Federated Learning Image Sensor with Neuroevolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image sensor technologies face high communication costs and lack interpretability in deep learning models, particularly in federated learning frameworks, and require adaptable neural architectures for efficient image recognition and analysis.
Innovation Solution
An adaptive federated learning framework that employs a lightweight convolutional fuzzy rough neural network, optimized through neuroevolution, which reduces communication burden and enhances model interpretability by using a fuzzification layer, rough layer, and output layer, and initializes parameters to facilitate efficient training and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep CNN models are used for image recognition in federated learning, then recognition accuracy is improved, but communication cost increases due to frequent uploads of model parameters to cloud server
Solution Approach 1:
The system segments the federated learning process into selective updates based on performance metrics. Instead of all clients uploading parameters every epoch, only clients meeting specific improvement thresholds or showing stagnation are selected for upload, reducing overall communication frequency while maintaining model quality.
Solution Approach 2:
The patent changes the parameter update strategy by introducing dynamic selection criteria based on training improvement frequency, stagnation frequency, and accuracy increase degree. This adaptive parameter update mechanism reduces unnecessary communications while preserving the benefits of deep CNN models for accurate recognition.
2Measurement precision
If deep CNN models are used for image recognition, then recognition performance is improved, but model interpretability deteriorates as the model becomes a black box
Solution Approach 1:
The patent introduces heat maps as an intermediary visualization tool that bridges the gap between the black box CNN model and human understanding. These heat maps provide visual feedback about which regions of input images contribute most to model decisions, making the model's reasoning process interpretable without changing the underlying high-performance CNN architecture.
Solution Approach 2:
The system implements feedback mechanisms where heat maps are generated and supplied to facilitate decision-making. This feedback loop allows users to understand model behavior and trust decisions while maintaining the performance benefits of deep CNN models.
3Ease of manufacture
If fixed neural network architecture is used, then model training is simplified, but adaptability to multiple practical application needs deteriorates
Solution Approach 1:
The patent implements dynamic architecture selection where different neural network configurations can be chosen based on specific application requirements. The system can adapt between lightweight models for resource-constrained devices and more complex models for applications requiring higher accuracy, making the federated learning system versatile across multiple practical scenarios.
Data Source
AI summary
The present invention discloses an intelligent image sensing device for sensing-computing-cloud integration based on a federated learning framework. The device comprises: intelligent image sensors, edge servers and a remote cloud, wherein the intelligent image sensor is used for perceiving and generating images, and uploading the images to the edge server; the edge server is used as a client; the remote cloud is used as a server; the clients train a convolutional fuzzy rough neural network based on the received images and the proposed federated learning framework; and the intelligent image sensors download the weight parameters of the trained convolutional fuzzy rough neural network from the clients, and classify and recognize the images based on the trained weight parameters. The present invention searches a lightweight deep learning architecture through neuroevolution, and deploys the lightweight deep learning architecture in the image sensors to automatically discriminate and analyze the perceived images.