
The system dynamically adjusts the operation of layer processors based on real-time comparison results between current and previous frames. When frames are identical, subsequent layer processors are stopped; when frames differ, they operate normally, making the system adaptive to input conditions.
2Stability of the object's composition
If hardware operates from input layer to output layer in the same manner regardless of input data states, then consistent processing is maintained, but power consumption becomes excessive
Solution Approach 1:
The system transitions from static uniform processing to dynamic conditional processing. The controller adjusts the operation state of layer processors based on the flag output from the flag layer processor, which is determined by comparing current and previous frame pooling output values.
Solution Approach 2:
The system uses feedback from the flag layer processor's comparison result to control the operation of subsequent layer processors. The comparison output serves as feedback that determines whether to activate or stop subsequent processing stages.
3Reliability
If feature generation to object recognition processes are completed in series, then comprehensive analysis is achieved, but power consumption exceeds battery performance of portable devices
Solution Approach 1:
The series of processing stages is segmented into a flag layer processor that performs pooling and comparison operations, and subsequent layer processors that perform full recognition. This segmentation allows selective execution based on whether frame changes are detected.
Solution Approach 2:
The system performs partial processing by executing only the flag layer processor when frames are identical, and performing full processing including all subsequent layers only when frame changes are detected. This avoids excessive computation when it is unnecessary.
Data Source
AI summary
An artificial neural network apparatus and an operating method including a plurality of layer processors for performing operations on input data are disclosed. The artificial neural network apparatus may include: a flag layer processor for outputting a flag according to a comparison result between a pooling output value of a current frame and a pooling output value of a previous frame; and a controller for stopping operation of a layer processor which performs operations after the flag layer processor among the plurality of layer processors when the flag is outputted from the flag layer processor, wherein the flag layer processor is a layer processor that performs a pooling operation first among the plurality of layer processors.