Adversarial Privacy-Preserving Representation Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques fail to adequately preserve user privacy while maintaining predictive power, as naive anonymization strategies compromise model performance.
Innovation Solution
A system and method for training a machine learning model to generate privacy-preserving representations by maximizing entropy among similar private attributes and maintaining separation between private and non-private attributes, using an adversarial training approach with a variational autoencoder to create a target portion for non-private attributes and a residual portion for private attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If naive anonymization strategies are used to preserve user privacy, then user privacy is protected, but model performance is compromised
Solution Approach 1:
The patent segments user data into private attributes and non-private attributes, processing them through separate pathways in the machine learning model. This segmentation allows the model to preserve privacy-sensitive information while maintaining access to task-relevant information, thereby protecting user privacy without compromising model performance on downstream tasks.
Solution Approach 2:
The patent introduces an intermediary representation layer that mediates between private and non-private attributes. This intermediary representation preserves the essential information needed for machine learning tasks while removing or obscuring private attributes, thus acting as a bridge that maintains both privacy protection and model effectiveness.
2Productivity
If all attributes are preserved in the representation, then model predictive power is maximized, but private attribute information is revealed
Solution Approach 1:
The patent applies local quality by treating different attributes of user data differently based on their privacy sensitivity and task relevance. Non-private attributes are preserved in full detail to maintain predictive power, while private attributes are transformed or obscured to protect privacy. This localized differentiation allows the model to optimize for both performance and privacy protection in respective domains.
Data Source
AI summary
A first machine learning model may be trained to generate a first representation of a first user data having private attributes and non-private attributes. The first representation may be generated to achieve a focal entropy by maximizing an entropy amongst similar private attributes. The first representation to preserve information associated with the non-private attributes but omit information associated with the private attributes. Moreover, the first user data may be classified based on a target portion of the first representation including the non-private attributes but not the residual portion of the first representation including the private attributes. The trained first machine learning model may be applied to generate a second representation of a second user data such that downstream tasks may be performed by applying a second machine learning model to the second representation of the second user data. Related systems and computer program products are also provided.