On-Device Multi-Party Attribution With Differential Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in securely combining multi-party attribution data from different sources on resource-constrained devices while ensuring data privacy, particularly in scenarios where access to certain data is restricted, hindering effective machine learning model training.
Innovation Solution
The solution involves obfuscating access-restricted event data using differential privacy algorithms on user devices, compressing data into bit formats, and adding noise to event labels, allowing secure transmission and use in machine learning model training without revealing individual identities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multi-party attribution data is combined from different sources on resource-constrained devices, then machine learning model training is enabled, but data privacy is compromised
Solution Approach 1:
The patent introduces a privacy-preserving intermediary layer between data sources and machine learning models. Differential privacy algorithms act as mediators that transform raw attribution data into privacy-protected representations, enabling model training without direct access to individual user data. This intermediary mechanism allows the system to achieve both productivity (model training) and reliability (data privacy protection).
Solution Approach 2:
The patent applies parameter changes by transforming data characteristics through differential privacy mechanisms. The system modifies data parameters (adding noise, perturbation) to change the statistical properties of attribution data while preserving useful patterns for model training. This parameter transformation enables the same data to serve both privacy protection and machine learning objectives simultaneously.
2Reliability
If access-restricted event data is obfuscated using differential privacy algorithms, then data privacy is protected, but data quality for training is degraded
Solution Approach 1:
The patent applies partial obfuscation strategies where differential privacy algorithms are selectively applied to specific data portions or events rather than uniformly to all data. This partial action approach protects privacy for access-restricted data while maintaining quality for public or less-sensitive data, thereby balancing privacy protection with training data quality.
Solution Approach 2:
The system incorporates feedback mechanisms that monitor the impact of privacy protection on model training quality. Based on this feedback, the system dynamically adjusts the level of obfuscation and noise addition to achieve optimal balance between privacy protection and data quality, ensuring that privacy measures do not excessively degrade training effectiveness.
3Productivity
If event data is compressed into bit formats, then data transmission efficiency is improved, but information loss occurs
Solution Approach 1:
The patent extracts only the essential information needed for machine learning training from the original event data during compression. By taking out and retaining only critical features and patterns while discarding redundant information, the system achieves efficient data transmission without significant information loss for the intended training purpose.
Data Source
AI summary
Embodiments of the disclosed technologies receive first event data associated with a first party application, receive second event data representing a click, in the first party application, on a link to a third party application, receive third event data from the third party application, convert the third event data to a label, map a compressed format of the labeled third event data to the first event data and the second event data to create multi-party attribution data, group multiple instances of the multi-party attribution data into a batch, add noise to the compressed format of the labeled third event data in the batch, and send the noisy batch to a second computing device. A debiasing algorithm can be applied to the noisy batch. The debiased noisy batch can be used to train at least one machine learning model.