Researchers have developed MOSAIC, a novel protocol for securely outsourcing AI computations. This method uses a matrix-multiplication masking technique that introduces minimal noise, allowing for efficient processing of large AI models like transformers. MOSAIC significantly reduces client overhead and offers runtimes orders of magnitude faster than previous approaches, with security based on LWE and LPN assumptions. An implementation demonstrates its potential for confidential AI in data centers, keeping the trusted computing base small while offloading heavy computation to untrusted hardware. AI
IMPACT Enables confidential AI computations by securely outsourcing workloads to untrusted servers.
RANK_REASON The cluster contains an academic paper detailing a new method for AI computation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
