PII Encoder Using Bi-gram Co-occurrence Matrices for Irreversible Tokenization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for transforming personal identifiable information (PII) into anonymized representations face challenges such as reversibility, loss of matching accuracy, and intolerance for variable data sizes, particularly in lossy transformation techniques, which fail to provide sufficient security and accuracy in biometric matching.
Innovation Solution
A system and process that encode PII into fixed-size vector representations using bi-grams, apply smoothing functions, and perform lossy transformations to generate irreversible tokens, ensuring privacy, security, and accurate matching capabilities while being agnostic to cultural and ethnic variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If lossless transformation techniques are used to translate PII into secondary representation, then security is improved, but the transformation is reversible allowing attackers to access original data
Solution Approach 1:
The patent transforms PII data by changing its parameter representation from original format to a derived format using cryptographic functions. The transformation modifies the data parameters irreversibly, converting structured PII into a format that cannot be reverted to the original state, thus preventing attackers from recovering original information even if they obtain the transformed representation.
Solution Approach 2:
The patent replaces traditional reversible encryption mechanisms with irreversible cryptographic transformation. Instead of using symmetric encryption that can be decrypted, the system employs one-way hash functions and cryptographic derivations that mathematically prevent reversal, substituting the mechanical encryption/decryption process with a one-way transformation process.
2Reliability
If lossy transformation techniques are used to transform PII, then security is improved through irreversibility, but matching accuracy deteriorates significantly
Solution Approach 1:
The patent segments PII data into distinct components (e.g., separating name, date of birth, location) and applies specific transformation rules to each segment. This segmentation allows the system to preserve essential matching characteristics in each component while applying irreversible transformation, enabling accurate matching without requiring the complete original data to be recoverable.
Solution Approach 2:
The patent introduces cryptographic intermediaries (hash functions, salt values, derivation keys) that mediate between the original PII and the transformed representation. These intermediaries enable the system to achieve irreversible transformation while maintaining the ability to match records through the intermediary structure, allowing accurate matching without direct access to original data.
3Quantity of substance
If variable-length PII data is transformed into fixed-size representations, then storage efficiency is improved, but transformation accuracy deteriorates
Solution Approach 1:
The patent transforms variable-length PII data by changing its dimensional representation. Instead of compressing the data along its original dimensional axis (which loses information), the system maps the data to a different dimensional space using cryptographic functions, where the fixed-size representation preserves the essential characteristics needed for matching while achieving storage efficiency.
Data Source
AI summary
A system can include a processor in communication with a data store. The processor can obtain personal identifiable information (PII) data and segregate the PII data into two or more secondary representations. The processor can generate a plurality of co-occurrence matrices based on the two or more secondary representations. The processor can perform a convolution between each of the plurality of co-occurrence matrices and one of a plurality of Gaussian kernels, wherein each of the plurality of Gaussian kernels comprises a different width. The processor can generate a tertiary representation of the PII data by performing a linear combination of the plurality of co-occurrence matrices. The processor can generate a vector based on the tertiary representation and perform a lossy tokenization process on the vector to generate a token. The processor can store the token at the data store.