Hybrid Cloud-Edge Pattern Recognition for Custom Wake Words
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual assistant systems on edge devices are limited in customizing pattern recognition functions due to hardware limitations, restricting users from defining specific wake patterns beyond pre-programmed settings, which hampers flexibility and user-specific needs.
Innovation Solution
A hybrid cloud-chip approach that enables user-defined pattern recognition on edge devices through a pattern recognition integrated circuit and a remote training service, allowing users to train and update machine learning models on a secondary device and transfer trained weights to the edge device for customized pattern recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If pattern recognition is implemented locally on edge devices with limited processing capabilities, then response time is improved, but adaptability deteriorates due to hardware limitations preventing user customization
Solution Approach 1:
The system divides pattern recognition into two segments: simple wake word detection performed locally on the edge device for fast response, and complex user-defined pattern training performed remotely in the cloud. This segmentation allows each component to operate within its capability constraints while achieving overall system goals.
Solution Approach 2:
A remote training service acts as an intermediary between the user and the edge device. The service receives training data from the user's device, processes it remotely to generate optimized model weights, and transfers them back to the edge device. This intermediary enables customization without burdening the limited edge hardware.
2Adaptability or versatility
If complex machine learning models are trained locally on edge devices, then adaptability is improved, but device complexity increases beyond hardware capabilities
Solution Approach 1:
The complex training functionality is extracted from the edge device and placed in a remote training service. The edge device retains only the lightweight inference capability, while the heavy lifting of model training and weight optimization is performed remotely where computational resources are abundant.
Solution Approach 2:
The system moves the training dimension to a different computational environment (cloud-based remote service) rather than attempting to fit it into the constrained edge device dimension. This dimensional shift allows complex operations to be performed without increasing edge device complexity.
3Ease of operation
If pre-programmed wake words are used on edge devices, then ease of operation is improved, but adaptability deteriorates as users cannot define custom patterns
Solution Approach 1:
The system transitions from a static, pre-programmed wake word approach to a dynamic, user-configurable system. Users can define and update their own wake patterns through the remote training service, allowing the system to adapt to individual preferences while maintaining the simplicity of local execution.
Data Source
AI summary
A system for configuring user-defined recognition patterns at an edge device using a hybrid cloud-edge device approach has a pattern recognition integrated circuit implementing a machine learning pattern recognizer that generates an event recognition output in response to an input thereto based upon pre-trained machine learning weights stored in a memory of the pattern recognition integrated circuit. A remote pattern recognition training service is in communication with a secondary user device receptive to a training input of the user-defined recognition patterns, and returns a set of training weights corresponding to the training input. An application interface connects the pattern recognition integrated circuit to the secondary user device, with the set of training weights returned to the secondary user device being transferable to the machine learning pattern recognizer for storage in the memory of the pattern recognition integrated circuit.