Search⌘ K
AI Features
Home
Courses
Data Science and Machine Learning Interview Handbook

Data Privacy and Security in ML

Explore key privacy preservation techniques in machine learning, including data anonymization and differential privacy. Understand how these methods protect individual data in sensitive datasets, and learn to balance data utility with privacy requirements during model development.

We'll cover the following...

Privacy in machine learning ensures that individual data contributions cannot be reverse-engineered, leaked, or exploited. This lesson introduces the core principles of data privacy in ML and demonstrates how to apply two foundational techniques: differential privacy and data anonymization.

Privacy preservation techniques for ML

You're building a machine learning model using user health records. Regulations like GDPR and HIPAA require that the model should not leak private information about any individual. You're tasked with applying privacy-preserving methods to protect this sensitive data during model development.

Explain the key principles of privacy preservation in machine learning and demonstrate how you would implement two fundamental techniques to protect individual data privacy.

Sample answer

This question is designed to test your understanding of privacy, security, coding skills, and awareness of privacy-utility trade-offs. Let’s look at a sample approach towards this:

Privacy preservation in machine learning is crucial for protecting individual identities while maintaining the utility of data for analysis. The two primary techniques ...

Python 3.10.4
import hashlib

import random

import numpy as np



class PrivacyPreservingML:

    """

    Demonstrates basic principles of privacy preservation

    in machine learning data handling.

    """

    

    @staticmethod

    def anonymize_data(personal_data):

        """

        Anonymize personal data using hashing.

        

        Args:

            personal_data (list): List of personal identifiers

        

        Returns:

            list: Anonymized identifiers

        """

        return [hashlib.sha256(str(item).encode()).hexdigest() for item in personal_data]

    

    @staticmethod

    def differential_privacy_noise(data, epsilon=1.0):

        """

        Add controlled noise to data to prevent individual identification.

        

        Args:

            data (list): Numerical data

            epsilon (float): Privacy budget (lower = more privacy)

        

        Returns:

            list: Data with added differential privacy noise

        """

        def laplace_mechanism(value, sensitivity=1.0):

            """Add Laplace noise for differential privacy"""

            return value + np.random.laplace(0, sensitivity/epsilon)

        

        return [laplace_mechanism(item) for item in data]



    

# Privacy Preservation

personal_ids = [1001, 1002, 1003, 1004, 1005]

sensitive_data = [75, 82, 90, 68, 95]



# Anonymize identifiers

anonymous_ids = PrivacyPreservingML.anonymize_data(personal_ids)

print("Anonymized IDs:", anonymous_ids)



# Apply differential privacy

privacy_protected_data = PrivacyPreservingML.differential_privacy_noise(sensitive_data)

print("Privacy-Protected Data:", privacy_protected_data)

Let’s walk through the code:

    ...