Federated Learning for Private Sponsored Content
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for providing sponsored content in networked environments face challenges in effectively targeting content while preserving user privacy, as they often rely on centralized data processing that may compromise user data security and privacy.
Innovation Solution
A system utilizing machine intelligence models with differential and homomorphic encryption techniques, deployed on mobile devices, to select and present sponsored content locally while ensuring privacy through mechanisms like local differential privacy and homomorphic encryption, which processes data on-device and encrypts sensitive information, ensuring that only encrypted data is transmitted.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized data processing is used for sponsored content targeting, then content targeting effectiveness is improved, but user data privacy and security are compromised
Solution Approach 1:
The system divides the centralized data processing architecture into distributed edge computing nodes deployed on user devices. Each device runs local machine intelligence models that process data locally, segmenting the monolithic centralized system into autonomous distributed units that can operate independently while maintaining collaborative learning through encrypted parameter updates.
Solution Approach 2:
Homomorphic encryption serves as an intermediary layer between user data and the machine intelligence model training process. This cryptographic mediator enables the system to process encrypted data without decryption, allowing centralized coordination of model updates while preserving user privacy. The encrypted parameters act as intermediaries that carry information necessary for model improvement without exposing sensitive user information.
2Measurement precision
If user data is collected and transmitted for model training, then model accuracy is improved, but data security and privacy protection are weakened
Solution Approach 1:
User devices perform preliminary data processing and feature extraction locally before transmitting any information to the server. The machine intelligence models are trained incrementally using locally computed gradients and encrypted parameter updates, allowing the central model to improve accuracy without ever receiving raw user data. This preliminary local processing ensures data security is maintained while still enabling model learning.
Solution Approach 2:
The system transforms the training paradigm from data-centric to parameter-centric learning. Instead of collecting and processing user data centrally, the system collects encrypted model parameter updates from distributed devices. The mathematical properties of homomorphic encryption allow these parameter changes to be aggregated and processed to improve model accuracy while the parameters themselves remain encrypted, preserving data security throughout the process.
3Object-affected harmful factors
If data is encrypted and processed locally, then user privacy is protected, but system complexity increases
Solution Approach 1:
User devices autonomously perform local model training and encrypted parameter update generation without requiring complex centralized coordination for each data processing operation. The devices self-manage their local machine intelligence models, automatically computing gradients from local data and generating encrypted updates that are transmitted to the server. This self-service approach distributes computational complexity to edge devices while simplifying the central server's role to aggregation and model distribution.
4Productivity
If centralized processing is used, then model training efficiency is improved, but user privacy control is reduced
Solution Approach 1:
The system implements dynamic privacy control where users can adjust their privacy settings and control the degree of local vs. centralized processing. The federated learning architecture allows flexible configuration of model update frequencies, encryption parameters, and local processing intensity, enabling users to dynamically balance privacy control with model training efficiency based on their preferences and device capabilities.
Data Source
AI summary
Systems and methods are shown for providing private local sponsored content selection and improving intelligence models through distribution among mobile devices. This allows greater data gathering capabilities through the use of the sensors of the mobile devices as well as data stored on data storage components of the mobile devices to create predicted models while offering better opportunities to preserve privacy. Locally stored profiles comprising machine intelligence models may also be used to determine the relevance of the data gathered and in improving an aggregated model for identifying the relevance of data and the selection of sponsored content items. Distributed optimization is used in conjunction with privacy techniques to create the improved machine intelligence models. Publishers may also benefit from the improved privacy by protecting the statistics of type or volume of sponsored content items shown with publisher content.