Automated Hypothesis Generation Using Secure Enclaves for AI Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of generating algorithms without access to large and sensitive datasets, particularly in healthcare, is hindered by the need for secure data transfer and compliance with existing technologies, which often requires extensive data transfer and compliance with privacy protections, which makes it challenging to share datasets, and the need for secure data transfer and compliance with privacy protections, which existing methods fail to address these challenges, and the lack of interoperability between data stewards and algorithm developers.
Innovation Solution
A system and method for automated hypothesis generation using secure enclaves and trusted execution environments, such as Intel SGX, to encrypt and process data within sequestered computing nodes, allowing algorithms to be developed without direct access to sensitive data, using generative AI to create data specifications and validate models within a zero-trust framework.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If datasets are shared with algorithm developers for model training, then algorithm development and validation can proceed, but data privacy and security are compromised
Solution Approach 1:
A trusted execution environment (TEE) acts as an intermediary between data stewards and algorithm developers. The TEE enables secure data processing by allowing algorithm developers to train models on sensitive data without directly accessing it, as the data remains encrypted and isolated within the TEE boundary. This resolves the contradiction by enabling algorithm development while maintaining data privacy through the TEE mediation layer.
Solution Approach 2:
Instead of sharing actual sensitive data, the system creates and processes copies of data within the trusted execution environment. The algorithm developer works with data copies that are isolated from the original sensitive datasets, allowing model training to proceed without compromising the security or privacy of the source data.
2Productivity
If large datasets are transferred to algorithm developers, then comprehensive model training is enabled, but transfer time and infrastructure requirements increase significantly
Solution Approach 1:
The system extracts only the essential data processing capabilities needed for model training and places them within the trusted execution environment. Rather than transferring entire large datasets, the TEE provides on-demand access to data processing functionality, allowing algorithm developers to train models without receiving or storing large volumes of data externally.
Solution Approach 2:
The solution moves the data processing operation to a different dimension - from external data transfer to cloud-based TEE processing. Instead of horizontally transferring data across networks, the system vertically integrates data access within the cloud infrastructure's trusted execution environment, eliminating the need for large-scale data transfers.
3Object-affected harmful factors
If data access restrictions are implemented to protect privacy, then data security is improved, but interoperability between data stewards and algorithm developers deteriorates
Solution Approach 1:
The trusted execution environment provides a universal interface that works across different data stewards and algorithm developers. The TEE's standardized attestation and secure processing capabilities enable any algorithm developer to work with any data steward's data without custom integrations, maintaining interoperability while enforcing security restrictions through the common TEE framework.
Data Source
AI summary
Systems and methods related to the generation of a hypothesis is presented. First a user inputs at least one research paper or study. The at least one research paper or study is mined by an AI algorithm for a set of data requirements. The set of data requirements may be combined with a tailored prompt into a generative AI system to generate a data specification. Next, within a secure enclave, the data specification may be combined with natural language (NL) prompts to interrogate data sets from a plurality of data stewards. A data set from the interrogated data sets that meets the data specification is selected. The data set is then encrypted, used to generate and train an AI model.