Differential Privacy Machine Learning for Streaming Security Threat Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Detecting and protecting against computer security threats while preserving the privacy of individual client devices is challenging, as existing methods may infringe on customer privacy by collecting sensitive data, and using perturbed data for machine learning may reduce the accuracy of threat detection.
Innovation Solution
A computer-implemented method using differential privacy machine learning where a remote server receives actual data values, generates perturbed values by adding noise, stores them without the actual values, trains a machine learning classifier, and uses actual values as input for more accurate threat identification and remedial actions, preserving privacy and enhancing detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If actual data values are collected and stored from client devices for security threat detection, then detection accuracy is improved, but client device privacy is compromised
Solution Approach 1:
A remote server acts as an intermediary between client devices and the security analysis system. The server receives actual data values from clients, generates perturbed data values by adding noise, stores only the perturbed values, and uses them for training machine learning classifiers. This intermediary approach allows the system to maintain detection accuracy while protecting client privacy by preventing direct access to actual data values.
Solution Approach 2:
The system transforms the data parameter from its original state to a perturbed state by adding noise. The perturbed data values retain sufficient statistical properties for effective security threat detection while making it impossible to reconstruct individual client data. This parameter transformation resolves the contradiction by maintaining utility for detection while eliminating privacy risks.
2Loss of information
If perturbed data values are used for machine learning training, then client device privacy is preserved, but threat detection accuracy deteriorates
Solution Approach 1:
The system performs preliminary actions by training multiple machine learning classifiers using different perturbed data values before actual threat detection. This pre-training phase allows the classifiers to learn from diverse perturbed datasets, improving their robustness and accuracy. When actual detection is needed, the system can use ensembles of pre-trained classifiers, maintaining detection accuracy while preserving privacy.
3Reliability
If actual data values are stored at the remote server, then threat detection capability is improved, but data exposure risk increases
Solution Approach 1:
The system converts the potential harm of data storage into a benefit by deliberately adding noise to create perturbed data values. This noise addition, which initially seems to degrade data quality, actually protects against data exposure risks while maintaining sufficient quality for threat detection. The harm of potential data exposure is transformed into the benefit of enhanced privacy protection.
Data Source
AI summary
Identifying and protecting against a computer security threat while preserving privacy of individual client devices using differential privacy machine learning for streaming data. In some embodiments, a method may include receiving first actual data values streamed from one or more first local client devices, generating first perturbed data values by adding noise to the first actual data values using a differential privacy mechanism, storing the first perturbed data values, training a machine learning classifier using the first perturbed data values, receiving a second actual data value streamed from a second local client device, generating a second perturbed data value by adding noise to the second actual data value, storing the second perturbed data value, identifying a computer security threat to the second local client device using the second actual data value as input to the trained machine learning classifier, and protecting against the computer security threat.