Distributed CNN for Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing convolutional neural networks (CNN) models are inefficient in processing multiple tasks on edge devices due to high energy consumption, memory requirements, and limited processing power, making it challenging to execute complex tasks or classifications that require sensor fusion.
Innovation Solution
A distributed CNN system is implemented, where activity spotting features from multiple sensors are extracted using convolutional layers and softmax layers, with the ability to generate interrupt signals to activate additional layers for activity classification, allowing for efficient processing across different sensors or hubs, including edge devices and cloud systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If CNN models are executed on a single entity (GPU or neuronal network accelerator), then processing power is sufficient for complex tasks, but energy consumption and memory requirements become excessively high
Solution Approach 1:
The patent divides the CNN model into multiple distributed instances across different entities (sensors, edge devices, cloud systems). Each instance processes a portion of the computation, allowing the system to distribute the processing load and reduce the energy consumption of any single entity while maintaining sufficient processing power for complex tasks.
Solution Approach 2:
The patent transitions from a single-entity vertical processing architecture to a multi-entity horizontal distributed architecture. By adding the dimension of spatial distribution across multiple entities, the system achieves both sufficient processing power through parallel computation and reduced energy consumption through localized processing at each node.
2Power
If CNN models are executed on a single entity, then processing capability is adequate, but memory requirements become excessively high
Solution Approach 1:
The patent segments the CNN model and its associated memory requirements across multiple distributed entities. Each entity stores and processes only the portion of the model it needs, significantly reducing the memory burden on any single entity while maintaining the full processing capability through coordinated computation across the distributed system.
3Use of energy by moving object
If CNN processing is distributed across multiple sensors and hubs, then energy consumption is reduced, but device complexity increases
Solution Approach 1:
The patent creates a universal distributed CNN framework that can be implemented across diverse entities (sensors, edge devices, cloud systems). Each entity follows the same architectural pattern and communication protocol, allowing the system to scale while managing complexity through standardized interfaces and modular design.
4Productivity
If distributed CNN processing is implemented across sensors and hubs, then multiple tasks can be handled efficiently, but system complexity increases
Solution Approach 1:
The patent segments both the CNN model and the task processing workflow across multiple entities. Each entity handles specific tasks using its portion of the distributed model, enabling efficient multi-task processing while managing complexity through clear division of labor and standardized inter-entity communication.
Data Source
AI summary
To efficiently execute deep convolutional neural networks (CNN) on edge devices (e.g., wearable device like an Apple Watch or FitBit) it may be necessary to split the output tasks across different entities. For edge devices with multiple sensors that are connected to multiple hubs, simple activity spotting may then be executed on a sensor while the hub resides in a sleep-like state. The hub may then be activated when an activity is detected by a sensor and further activity classification may then be performed. It is also contemplated that the edge device may include multiple hubs for simultaneous processing of multiple classification tasks.