Encrypted Vector Retrieval Using Rotation to Resist Vec2Text Attacks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vector retrieval methods lack security protection for embedding vectors, making them vulnerable to vec2text attacks that leak user privacy.
Innovation Solution
Implement a vector retrieval method involving linear transformation with a random number and rotation using an orthogonal matrix to encrypt vectors, generating ciphertext vectors that maintain similarity while masking plaintext information, thereby resisting vec2text attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If embedding vectors are stored in knowledge vector base for retrieval, then vector retrieval functionality is enabled, but user privacy is exposed to vec2text attacks
Solution Approach 1:
The patent introduces ciphertext vectors as an intermediary representation between plaintext and stored embeddings. The encryption module transforms plaintext vectors into ciphertext vectors using cryptographic algorithms, which are then stored in the knowledge vector base. During retrieval, ciphertext vectors are used for similarity search, and only after successful retrieval is the plaintext reconstructed through decryption. This intermediary mechanism enables retrieval functionality while preventing direct access to plaintext during storage and search operations.
Solution Approach 2:
The patent changes the parameter state of vectors from plaintext form to encrypted ciphertext form through cryptographic transformation. The encryption process modifies the vector representation by applying cryptographic algorithms that transform the original embedding parameters into encrypted parameters. This parameter change ensures that the stored vectors cannot be directly attacked through vec2text methods, as the attack requires access to plaintext or unencrypted representations.
2Reliability
If encryption is applied to protect embedding vectors, then privacy security is improved, but retrieval accuracy may deteriorate
Solution Approach 1:
The patent applies preliminary encryption action to transform plaintext vectors into ciphertext vectors before storage. The encryption module performs cryptographic transformation on the embedding vectors prior to storing them in the knowledge vector base. This preliminary encryption ensures that the vectors are protected from attacks during storage and retrieval operations, while the cryptographic design maintains the structural properties necessary for accurate similarity search and retrieval.
Data Source
AI summary
A vector retrieval method and apparatus, a medium, an electronic device and a program product. The method includes: acquiring a target vector to be retrieved and a target key; performing linear transformation on the target vector by using the target random number; performing a rotation operation on the target vector obtained after the linear transformation by using the target orthogonal matrix to obtain a first ciphertext vector; and sending the first ciphertext vector to a first server, so that the first server retrieves at least one second ciphertext vector that is similar to the first ciphertext vector from a knowledge vector base, ciphertext vectors in the knowledge vector base are generated based on the target key and a reference vector, and a generation manner of the ciphertext vectors is the same as a generation manner of the first ciphertext vector.