Distributed Executable Packages for Private Fraud Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML model training systems in untrusted environments fail to address privacy concerns, as financial institutions are reluctant to share sensitive data with cloud service providers due to privacy and regulatory issues, limiting the development of robust fraud detection models.
Innovation Solution
Implementing distributed executable file packages that enable federated learning, split learning, or federated split learning, allowing clients to train partial ML models locally and share trained parameters with a server for aggregation, maintaining data privacy while constructing a more robust fraud prediction model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud service providers collect data from multiple clients to train a global fraud detection model, then model accuracy is improved, but client data privacy is compromised
Solution Approach 1:
The ML model is divided into multiple partial models that are distributed to different client devices. Each client trains their local partial model using their own data without sharing it with others. The trained partial models are then aggregated by the server to construct the global fraud detection model, achieving high accuracy while maintaining data privacy.
Solution Approach 2:
The server acts as an intermediary that coordinates the federated learning process. It distributes model parameters to clients, collects trained partial models, and aggregates them into a global model. This intermediary structure enables collaborative training without direct data sharing between clients, resolving the privacy-accuracy contradiction.
2Reliability
If clients share sensitive financial data with cloud service providers, then robust fraud detection models can be developed, but privacy and regulatory compliance issues arise
Solution Approach 1:
The training process is segmented so that each client keeps their sensitive financial data locally and only shares trained model parameters with the server. This segmentation eliminates the need for direct data sharing while still enabling the development of robust fraud detection models through collective learning from multiple clients' data patterns.
Solution Approach 2:
Instead of sharing original sensitive data, the system shares copies in the form of trained model parameters and weights. These parameter copies contain learned patterns from the data without revealing the actual sensitive financial information, thus maintaining privacy compliance while achieving robust model performance.
3Measurement precision
If conventional centralized ML training is used, then model training can be performed with large datasets, but data privacy cannot be preserved in untrusted environments
Solution Approach 1:
The centralized training approach is transformed into a distributed segmented training process. The global model is split into partial models distributed across multiple client devices. Each segment trains locally on its own data, and the segments are aggregated to form the complete model. This maintains training quality while ensuring privacy in untrusted environments.
Solution Approach 2:
The training paradigm shifts from a single centralized dimension to a distributed multi-dimensional architecture. Training occurs simultaneously across multiple client devices and the server in different dimensional spaces, then the results are aggregated. This dimensional transformation enables both high training quality and privacy preservation.
Data Source
AI summary
Aspects of the present disclosure provide systems, methods, and computer-readable storage media that support cooperative training of machine learning (ML) models that preserves privacy in untrusted environments using distributed executable file packages. The executable file packages may include files, libraries, scripts, and the like that enable a cloud service provider configured to provide server-side ML model training to also support cooperative ML model training with multiple clients, particularly for a fraud prediction model for financial transactions. Because the cooperative training includes the clients training respective ML models and the server aggregating the trained ML models, private client data such as financial transaction data may be used to train the fraud prediction model without exposing the client data to others. Such cooperative ML model training enables offloading of computing resource-intensive training from client devices to the server and may train a more robust fraud detection model.