Federated Recommender System for Engineering Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommender systems for engineering design processes face challenges due to data privacy concerns, which prevent centralized training of models across multiple users, leading to poor user experience and inefficient component selection, as they rely on collaborative filtering without personalized data.
Innovation Solution
A global shared recommender system using a graph neural network architecture is implemented, where users receive initial parameters from a central server, allowing for user-specific training with local data, and gradient information is shared to update the model without exposing personal data, enabling personalized recommendations while maintaining privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If collaborative filtering is used for recommender system, then data privacy is protected, but recommendation quality and personalization deteriorate
Solution Approach 1:
The system segments the training process into local federated training at each user's device and global aggregation at the server. This segmentation allows personalization through local training while protecting privacy by never sharing raw data, only model updates. The recommender system is divided into local models trained on individual user data and a global model that aggregates these local learnings.
Solution Approach 2:
The server acts as an intermediary that coordinates federated learning between users without accessing their raw data. It aggregates model updates from multiple users and distributes improved global models back to users, enabling collective learning while maintaining data privacy. The intermediary facilitates knowledge transfer without direct data sharing.
2Measurement precision
If individual user data is collected for personalization, then recommendation accuracy improves, but data privacy concerns increase
Solution Approach 1:
Instead of sharing raw user data, the system shares copies in the form of model weight updates and gradients. These copies contain learned patterns and preferences without revealing actual user data. The server receives model updates as copies that can be aggregated to improve the global model without exposing sensitive information.
Solution Approach 2:
The system transforms raw user data into model parameters (weights and gradients) that can be shared safely. By changing the representation from raw data to aggregated model parameters, the system enables personalization while reducing privacy risks. The parameters capture user preferences without containing identifiable information.
3Productivity
If centralized training data aggregation is performed, then model training efficiency improves, but user privacy protection deteriorates
Solution Approach 1:
The training process is segmented into distributed local training sessions at user devices and centralized aggregation at the server. Each user trains locally on their own data, maintaining privacy, while the server aggregates results efficiently. This segmentation achieves both privacy protection and training efficiency by parallelizing the training process across multiple devices.
Solution Approach 2:
The system implements feedback loops where users train local models, send updates to the server, receive aggregated global models, and use them for improved recommendations. This feedback mechanism enables continuous model improvement through collective learning while maintaining privacy through iterative updates rather than centralized data collection.
Data Source
AI summary
A computer implemented method for providing a recommender system for a design process of a complex system is provided, wherein the recommender system is shared by a plurality of users, wherein the complex system includes a plurality of connectable components and is designed in a design process by a sequence of design steps wherein in each design step a partial design is created until a completed design is obtained, wherein a partial design of one step and a partial design of a subsequent step differ in a design difference reflecting a difference in at least one element including a component or/and connection of the components, and wherein the shared recommender system provides at each design step a prediction of the subsequent design difference.