Homomorphic Matrix Processing Using Ciphertext Transpose

The patent divides the CCMM operation into two distinct phases: a ciphertext phase where matrix multiplication is performed on encrypted data to maintain security, and a plaintext phase where the result is decrypted and further computation is performed. This segmentation allows the system to leverage efficient plaintext operations for the final computation while maintaining security during the encrypted phase.

The patent introduces a decryption intermediate step as a mediator between ciphertext operations and final plaintext results. By decrypting the ciphertext matrix multiplication result to obtain a plaintext matrix, the system can then perform additional efficient plaintext operations. This intermediary decryption step enables the system to combine the security benefits of homomorphic encryption with the computational efficiency of plaintext operations.

The patent segments the computation into encrypted-phase operations (which consume resources but maintain security) and plaintext-phase operations (which are more efficient). By performing the final computation in the plaintext phase after decryption, the system reduces overall resource consumption compared to performing all operations in the ciphertext phase.

The decryption intermediate step acts as a resource optimization mediator. It converts ciphertext data to plaintext, enabling the use of highly optimized plaintext linear algebra libraries and operations that consume significantly fewer resources than homomorphic operations, thereby reducing total resource consumption while maintaining security during the encrypted computation phase.