LATIDIA · Ciberseguridad

MOAI: Arquitectura de optimización de módulos para la inferencia de transformadores seguros no interactivos

arXiv: 2611.13641v1Tipo de Anuncio: nuevo Resumen: Se han planteado problemas de privacidad en la inferencia de Modelos de Lenguaje Grande (LLM) cuando los modelos se implementan en Proveedores de Servicios en la Nube (CSP). El cifrado homomórfico (HE) ofrece una p

La noticia

arXiv:2611.13641v1 Announce Type: new Abstract: Privacy concerns have been raised in Large Language Models (LLM) inference when models are deployed in Cloud Service Providers (CSP). Homomorphic encryption (HE) offers a promising solution by enabling secure inference directly over encrypted inputs. However, the high computational overhead of HE remains a major bottleneck. To address this challenge, we propose MOAI, an efficient HE-based, non-interactive framework for secure transformer inference. MOAI gains significant efficiency improvement from: (1) a novel evaluation flow that combines column and diagonal packing with consistent strategies across all layers, eliminating expensive format conversions. (2) rotation-free algorithms for Softmax and LayerNorm that significantly reduce the number of costly HE rotations,