Federated Learning Knowledge Augmentation for IoT
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing federated learning methods struggle to generate models optimized for individual devices' data distributions, especially when low-end IoT devices are involved, leading to limited performance due to model capacity constraints.
Innovation Solution
A knowledge-augmenting method and apparatus that uses federated learning information to combine local information from individual devices, including low-end devices, with server-based global models, generating a true label approximate value for learning a large model without exposing private data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If personalized federated learning is used to optimize models for individual devices, then model performance for local data is improved, but device complexity and computational requirements increase beyond the capacity of low-end IoT devices
Solution Approach 1:
The model is segmented into two parts: global parameters that are trained through federated learning and transmitted to devices, and local parameters that are trained locally on low-end devices. This segmentation allows the complex personalized model to be distributed across server and device, reducing the computational burden on individual low-end devices while maintaining personalized performance.
Solution Approach 2:
The server acts as an intermediary that collects local parameters from devices, combines them with global parameters, and generates updated global parameters that are then transmitted back to devices. This intermediary approach enables low-end devices to benefit from personalized models without needing to perform complex computations locally.
2Productivity
If local data is shared with the server for model training, then model learning capability is improved, but data security and privacy are compromised
Solution Approach 1:
The invention extracts only the necessary model parameters (global and local parameters) from the training process and transmits them between server and devices, while the actual sensitive local data remains on-device. This extraction approach enables model training without exposing private data, maintaining both learning capability and data security.
Solution Approach 2:
The server acts as an intermediary that processes model parameters rather than raw data. It receives local parameters from devices, combines them with global parameters to generate updated global parameters, and transmits only these parameter updates back to devices. This intermediary parameter-based communication enables collaborative training while preserving data privacy.
3Object-affected harmful factors
If only model parameters are transmitted in federated learning, then data security is maintained, but the server lacks sufficient information to generate optimized global models for diverse data distributions
Solution Approach 1:
The parameter transmission is segmented into two types: global parameters that capture general patterns from multiple devices and local parameters that capture device-specific characteristics. This segmentation allows the server to receive differentiated information that reflects diverse data distributions without transmitting raw data, enabling optimized global model generation while maintaining security.
Solution Approach 2:
The invention applies local quality by allowing different types of parameters to be transmitted and processed differently. Local parameters from devices with diverse data distributions are combined with global parameters in a way that preserves the unique characteristics of each device's data distribution, enabling the server to generate globally optimized models that adapt to local variations.
Data Source
AI summary
A method and apparatus for augmenting knowledge using federated learning information. An apparatus for augmenting knowledge using federated learning information comprises a transceiver unit that receives local information including a global parameter of a local model, a local latent vector, and a local loss value from each of a plurality of individual devices, a data storage unit that stores the local information, a federated learning execution unit that collects a global parameter of the local model and generates a federated global parameter for a global model, and a large model learning unit that generates a true label approximate value for learning a large model using the local information and the federated global parameter, and learns the large model using a prediction result obtained by inputting the local latent vector into the large model and the true label approximate value.