Transformer Block Estimators for Input Data Obfuscation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models are vulnerable to data interception and misuse, particularly when sensitive data is transmitted over untrusted networks, necessitating effective input obfuscation techniques to protect privacy.
Innovation Solution
The use of estimators, specifically trained neural networks with transformer block architecture, to apply obfuscation to input data for machine learning models, ensuring that the obfuscated data maintains the quality of output while concealing the original data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If input data is transmitted over untrusted networks to machine learning models, then model functionality and accessibility are improved, but data privacy and security deteriorate
Solution Approach 1:
The system applies obfuscation transformations to input data before transmission over untrusted networks. Estimators (neural networks) pre-process the data to embed protective transformations that conceal sensitive information while preserving model functionality, preventing data interception risks before they can occur during transmission.
Solution Approach 2:
The patent introduces estimators as intermediary components between the user and the machine learning model. These estimators apply obfuscation transformations that act as a protective layer, allowing data to be transmitted over untrusted networks while the intermediary transformation prevents direct exposure of sensitive information to potential interceptors.
2Object-affected harmful factors
If obfuscation is applied to input data, then data privacy is improved, but model input quality and processing accuracy may deteriorate
Solution Approach 1:
The system dynamically adjusts obfuscation parameters based on the specific data characteristics and model requirements. The estimators learn optimal transformation parameters that balance privacy protection with maintaining sufficient data quality for accurate model processing, changing parameters adaptively rather than applying fixed obfuscation.
Solution Approach 2:
The patent implements feedback mechanisms where the system evaluates the impact of obfuscation transformations on model performance. This feedback is used to refine and adjust the obfuscation strategy, ensuring that privacy protection measures do not excessively degrade input quality or model processing accuracy.
3Object-affected harmful factors
If complex obfuscation techniques are implemented, then data security is improved, but system complexity and computational overhead increase
Solution Approach 1:
The estimators are trained to automatically learn and apply appropriate obfuscation transformations without requiring complex manual configuration or intervention. The system self-adjusts to find effective obfuscation strategies, reducing the operational complexity of implementing and maintaining security measures while maintaining strong data protection.
Data Source
AI summary
Provided are methods and systems for obtaining, by a computer system, a machine learning model, the machine learning model comprising at least one transformer block; generating, by the computer system, one or more estimator based on the at least one transformer block, wherein at least one estimator comprises a mean shift estimator; and wherein at least one estimator comprises a dispersion shift estimator; training, by the computer system, the one or more estimators to obfuscate input data for the machine learning model; and storing, by the computer system, the trained one or more estimators in memory.