Privacy-Preserving Anomaly Detection on Distributed Ledgers
Solution Overview
Problem
Decentralized, private permissioned distributed ledger networks face challenges in managing transactions with personal identifiable information, as network operators lack visibility into metadata and activities, making them vulnerable to malicious activities.
Innovation Solution
Implementing a method for privacy-preserving machine learning model sharing and anomaly detection using local and aggregated machine learning models, where institutions train local models, aggregate parameters securely, and update models without sharing raw data, leveraging DeepAnT models and secure aggregation protocols to detect anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If all transactions are kept private on the distributed ledger network, then data privacy is improved, but network operator visibility and security monitoring deteriorate
Solution Approach 1:
The patent introduces a trusted entity as an intermediary that collects model parameters from multiple institutions, aggregates them into a global machine learning model, and enables anomaly detection without requiring the network operator to see raw transaction data. This mediator allows privacy preservation while enabling security monitoring through the aggregated model.
Solution Approach 2:
The patent replaces direct human/operator monitoring of transactions with an automated machine learning system. The global model, trained on aggregated parameters from local models, automatically detects anomalies and suspicious activities, substituting manual security review with intelligent automated detection that works within private transaction constraints.
2Measurement precision
If institutions share raw transaction data for model training, then model accuracy is improved, but data privacy and security are compromised
Solution Approach 1:
The patent extracts only the essential model parameters needed for training the global machine learning model, while leaving the sensitive raw transaction data localized at each institution. By taking out only the necessary computational elements (model parameters) and leaving the sensitive data behind, the system achieves collaborative model improvement without data sharing.
Solution Approach 2:
The patent transforms raw transaction data into model parameters through local model training, then aggregates these parameters to build a global model. This parameter transformation allows the system to work with transformed representations rather than original sensitive data, maintaining privacy while enabling collaborative learning.
3Reliability
If the network operator monitors all transactions for security, then anomaly detection capability is improved, but system complexity and computational overhead increase
Solution Approach 1:
The patent segments the anomaly detection function into distributed local models at each institution and a centralized global model for aggregation. Instead of one centralized monitoring system processing all transactions, the detection capability is divided across multiple independent local models that contribute parameters to a global model, reducing the complexity burden on any single component.
Solution Approach 2:
Each institution maintains and trains its own local machine learning model using its local data, performing self-service anomaly detection at the institutional level. This distributes the computational workload and reduces the burden on the network operator, while the aggregated global model provides additional collaborative detection capabilities.
Data Source
AI summary
A method for privacy preserving machine learning model sharing may include a computer program for a first institution of a plurality of institutions in a distributed ledger network: receiving transaction data for a transaction; training a local machine learning model using the transaction data; submitting parameters for the local machine learning model to the distributed ledger network as a private transaction with a trusted entity, wherein the trusted entity receives parameters for a plurality of local machine learning models from the distributed ledger network for the plurality of institutions in the distributed ledger network and aggregates the parameters into an aggregated machine learning model and submits the aggregated parameters to the distributed ledger network as one or more transactions; receiving, from the distributed ledger network, the aggregated parameters for the aggregated machine learning model; and updating the local machine learning model with the aggregated parameters.