Neural Network Deployment in Resource-Constrained Environments

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Deploying neural networks in resource-constrained environments, such as automobiles, is challenging due to the need for significant processing power and the presence of intermittent communications, which limits their application in embedded devices with minimal processing capabilities and variable telecommunications infrastructure.

Innovation Solution

A system that includes a sensor in the resource-constrained environment generating data, a first computing device outside the environment producing a neural network structure based on this data, and a second computing device within the environment using the neural network to determine the environment's state, with a transceiver for wireless communication, enabling the use of neural networks in environments with limited resources and intermittent connectivity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If neural networks are deployed in resource-constrained environments, then monitoring and control capabilities are improved, but processing power requirements and communication infrastructure demands worsen

Engineering Contradiction:

Improvemonitoring and control capabilityVSAvoidprocessing power requirement

Core Design Contradiction:

ProductivityVSPower

Solution Approach 1:

The system divides the neural network processing into two segments: training is performed on external computing devices with sufficient processing power, while only the trained model (weights and biases) is deployed to the resource-constrained embedded device. This segmentation allows the embedded device to perform inference without requiring high processing power for training operations.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The neural network is trained in advance on external computing devices before deployment. The training process, which requires significant computational resources, is completed beforehand, allowing the embedded device to only perform lightweight inference operations using the pre-trained model parameters.

Inventive Principle:

Principle #10Preliminary action

2Productivity

If neural networks are deployed in resource-constrained environments, then monitoring and control capabilities are improved, but communication infrastructure requirements worsen

Engineering Contradiction:

Improvemonitoring and control capabilityVSAvoidcommunication infrastructure requirement

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The complex communication requirements are extracted from the embedded device and handled externally. The system downloads trained model parameters from external computing devices and uploads sensor data for inference, minimizing the communication burden on the resource-constrained embedded device while maintaining monitoring and control capabilities.

Inventive Principle:

Principle #2Taking out (Extraction)

3Speed

If neural networks are trained locally in resource-constrained environments, then real-time processing is improved, but device resource consumption worsens

Engineering Contradiction:

Improvereal-time processing speedVSAvoiddevice energy consumption

Core Design Contradiction:

SpeedVSUse of energy by moving object

Solution Approach 1:

The computational workload is segmented between external computing devices (training) and embedded devices (inference). This allows real-time processing at the embedded device level while energy-intensive training operations are performed externally, maintaining real-time responsiveness without excessive energy consumption.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The energy-intensive training process is performed in advance on external computing devices. The embedded device only needs to perform lightweight inference using pre-trained models, achieving real-time processing speed while minimizing energy consumption during operation.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20230023870A1Neural network applications in resource constrained environments

Publication Date: 2023.01.26 ALPINE ELECTRONICS OF SILICON VALLEY INC

AI summary

Systems and methods are disclosed for applying neural networks in resource-constrained environments. A system may include a sensor located in a resource-constrained environment configured to generate first sensor data and second sensor data of the resource-constrained environment. The system may also include a first computing device not located in the resource-constrained environment configured to produce a neural network structure based on the first sensor data. The system may also include a second computing device configured to determine a state of the resource-constrained environment based on input of the second sensor data to the neural network structure. The system may also include a controller located in the resource-constrained environment configured to control a device in the resource-constrained environment based on the state of the resource-constrained environment determined by the second computing device. The second computing device may be further configured to calculate an activation area for the neural network structure.