Adaptable Filtering for Edge Deep Learning Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Implementing machine learning models on edge devices with limited resources is challenging due to processing power and memory constraints, leading to inefficiencies and the need for data to be sent to cloud networks for inference, which incurs costs and delays.
Innovation Solution
The use of adaptable filters, such as differential and smart filters, on edge devices to selectively process and filter data streams, allowing local execution of machine learning models without the need for external resources, by intelligently discarding unessential data and enabling the device to handle higher data stream rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If machine learning models are executed on edge devices with limited resources, then cloud dependency is reduced and inference speed is improved, but processing capability and memory capacity are insufficient
Solution Approach 1:
The patent segments the data stream into essential and unessential portions using filtering mechanisms. By dividing the data processing task into filtered subsets, the edge device can process only critical data points, reducing computational load while maintaining inference speed. The segmentation allows the model to operate on a subset of data rather than the complete stream.
Solution Approach 2:
The patent extracts and removes unessential data from the data stream before processing. By taking out redundant or less important data points through filtering, the system reduces the volume of data requiring processing on the edge device, thereby reducing computational requirements while preserving the essential information needed for accurate inference.
2Productivity
If data streams are processed at higher rates on edge devices, then efficiency is improved and cloud transmission is reduced, but resource utilization exceeds device capabilities
Solution Approach 1:
The patent applies partial action by processing only a filtered subset of the data stream rather than the complete stream. The filtering mechanism selectively passes only essential data points to the machine learning model, allowing the device to operate at higher effective processing rates for critical data while keeping overall resource utilization within acceptable limits.
3Measurement precision
If complete data streams are transmitted to cloud networks for inference, then processing accuracy is maintained, but transmission costs and delays increase
Solution Approach 1:
The patent extracts essential data from the complete data stream and transmits only this filtered subset to the cloud network for inference. This extraction process removes redundant information that would consume bandwidth and transmission time, while preserving the critical data needed for accurate inference, thereby reducing transmission delays without sacrificing measurement precision.
4Power
If filtering is applied to reduce data volume, then resource requirements are reduced, but information loss may occur
Solution Approach 1:
The patent applies local quality by using different filtering strategies for different portions or types of data within the stream. The filtering mechanism is designed to preserve essential information while removing unessential data, applying selective quality preservation based on the importance and characteristics of different data points rather than uniform filtering across all data.
Data Source
AI summary
Techniques for utilizing adaptable filters for edge-based deep learning models are described. Filters may be utilized by an edge electronic device to filter elements of an input data stream so that only a subset of the elements are used as inputs to a machine learning model run by the electronic device, enabling successful operation despite the input data stream potentially being generated at a higher rate than a rate in which the ML model can be executed. The filter can be a differential-type filter that generates difference representations between consecutive elements of the data stream to determine which elements are to be passed on for the ML model, a “smart” filter such as a neural network trained using outputs from the ML model allowing the filter to “learn” which elements are the most likely to be of value to be passed on, or a combination of both.