Certifying Node for Edge ML Model Integrity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning model training architectures face privacy and security concerns due to the transmission of user data to servers, slow learning processes, and large file sizes that burden networks, leading to user reluctance in data sharing and potential data leaks.
Innovation Solution
A certifying node in a peer-to-peer network is introduced to maintain network integrity by receiving incrementally updated ML models, verifying contributions, and transmitting certified models to edge devices for incremental training, ensuring data security and preventing skewing or denial-of-service attacks through encrypted identifiers and threshold-based validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If user data is sent to server for ML model training, then model training can be performed, but privacy and security concerns arise and data may be leaked
Solution Approach 1:
Instead of sending data to the server for training, the patent inverts the architecture so that ML models are trained at the edge devices (users' devices) using local data. The trained models or model updates are then shared with the server and other users. This inversion eliminates the need to transmit sensitive user data while still enabling collaborative model training across the network.
Solution Approach 2:
The patent introduces a server as an intermediary that facilitates peer-to-peer model sharing and training coordination without requiring direct data transmission between users. The server acts as a trusted mediator that verifies model updates, manages the peer-to-peer network, and coordinates training processes while maintaining data privacy at the edge devices.
2Productivity
If ML model training is done periodically on server, then data aggregation is simple, but learning process is slow and not continuous
Solution Approach 1:
The patent implements continuous incremental training where ML models are continuously updated at edge devices as new data becomes available, rather than periodic batch training. The peer-to-peer network continuously shares model updates between users, enabling the system to learn continuously without interruption or aggregation delays.
Solution Approach 2:
The patent segments the centralized training process into distributed incremental training across multiple edge devices. Instead of aggregating all data periodically at the server, each user's device independently trains models incrementally on local data and shares updates with the peer-to-peer network, enabling parallel continuous learning across the distributed system.
3Ease of operation
If ML models are repeatedly transmitted to server, then model updates can be deployed, but network load increases due to large file sizes
Solution Approach 1:
The patent extracts only the essential model updates (such as weight changes or delta parameters) rather than transmitting complete ML models. This extraction approach significantly reduces the size of transmitted data while preserving the ability to deploy model updates, thereby reducing network bandwidth consumption for model deployment operations.
4Productivity
If peer-to-peer network allows free model updates, then incremental training is efficient, but network integrity may be compromised by malicious updates
Solution Approach 1:
The patent implements feedback mechanisms where model updates are verified through consensus among peer nodes before being accepted into the network. Each update undergoes validation checks and is cross-verified by multiple users' devices, providing feedback loops that detect and reject malicious or erroneous updates while maintaining efficient incremental training for legitimate model improvements.
Data Source
AI summary
A method and a system for maintaining network integrity for incrementally training machine learning (ML) models at edge devices is provided. The method includes registering, by a certifying node, one or more edge devices with a peer to peer network. Upon registration, an incrementally updated ML model is received from a first registered device at the certifying node. The certifying node accepts the incrementally updated ML model if a contribution of the first edge device is within a predetermined threshold, and else rejects the updated ML model if the contribution is beyond the predetermined threshold. Limiting the contribution by each edge device enables prevention of skew by any of the edge devices at the certifying node. Upon accepting the updated ML model, the certifying node certifies the updated ML model and transfers the certified ML model to one or more other edge devices in the peer to peer network.