Privacy Preserving Transfer Learning for User Data Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning technologies face challenges in predicting user interests and attributes without compromising user privacy, especially with the decline of third-party cookies, which limits data collection and inference capabilities.
Innovation Solution
The use of transfer learning techniques to train machine learning models based on aggregated data, applying k-anonymity and differential privacy to ensure user data privacy, allowing for prediction of user interests and attributes without accessing individual user data, and leveraging richer data from known users to provide relevant content to unknown users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If third-party cookies are used for data collection, then user data can be collected and analyzed, but user privacy is compromised
Solution Approach 1:
The patent extracts only the essential aggregated features from user data without collecting or storing individual user information. The system takes out and processes only the necessary aggregated statistics (e.g., feature counts, histograms) while leaving individual user data behind, thus enabling data collection while preserving privacy.
Solution Approach 2:
The patent introduces an intermediary aggregation process that acts as a mediator between raw user data and the machine learning model. This intermediary layer processes data locally on client devices and transmits only aggregated results to the server, preventing direct access to individual user data while still enabling model training.
2Object-affected harmful factors
If aggregated data is used for model training, then user privacy is protected, but data richness and model training quality may be reduced
Solution Approach 1:
The patent merges data from multiple users into aggregated statistics (e.g., combining feature counts, histograms, and distributions across many users). This merging process preserves privacy by eliminating individual identifiers while retaining rich statistical patterns that are useful for machine learning model training.
Solution Approach 2:
The patent transforms individual user data into different parameter representations (aggregated statistics such as mean, variance, histograms, and feature counts). This parameter transformation maintains the essential information needed for model training while removing personally identifiable information, thus preserving both privacy and data richness.
3Measurement precision
If more user data is collected for better predictions, then prediction accuracy improves, but bandwidth usage and data storage requirements increase
Solution Approach 1:
The patent applies partial action by collecting and transmitting only the necessary aggregated features rather than complete user data sets. The system transmits a subset of processed statistical information (e.g., feature histograms and counts) that is sufficient for model training, reducing bandwidth usage while maintaining prediction accuracy.
4Measurement precision
If complex machine learning models are deployed, then prediction capability improves, but computational resources and model training time increase
Solution Approach 1:
The patent performs preliminary data aggregation and processing on client devices before transmitting data to the server. By pre-computing aggregated statistics (feature counts, histograms, distributions) locally, the system reduces the computational burden on the server and enables the use of more complex models without proportionally increasing overall system complexity.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training and using machine learning models to predict data in privacy preserving manners are described. In one aspect, a method includes receiving, from a client device of a user, a digital component request including one or more contextual signals that describe an environment in which a selected digital component will be presented. The contextual signals are provided as input to a trained machine learning model that is trained to output, based on input contextual signals, predicted data about the user. The trained machine learning model is trained using a set of aggregated data including, for each of a set of aggregation keys, aggregated data for a plurality of users having electronic resource views that match the aggregation key. The predicted data about the user is received as an output of the trained machine learning model.