Naive Bayes Classifier Parameter Adaptation for Privacy Compliance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification models, such as Naive Bayes classifiers, face challenges in ensuring independence from specific training instances, particularly in compliance with privacy regulations like GDPR, where adding noise to parameters to achieve independence significantly decreases accuracy and may not sufficiently address dependence on multiple training records.
Innovation Solution
A method to make Naive Bayes-type classifiers independent from specific undesired training instances by adapting their parameters and prior probabilities without accessing the original training dataset, allowing for optimal accuracy and compliance with privacy regulations by handling removal requests efficiently and securely.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If noise is added to classifier parameters to achieve independence from training instances, then privacy compliance is improved, but classification accuracy deteriorates
Solution Approach 1:
The patent extracts and removes specific training instances from the training dataset upon receiving removal requests, rather than adding noise to all parameters. This selective extraction allows the classifier to maintain independence from specific instances while preserving accuracy on the remaining data.
Solution Approach 2:
The patent changes the parameters of the classifier by retraining or adapting them after removing specific training instances. This parameter adjustment allows the classifier to maintain optimal performance on the remaining dataset while achieving independence from removed instances, avoiding the accuracy loss associated with noise addition.
2Reliability
If noise is added to make the classifier independent from multiple training records, then privacy compliance is improved, but classification accuracy significantly deteriorates
Solution Approach 1:
The patent removes specific training records from the training dataset upon receiving removal requests, enabling the classifier to be independent from those specific records without adding noise that would degrade accuracy across all records.
Solution Approach 2:
The patent implements a dynamic approach where the training dataset and classifier parameters can be updated in response to removal requests. This allows the classifier to adapt its independence requirements dynamically rather than statically adding noise, maintaining accuracy while achieving independence when needed.
3Adaptability or versatility
If the classifier is made independent from specific training instances by adding noise, then privacy control is improved, but model utility deteriorates
Solution Approach 1:
The patent extracts and removes specific training instances from the dataset upon receiving privacy control requests, allowing users to control their data usage without degrading overall model utility through noise addition.
Solution Approach 2:
The patent adjusts classifier parameters by retraining or adapting them after removing specific instances, maintaining model utility on the remaining data while providing users with privacy control over their specific contributions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a system (100) for processing a classifier. The classifier is a Naïve Bayes-type classifier classifying an input instance into multiple classes based on multiple continuous probability distributions of respective features of the input instance and based on prior probabilities of the multiple classes. Upon receiving a removal request message identifying one or more undesired training instances, the classifier is made independent from one or more undesired training instances. To this end, for a continuous probability distribution of a feature, adapted parameters of said probability distribution are computed based on current parameters of said probability distribution and the one or more undesired training instances. Further, an adapted prior probability of a class is computed based on a current prior probability of the class and the one or more undesired training instances.