Federated Learning Neural Network Weight Updates for Financial Crime Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The financial industry's decentralized system, where data is stored in secure 'silos,' prevents the sharing of information between entities, hindering the detection of illicit financial crimes like money laundering, despite the interconnected nature of financial markets.
Innovation Solution
A cloud-based federated learning system that allows financial entities to share updates in neural network weights without exchanging raw data, enabling improved detection of money laundering and other financial crimes through aggregated feature importance differential values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data is stored in secure encrypted silos to maintain data security and prevent data leakage, then data security and confidentiality are improved, but the ability to share information between entities for detecting illicit financial crimes deteriorates
Solution Approach 1:
The patent extracts only the essential learning information (weight updates and feature importance differential values) from the neural networks while leaving the raw customer data confined within secure silos. This extraction enables information sharing for crime detection without compromising data security or requiring data to leave the encrypted environments.
Solution Approach 2:
The patent introduces federated learning as an intermediary mechanism that facilitates collaboration between entities with encrypted data silos. Through this intermediary system, entities can jointly train detection models and share insights about financial crimes without directly sharing their underlying customer data, thus mediating between security requirements and information sharing needs.
2Adaptability or versatility
If entities maintain isolated data silos to prevent competitors from gaining insights, then commercial confidentiality is improved, but the accuracy of illicit activity detection across the financial system deteriorates
Solution Approach 1:
The patent merges the detection capabilities of multiple entities by combining their neural network weight updates and feature importance differential values through federated learning. This merging enhances detection accuracy by aggregating knowledge from multiple sources while maintaining the commercial confidentiality of each entity's raw data through the federated approach.
3Measurement precision
If raw customer data is shared between entities to improve detection accuracy, then detection accuracy is improved, but data leakage risk and regulatory compliance issues worsen
Solution Approach 1:
The patent extracts only the learned patterns and weight adjustments from neural networks trained on local data, rather than sharing the raw customer data itself. This extraction enables detection accuracy improvement while eliminating data leakage risks associated with sharing sensitive customer information between entities.
Solution Approach 2:
The patent creates and shares copies of the learned knowledge in the form of weight updates and feature importance differential values, rather than copying or sharing the original raw customer data. These copies contain the essential detection insights without revealing sensitive customer information, thus improving detection while preventing data leakage.
Data Source
AI summary
A method of updating a first neural network is disclosed. The method includes providing a computer system with a computer-readable memory that stores specific computer-executable instructions for the first neural network and a second neural network separate from the first neural network. The method also includes providing one or more processors in communication with the computer-readable memory. The one or more processors are programmed by the computer-executable instructions to at least process a first data with the first neural network, process a second data with the second neural network, update a weight in a node of the second neural network by a delta amount as a function of the processing of the second data with the second neural network, and update a weight in a node of the first neural network as a function of the delta amount. A computer system for updating a first neural network is also disclosed. Other features of the preferred embodiments are also disclosed.