Homomorphic Encryption for Secure AI Credit Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Financial institutions face challenges in transmitting unencrypted confidential data to public cloud-based computing clusters for machine learning and AI processes due to confidentiality and privacy restrictions, limiting the ability to operate on encrypted data during training and deployment phases.
Innovation Solution
Implementing a homomorphic encryption scheme that allows secure computations on encrypted data, enabling a third-party computing system to train and deploy machine learning models using homomorphically encrypted customer data, while maintaining privacy and confidentiality, and decrypting output data for notification to customers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If unencrypted confidential data is transmitted to public cloud-based computing clusters for machine learning and AI processes, then computational efficiency and productivity are improved, but data confidentiality and privacy are compromised
Solution Approach 1:
Homomorphic encryption serves as an intermediary mechanism that enables cloud-based AI processing while preserving data confidentiality. The encryption scheme allows computations to be performed on encrypted data without decryption, acting as a mediator between the need for computational efficiency and data privacy protection
Solution Approach 2:
The patent changes the state of data from unencrypted to homomorphically encrypted form, transforming the parameter of data representation. This parameter change enables the data to maintain confidentiality while still being processable by AI models, resolving the contradiction between security and computational efficiency
2Object-affected harmful factors
If encrypted data is used for training and deployment of machine learning models, then data privacy is maintained, but computational complexity increases
Solution Approach 1:
The patent replaces traditional decryption-and-process mechanics with homomorphic encryption mechanics. Instead of decrypting data before processing (traditional mechanical approach), the system performs computations directly on encrypted data, substituting the operational mechanism while maintaining data privacy
3Reliability
If confidential data is not transmitted to third-party computing systems, then data security is maintained, but the ability to leverage distributed computing resources is limited
Solution Approach 1:
Homomorphic encryption provides multi-functionality by enabling both data security and cloud computing access simultaneously. The same encryption mechanism that protects data confidentiality also enables the data to be processed by third-party AI systems, making the solution adaptable to distributed computing environments
Data Source
AI summary
The disclosed embodiments include computer-implemented processes that predict a credit score for a customer in real-time based on an application of a trained machine-learning or artificial-intelligence process to encrypted event data at a third-party computing cluster. For example, an apparatus may transmit encrypted event data to a third-party computing system. The third-party computing system may apply a trained artificial intelligence process to encrypted feature data that includes the encrypted event data, and the apparatus may receive, from the third-party computing system, encrypted output data representative of a predicted credit score during at least one temporal interval. The apparatus may decrypt the encrypted output data using a homomorphic decryption key, and transmit a notification that includes the decrypted output data to a device. An application program executed at the device may present a graphical representation of at least a portion of the decrypted output data within a digital interface.