Federated Learning Anomaly Detection via Predictive Score Ratios

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems face high processing costs and data security risks when processing and analyzing large amounts of financial data across different entities, particularly due to the need for data transmission and storage at central locations, which are vulnerable to attacks like adversarial attacks and backdoor attacks.

Innovation Solution

Implementing a federated learning system that allows data to be stored and processed at local devices, using techniques such as clustering and training data analysis to detect anomalies and prevent attacks, thereby reducing the risk of data breaches and maintaining data privacy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If data is transmitted and stored at central locations for processing and analysis, then processing capabilities and model training can be performed, but data security risks increase and systems become vulnerable to adversarial attacks and backdoor attacks

Engineering Contradiction:

Improvedata securityVSAvoidadversarial attacks

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent divides the centralized data processing system into distributed federated learning nodes. Each participant maintains local data and models independently, transmitting only model updates rather than raw data. This segmentation eliminates the central vulnerable point while maintaining collaborative model training capabilities across multiple distributed entities.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces intermediary mechanisms including secure aggregation protocols and trusted execution environments that mediate between distributed participants. These intermediaries enable model update exchange and coordination without requiring direct data sharing or centralized data storage, thus preventing adversarial attacks while facilitating collaborative learning.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If a federated learning system is implemented to store and process data at local devices, then data security and privacy are improved, but the system complexity increases due to the need for anomaly detection and attack prevention mechanisms

Engineering Contradiction:

Improvedata privacyVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent implements preliminary anomaly detection mechanisms that analyze model updates before they are aggregated into the global model. By performing detection actions in advance rather than reactively, the system prevents compromised updates from corrupting the model while maintaining a relatively simple architecture compared to continuous monitoring approaches.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent incorporates feedback loops where model updates are evaluated against detection thresholds and historical patterns. This feedback mechanism enables automatic identification and rejection of anomalous updates without requiring complex real-time analysis, thus managing system complexity while ensuring data privacy and model integrity.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If clustering techniques and training data analysis are used to detect anomalies, then the detection accuracy improves, but the processing time and computational costs increase

Engineering Contradiction:

Improveanomaly detection accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial analysis by focusing detection efforts on critical model update parameters rather than analyzing entire model architectures. This selective approach achieves sufficient detection accuracy for identifying adversarial attacks while significantly reducing computational overhead and processing time compared to comprehensive analysis methods.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The patent transforms complex anomaly detection into parameter-based threshold comparisons by analyzing specific statistical properties of model updates. This parameter transformation converts computationally intensive pattern recognition tasks into simple threshold evaluations, maintaining detection accuracy while minimizing processing time and computational resources required.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250175483A1Method to detect and defend against targetted adversarial attacks on a federated learning system

Publication Date: 2025.05.29 U S BANCORP NAT ASSOC

AI summary

A method comprises transmitting a first client model to a first computing device and a second client model to a second computing device; determining (i) a first predictive score ratio for the first computing device, and (ii) a first predictive score ratio for the second computing device; determining the first computing device and the second computing device match; determining (i) a second predictive score ratio for the first computing device, and (ii) a second predictive score ratio for the second computing device; and detecting an anomaly in the first computing device responsive to (i) the determining the first computing device and the second computing device match, and (ii) determining the second predictive score ratio for the first computing device exceeds the second predictive score ratio for the second computing device by an amount above a difference threshold.