Open Source North
Building Trustworthy Tech: Balancing Innovation with Privacy in the AI Age

The power of machine learning (ML) has transformed numerous industries, but its reliance on vast amounts of data raises critical privacy concerns. This talk dives into real-world examples where ML poses privacy risks, showcasing how seemingly harmless data collection can lead to sensitive information exposure. Through compelling case studies, we’ll explore scenarios where:

– Personal attributes are inferred: From browsing history to social media activity, seemingly anonymized data can be used to reveal sensitive attributes like political views, health conditions, or financial situations.
– Bias creeps in: Algorithmic bias based on training data can lead to discriminatory outcomes, impacting areas like loan approvals, job applications, and even criminal justice.
– Surveillance gets sophisticated: Facial recognition, location tracking, and other ML-powered tools raise concerns about mass surveillance and potential misuse of personal data.

We’ll then delve into the world of privacy-preserving machine learning (PPML) as the crucial next step. We’ll explore promising techniques like:

1. Differential privacy: Adding carefully crafted noise to data protects individual privacy while preserving its usefulness for analysis.
2. Federated learning: Training models on decentralized devices without sharing raw data, keeping information secure within its source.
3. Homomorphic encryption: Performing computations on encrypted data, allowing analysis without decrypting sensitive information.

By showcasing real-world problems and presenting potential solutions, this talk aims to raise awareness about the importance of PPML for both beginners and intermediate audiences. We’ll conclude by highlighting the exciting future trends in PPML, from advancements in technology to evolving regulations, emphasizing its crucial role in building a responsible and ethical ML landscape.

Sri Harsha Gajavalli

PPML Researcher

Arizona State University