Inference Processing Apparatus Data Filtering for Embedded Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inference processing on small embedded devices faces challenges in increasing speed while reducing power consumption due to the need to process all input data, which is inefficient and power-intensive.
Innovation Solution
An inference processing device that uses a learned neural network to infer features, incorporating a data filtering unit to extract only specific input data and reduce unnecessary processing, thereby optimizing speed and power usage by comparing input data against previous data and performing inference only when significant differences are detected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If inference operation is performed on all input data using conventional methods, then processing completeness is maintained, but power consumption increases and processing speed decreases
Solution Approach 1:
The patent extracts only the necessary input data that requires inference processing by comparing it with stored historical data. The data filtering unit extracts input data only when changes are detected, removing unnecessary processing of redundant data and thereby reducing power consumption while maintaining processing speed.
Solution Approach 2:
The patent performs preliminary comparison of input data with historical data stored in memory before executing inference operations. This preliminary action filters out redundant data and identifies only the necessary data for processing, reducing both power consumption and processing time while maintaining accuracy.
2Measurement precision
If all input data is processed through the neural network, then comprehensive analysis is achieved, but processing time increases
Solution Approach 1:
The patent extracts only the essential input data that has changed compared to historical data. By using the data filtering unit to identify and extract only necessary data for inference, the system maintains measurement precision for critical changes while significantly reducing processing time by skipping redundant data.
Solution Approach 2:
The patent performs preliminary comparison of current input data with stored historical data before inference processing. This preliminary action identifies only the necessary data points that require analysis, maintaining accuracy for significant changes while reducing overall processing time by avoiding unnecessary inference operations on redundant data.
Data Source
AI summary
An inference processing device uses a learned neural network to infer a feature of input data, the inference processing device including: a first storage unit that stores the input data; a second storage unit that stores a weight of the learned neural network; a data filtering unit that extracts only specific input data from pieces of the input data; and an inference operation unit that uses the specific input data extracted by the data filtering unit and the weight as inputs, performs inference operation of the learned neural network, and infers the feature of the input data.