#South Korea #Seoul #Desilo #Cornami #Encrypted AI

DESILO and Cornami's Groundbreaking Research on Encrypted AI Computation

In a remarkable turn of events in the tech industry, DESILO, a pioneering Privacy Enhancing Technology (PET) firm, and Cornami, known for its prowess in scalable compute acceleration, have announced a formidable advancement in the realm of encrypted artificial intelligence (AI) computation. This breakthrough focuses on full homomorphic encryption (FHE), a technology that ensures data privacy while processing sensitive information.

The research paper, co-authored by the renowned Craig Gentry—often hailed as the father of FHE—and Yongwoo Lee, who heads cryptography at DESILO, details a newly developed approach aimed at enhancing the efficiency of encrypted matrix arithmetic. This improvement boasts an impressive capability of accelerating encrypted matrix multiplication by up to 80 times faster when compared to existing cutting-edge technologies, thereby addressing a critical gap in the practical application of privacy-preserving AI.

Matrix multiplication forms the cornerstone of many contemporary machine learning models. The new method introduced by DESILO and Cornami is specifically tailored to optimize these operations across varying scales and real-world workloads. This advancement promises to bring a much-needed transition from theoretical cryptographic applications into functional AI systems that can operate effectively without compromising data privacy.

Seungmyung Lee, CEO of DESILO, commented on the significance of this development: _“This research shows that privacy-preserving computation can be both efficient and practical. It forms a crucial foundation for our encrypted AI stack, allowing organizations to analyze sensitive data securely.”_ The significance of such a secure yet efficient computation model cannot be overstated, especially as industries grapple with the need to maintain confidentiality amid rising data privacy concerns.

Collaboration has been a significant aspect of this research, bridging DESILO's innovations in encryption with Cornami's high-performance computing architecture. This strategic partnership is geared towards making fully homomorphic encryption a viable option in enterprise and AI sectors—an area where the need for both confidentiality and efficiency in computation cannot be overlooked.

Dr. Craig Gentry, Chief Scientist of Algorithms at Cornami, emphasized the historical challenges associated with deploying FHE in real-world applications due to its computational demands. He stated, _“For decades, FHE has been the benchmark for ensuring data privacy; however, high computational costs have limited its practical use.”_ He explained that the latest research aims to streamline encrypted matrix multiplication, a fundamental operation constituting over 90% of AI workloads. With access to Cornami's scalable Compute Fabric, the new method achieves unprecedented speeds in encrypted processing.

This development not only positions encrypted AI as a more practical application but also opens doors for utilizing large language models (LLMs) in environments demanding secure data handling. Efficient and secure matrix multiplication enables these models to perform inference on encrypted data at speeds approximating plaintext operations, thus reinforcing essential elements such as compliance with data regulations, data sovereignty, and post-quantum security.

The research paper detailing these groundbreaking findings has been made publicly available through the IACR ePrint Archive, signifying an epoch-altering moment in AI and data security. The implications of these advancements extend across various sectors, including healthcare and finance, where sensitive information must be shared and analyzed securely among multiple parties. DESILO is actively expanding its influence in these fields, developing frameworks for collaborative data initiatives without compromising privacy.

As we look towards the future, the partnership between DESILO and Cornami not only represents a major leap forward in encrypted computation but also lays the groundwork for advancing privacy-preserving AI as a standard practice across industries. The merging of security with performance reflects a crucial evolution in how technology can adapt to meet the demands of modern data handling and privacy requirements.