Federated Anomaly Detection Models for Privacy-Constrained Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing federated learning systems face challenges in training machine learning models on sensitive and distributed datasets without violating privacy regulations or requiring data centralization, particularly in environments with stringent data privacy laws, banking regulations, and confidentiality agreements.
Innovation Solution
Implementing a distributed computing architecture that allows authenticated client devices to locally execute model training scripts using private data, aggregating locally trained models into a federated champion model on the server without transferring raw data, leveraging techniques like differential privacy and specialized server software to ensure privacy and compliance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data is aggregated into a central repository for model training, then model training effectiveness is improved, but data privacy compliance deteriorates
Solution Approach 1:
The patent introduces federated learning as an intermediary mechanism that enables model training across distributed data sources without centralizing the data itself. The global model acts as a mediator that aggregates knowledge from local models trained on private data, achieving training effectiveness while preserving data privacy compliance through cryptographic techniques and secure aggregation protocols
Solution Approach 2:
The patent segments the model training process into distributed local training operations and centralized model aggregation operations. Each participating organization trains local models on their own data segments, and only model parameters (not raw data) are shared and aggregated, thus maintaining both training effectiveness and data privacy
2Measurement precision
If raw data is transferred to the server for model training, then model accuracy is improved, but bandwidth consumption and data security deteriorate
Solution Approach 1:
The patent extracts only the essential model parameters and gradients from local training processes, transmitting only these compressed representations to the server for aggregation. This extraction approach achieves model accuracy improvement while dramatically reducing bandwidth consumption compared to transferring raw data
Solution Approach 2:
The patent uses model parameter copies and gradients as surrogates for raw data. Instead of transferring actual data, the system transfers replicated and transformed versions (model updates) that contain the necessary training information but consume minimal bandwidth and preserve security
3Measurement precision
If data is shared across multiple organizations for model training, then anomaly detection accuracy is improved, but data sovereignty and privacy protection deteriorate
Solution Approach 1:
The patent inverts the traditional data sharing paradigm by bringing the model to the data instead of bringing the data to the model. Local models are trained on-site using private data, and only model updates are shared, thus achieving improved anomaly detection accuracy while maintaining data sovereignty and privacy protection
4Object-affected harmful factors
If federated learning is implemented across distributed devices, then data privacy is preserved, but system complexity and coordination overhead increase
Solution Approach 1:
The patent merges the coordination functions into a centralized server that manages the federated learning process. The server handles model distribution, aggregation, and synchronization, thereby preserving data privacy across distributed devices while reducing individual device complexity through centralized coordination management
Data Source
AI summary
Systems and methods are described herein for generating, training, and federating machine learning models to detect anomalous or rare-event activity in sensitive electronic data, such as financial transactions. In some implementations, configuration information for a machine learning script (MLS) is obtained. The MLS is generated by a server based on the configuration. Data representing the MLS is provided to a user device. The user device is caused to perform operations when executing the MLS. An instance of a machine learning model is trained based on the MLS. One or more parameters associated with the instance of the machine learning model are identified. Data representing the one or more parameters associated with the instance of the machine learning model are received by the server from the user. A federated model is generated based at least in part on the received data representing the identified parameters and provided for output.