Fhela

Machine Learning on Encrypted data - protection sensitive datasets We are developing own Hardware acceleration and Software of Fully Homomorphic Encryption (FHE). Hardware acceleration (ASIC) ML FHE - will allow to accelerate FHE operations for training AI on encrypted datasets, making it possible to apply FHE methods to Machine Learning. Software FHE There are two types of ML FHE: (a) Confidential inference (operations of encrypted neural network). The trained neural network and user queries remain encrypted throughout the process. (b) Confidential AI training. Addresses the needs of various market participants by allowing them to train AI models on encrypted data, ensuring that the datasets, models, and user queries remain secure and proprietary. FHE allows owners of datasets, models, and queries to protect their assets from theft. Use cases: - Biotech AI for drug development. Handles large, expensive datasets and models under heavy regulation, ensuring secure data sharing and collaboration. - Science AI - Military AI - Fintech AI - Intellectual property AI. What is Fully Homomorphic Encryption (FHE)? - FHE allows computation on encrypted data without decrypting it. - Proven to be quantum resistant We have: - Ready 3rd hardware prototype - LWE core, NTT core, Programmable Bootstrapping core - Patent US 12095896 - 6 patents of FHE in preparation - The benchmark of 2nd hardware prototype showed the best resource savings and speed compared to competitors - Faster polynomial multipliers (5 times) than Montgomery multipliers - own math solution Machine Learning on Encrypted data - protection sensitive datasets We are developing own Hardware acceleration and Software of Fully Homomorphic Encryption (FHE). Hardware acceleration (ASIC) ML FHE - will allow to accelerate FHE operations for training AI on encrypted datasets, making it possible to apply FHE methods to Machine Learning. Software FHE There are two types of ML FHE: (a) Confidential inference (operations of encrypted neural network). The trained neural network and user queries remain encrypted throughout the process. (b) Confidential AI training. Addresses the needs of various market participants by allowing them to train AI models on encrypted data, ensuring that the datasets, models, and user queries remain secure and proprietary. FHE allows owners of datasets, models, and queries to protect their assets from theft. Use cases: - Biotech AI for drug development. Handles large, expensive datasets and models under heavy regulation, ensuring secure data sharing and collaboration. - Science AI - Military AI - Fintech AI - Intellectual property AI. What is Fully Homomorphic Encryption (FHE)? - FHE allows computation on encrypted data without decrypting it. - Proven to be quantum resistant We have: - Ready 3rd hardware prototype - LWE core, NTT core, Programmable Bootstrapping core - Patent US 12095896 - 6 patents of FHE in preparation - The benchmark of 2nd hardware prototype showed the best resource savings and speed compared to competitors - Faster polynomial multipliers (5 times) than Montgomery multipliers - own math solution