Privacy-Preserving ML Classifiers via Aggregated Statistics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in developing customized machine learning (ML) classifiers that balance data privacy and effective model performance, particularly when proprietary data is used for training, as it often contains sensitive information and intellectual property.
Innovation Solution
The proposed solution involves generating ML classifiers using aggregated statistics objects, such as bin frequencies and covariances, which are calculated for each predictive feature conditioned on a target value. These aggregated statistics are anonymized and difficult to re-identify, allowing for classifier training without compromising data privacy or intellectual property.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If proprietary data is shared with external classifier developers for training, then model performance is improved, but data privacy and security are compromised
Solution Approach 1:
The patent introduces an intermediary mechanism where only aggregated statistics (bin frequencies, covariances) are shared with external developers, not the raw proprietary data. This intermediary layer preserves data privacy while enabling effective model training through the use of Generalized Additive Models that can learn from aggregated statistical patterns.
Solution Approach 2:
The patent extracts only the necessary statistical information (bin frequencies and covariances) from the proprietary data, separating the essential training signals from the sensitive raw data. This extraction allows model development while leaving the proprietary data secure within the organization's firewall.
2Measurement precision
If disaggregated data records are used for training, then classifier accuracy is improved, but privacy vulnerabilities increase
Solution Approach 1:
The patent uses aggregated statistics as an intermediary that preserves the essential patterns needed for accurate classification while eliminating direct privacy vulnerabilities. The bin frequencies and covariances capture statistical relationships without exposing individual records.
3Productivity
If high-fidelity synthetic datasets are shared, then model training capability is improved, but intellectual property disclosure risk increases
Solution Approach 1:
The patent extracts only the statistical essence (bin frequencies and covariances) needed for model training, removing all extraneous information that could constitute intellectual property. This extraction provides sufficient training capability while minimizing IP disclosure.
4Object-affected harmful factors
If proprietary data is kept private, then data security is improved, but access for model development is reduced
Solution Approach 1:
The patent creates an intermediary data representation (aggregated statistics) that maintains security while enabling access. The bin frequencies and covariances serve as a secure interface that allows model developers to work with protected data.
Data Source
AI summary
A computer-implemented method for generating a classifier, comprising: receiving aggregated statistics objects, wherein the aggregated statistics objects comprise bin frequencies F0 and F1, wherein the bin frequencies F0 and F1 are calculated for each of a plurality of predictive features, conditioned on a target value being 0 or 1, respectively; bin-level covariances C0 and C1, wherein the bin-level covariances C0 and C1 are calculated for each pair of bins, conditioned on the target value being 0 or 1, respectively; feeding the aggregate statistics objects F0, F1, C0, and C1 into the classifier, wherein the classifier generates a score calculated as a sum of a plurality of flexible nonlinear shape functions applied to the plurality of predictive features, respectively; and training the classifier by fitting the plurality of shape functions to maximize a divergence for score separation between target value 0 and target value 1.