Google Research has developed new algorithms for differentially private partition selection, a technique that enables the safe sharing of large datasets for AI and machine learning advancements. Their parallel algorithm, detailed in the paper "Scalable Private Partition Selection via Adaptive Weighting" at ICML2025, can process datasets with hundreds of billions of items, significantly outperforming previous sequential methods. This innovation aims to balance robust privacy guarantees with data utility, allowing for the extraction of valuable information like common words from private text corpora without compromising individual user data. Google is open-sourcing this technology to foster community collaboration. AI
IMPACT Enables safer and more scalable use of large datasets for AI training, potentially accelerating model development.
RANK_REASON Publication of a research paper detailing novel algorithms for differentially private partition selection. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Google AI / Research →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
