Feature-Specific Differential Privacy for Accurate ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models trained with sensitive data are vulnerable to membership inference attacks, as existing differential privacy techniques apply uniform privacy budgets across all features, leading to reduced accuracy and inefficiency.
Innovation Solution
Implementing differential privacy by feature, where different privacy budgets are assigned to varying levels of sensitivity of data features, enhancing model accuracy by applying noise based on feature-specific privacy budgets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If uniform differential privacy is applied across all features, then privacy protection is maintained, but model accuracy decreases
Solution Approach 1:
The patent applies differential privacy differently to different features based on their sensitivity levels. Each feature is assigned a privacy budget proportional to its sensitivity, allowing high-sensitivity features to receive stronger privacy protection while low-sensitivity features receive less protection, thereby maintaining overall model accuracy while preserving privacy for critical data.
2Reliability
If higher privacy budgets are allocated to sensitive features, then privacy protection improves, but computational efficiency decreases
Solution Approach 1:
Instead of applying uniform privacy protection across all features, the system allocates privacy budgets locally based on feature sensitivity. This ensures that computational resources are not wasted on low-sensitivity features while maintaining adequate protection for high-sensitivity features, optimizing the balance between privacy and computational efficiency.
3Reliability
If differential privacy noise is applied to all features equally, then privacy is protected, but information utility is reduced
Solution Approach 1:
The patent implements feature-sensitive differential privacy where the amount of noise added to each feature is proportional to its sensitivity. This preserves information utility for low-sensitivity features while maintaining privacy protection for high-sensitivity features, thereby reducing overall information loss compared to uniform noise application.
Data Source
AI summary
Various examples are directed to systems and methods for using a machine learning model. A computing system may access training data comprising a plurality of training data items. Each of the plurality of training data items may comprise a plurality of features. From a first training data item of the plurality of training data items, the computing system may generate a first transformed training data item using a first privacy budget corresponding to a first portion of the first training data item and a second privacy budget corresponding to a second portion of the first training data item. The computing system may train a machine learning model using the first transformed training data item and use the trained machine learning model to generate at least one class probability for a data item.