Semi-Federated Learning via STAR-RIS for IoT Energy Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale wireless IoT scenarios, heterogeneous computing capabilities among devices hinder efficient model training, and limited battery resources pose challenges for prolonged network operation, leading to reduced network performance and energy inefficiency.
Innovation Solution
A semi-federated learning method based on next-generation multiple access (NGMA) technology integrates communication-centric and computing-centric approaches, utilizing a STAR-RIS to dynamically adjust channel environments and optimize power allocation, enabling devices with weak computing capabilities to participate in global model training while minimizing energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If federated learning is used to reduce communication overheads and training time, then communication efficiency is improved, but training accuracy deteriorates due to distributed characteristic
Solution Approach 1:
The patent segments users into two types: communication-centric users who upload local datasets and computing-centric users who compute and transmit model gradients. This segmentation allows each user type to contribute differently to the global model, with communication-centric users providing data diversity and computing-centric users providing computational contributions, thereby maintaining both efficiency and accuracy.
Solution Approach 2:
The patent applies local quality by allowing different users to have different roles based on their capabilities. Communication-centric users focus on data transmission while computing-centric users focus on gradient computation. This differentiation ensures that each user contributes according to their local capabilities, improving overall system efficiency without compromising accuracy.
2Adaptability or versatility
If devices with weak computing capabilities participate in federated learning, then system inclusivity is improved, but training performance deteriorates due to heterogeneous computing capabilities
Solution Approach 1:
The patent segments users based on computing capabilities into communication-centric and computing-centric types. This segmentation allows devices with weak computing capabilities to participate as communication-centric users, contributing data without being hindered by limited computational power, while devices with strong capabilities participate as computing-centric users. This maintains inclusivity while preserving training performance.
Solution Approach 2:
The patent creates a universal federated learning framework where users can perform different functions based on their capabilities. Both communication-centric and computing-centric users contribute to the global model in their respective ways, making the system universally applicable to heterogeneous devices while maintaining high training performance through diverse contributions.
3Productivity
If communication and computing operations are performed by IoT devices, then model training is enabled, but energy consumption increases due to limited battery capacity
Solution Approach 1:
The patent implements periodic action through round-based federated learning operations. Users perform communication and computing tasks in periodic rounds, uploading data or gradients at scheduled intervals rather than continuously. This periodic operation pattern reduces peak energy consumption and allows devices to manage battery resources more effectively while maintaining model training capability.
Solution Approach 2:
The patent segments tasks into communication-centric operations (data upload) and computing-centric operations (gradient computation), allowing devices to participate in model training based on their energy availability and capabilities. This segmentation enables energy-efficient participation where devices can choose to perform lighter communication tasks or more intensive computing tasks depending on their energy state.
Data Source
AI summary
A semi-federated learning (semiFL) method based on a next-generation multiple access (NGMA) technology is provided. Centralized learning (CL) and FL are integrated such that devices with weak computing capabilities can also participate in training of a global model. A simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) is deployed to dynamically change a channel environment such that a system can meet different task requirements of heterogeneous users. Communication-centric CL users and computing-centric FL users can transmit data in parallel on a same time-frequency resource. This avoids a waste of data resources, enriches data obtaining of a base station (BS), and improves accuracy of the global model. The semiFL method also integrates a strategy for jointly optimizing user power allocation and a configuration of the STAR-RIS to reduce total uplink transmit power consumption of the system and prolong a life cycle of an intelligent Internet of Things (IoT) network.