Google TechTalks — searchable YouTube transcripts & AI Q&A

Privacy Amplification for Correlated-Noise Mechanisms via b-Min-Sep Subsampling

This talk explores privacy amplification for correlated noise mechanisms (DPMF) using b-min-sep subsampling, addressing challenges in combining DP-SGD with DPMF.

Local Node Differential Privacy

This video introduces local node differential privacy, a new model for private analysis of distributed network data without a trusted third party.

Understanding Trade-offs Between Worst-case Differential Privacy Guarantees & Real Threat Models

This video explores the gap between theoretical differential privacy guarantees and real-world attack success, focusing on trade-offs in threat models and DP-SGD.

Atomic Facts to Structured Knowledge: Rethinking Unlearning & Jailbreaking in Large Language Models

This talk explores how structured knowledge in LLMs creates vulnerabilities for unlearning and jailbreaking, proposing graph-based modeling for trust and safety.

Differentially Private Table-Image Multimodal Data Generation

This video explains differentially private synthetic data generation for tables and images, discussing DP mechanisms and why synthetic data is needed.