Homomorphic Encryption
An encryption technique that allows computations to be performed directly on ciphertext without decrypting the data, theoretically achieving the highest level of data privacy protection. However, its computational overhead remains substantial, and it has not yet been widely adopted in production-grade AI inference.
Core Facts
Timeline (last 90 days)
If Proton wants to embed AI in its products, it can often only take paths such as local inference, client-side processing or privacy-enhancing technologies (such as federated learning and homomorphic encryption), and these approaches still lag cloud-based large models in cost and effectiveness.
Achieving end-to-end verifiability usually requires cryptographic techniques such as zero-knowledge proofs, mix-nets, or homomorphic encryption.
Privacy-enhancing technologies (PETs) include homomorphic encryption, federated learning, differential privacy and secure multi-party computation, with the goal of maintaining privacy protection at the data usage stage as well, but they still face deployment challenges such as high performance overhead and high engineering complexity.
In actual products, the combination of TEE and traditional encryption is more practically feasible than homomorphic encryption.
Traditional multi-party computation (MPC) is implemented through cryptographic protocols such as secret sharing and homomorphic encryption, but the computational overhead is enormous, making it difficult to apply to large-scale neural network inference.
Homomorphic encryption theoretically allows inference to be run directly on encrypted data, but the computational overhead is still too large and it has not yet become practical.
Homomorphic encryption can perform computations directly on encrypted data without first decrypting it, and the results are also encrypted and can only be decrypted by those who hold the key.
谷歌近期推动同态加密技术，目标是让同态加密变得实用，通过算法优化、硬件加速和开发工具链完善来压缩性能开销
在密文上做运算的计算量往往是明文运算的成千上万倍，是同态加密工程化的最大障碍
CKKS方案支持近似浮点数运算，特别适合机器学习中的矩阵乘法和激活函数计算，成为AI领域应用同态加密的重要基础
1 more timeline events
All Facts (11)
Homomorphic encryption theoretically allows inference to be run directly on encrypted data, but the computational overhead is still too large and it has not yet become practical.
80%UnverifiedHomomorphic encryption can perform computations directly on encrypted data without first decrypting it, and the results are also encrypted and can only be decrypted by those who hold the key.
70%UnverifiedIf Proton wants to embed AI in its products, it can often only take paths such as local inference, client-side processing or privacy-enhancing technologies (such as federated learning and homomorphic encryption), and these approaches still lag cloud-based large models in cost and effectiveness.
50%UnverifiedAchieving end-to-end verifiability usually requires cryptographic techniques such as zero-knowledge proofs, mix-nets, or homomorphic encryption.
50%UnverifiedPrivacy-enhancing technologies (PETs) include homomorphic encryption, federated learning, differential privacy and secure multi-party computation, with the goal of maintaining privacy protection at the data usage stage as well, but they still face deployment challenges such as high performance overhead and high engineering complexity.
50%UnverifiedIn actual products, the combination of TEE and traditional encryption is more practically feasible than homomorphic encryption.
50%UnverifiedTraditional multi-party computation (MPC) is implemented through cryptographic protocols such as secret sharing and homomorphic encryption, but the computational overhead is enormous, making it difficult to apply to large-scale neural network inference.
50%UnverifiedCKKS方案支持近似浮点数运算，特别适合机器学习中的矩阵乘法和激活函数计算，成为AI领域应用同态加密的重要基础
50%Unverified同态加密的理论基础可以追溯到2009年Craig Gentry的工作，他基于理想格数学结构首次证明了可以在密文上执行任意计算
50%Unverified在密文上做运算的计算量往往是明文运算的成千上万倍，是同态加密工程化的最大障碍
50%Unverified谷歌近期推动同态加密技术，目标是让同态加密变得实用，通过算法优化、硬件加速和开发工具链完善来压缩性能开销
50%