Transaction Graph Assessment Using Generator-Discriminator Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional illicit activity detection systems, such as rule-based and supervised machine learning models, face challenges due to label scarcity and high false-positive rates, while unsupervised methods are evaded by adversaries mimicking legitimate behaviors.
Innovation Solution
A generator-discriminator network architecture that models the final objective of illicit actors, generating synthetic illicit activity without labeled data, and a differentiable rule-based proxy provides gradient information to train a discriminator to distinguish between legitimate and illicit transactions, using tensor representations of transactional graphs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based detection systems are used, then detection coverage is improved, but false positive rate increases
Solution Approach 1:
The patent introduces a generator-discriminator network as an intermediary between rule-based systems and transaction assessment. The generator creates synthetic illicit transaction patterns that train the discriminator, which then serves as a smart filter before rule-based detection, reducing false positives while maintaining detection coverage
Solution Approach 2:
The discriminator provides feedback signals to the generator during training, creating a closed-loop system where detection performance continuously improves. The discriminator's assessments of generated samples feed back to refine the generator's ability to create realistic illicit patterns, enhancing overall detection accuracy
2Measurement precision
If supervised machine learning models are used, then detection accuracy is improved, but label availability decreases
Solution Approach 1:
The generator creates synthetic copies of illicit transaction patterns through adversarial training. These copied patterns serve as artificial labels that augment the scarce real labeled data, enabling supervised learning models to achieve high detection accuracy without requiring large amounts of actual labeled illicit transactions
Solution Approach 2:
The generator-discriminator system performs preliminary action by pre-training the discriminator on synthetic data before deployment. This preliminary training phase creates a well-performing model that can then be applied to real transactions, overcoming the limitation of scarce labels in advance
3Quantity of substance
If unsupervised training techniques are used, then labeled data requirement is reduced, but detection effectiveness decreases
Solution Approach 1:
The system transitions from static unsupervised pattern discovery to dynamic adversarial training. The generator and discriminator continuously adapt to each other, with the generator learning to create more sophisticated illicit patterns and the discriminator learning to detect them, maintaining high detection effectiveness with minimal labeled data
Solution Approach 2:
The patent changes the training paradigm from unsupervised parameter optimization to adversarial parameter optimization. By introducing the generator-discriminator framework, the system optimizes parameters through competitive learning rather than simple pattern clustering, significantly improving detection effectiveness while keeping labeled data requirements low
4Quantity of substance
If GANs are used to generate labeled data, then data availability is improved, but model adaptability decreases
Solution Approach 1:
The generator is designed to be dynamic and adaptive, continuously evolving its ability to create new illicit patterns through adversarial training. This dynamic nature allows the model to adapt to emerging money laundering strategies while maintaining adequate data availability for training
Solution Approach 2:
The system performs preliminary adversarial training to establish a baseline generator capability. This preliminary action creates a foundation that can be further adapted to specific domains and emerging threats, balancing data generation capability with model adaptability
Data Source
AI summary
In various embodiments, a process for assessing transactional graphs based on generator-discriminator networks includes using a generator network to generate a first set of transaction graph samples that are of a generated type, wherein the generator network is trained to optimize a predetermined objective function. The process includes sampling, from a collected dataset, a second set of transaction graph samples that are of a non-generated type; and providing the first set of transaction graph samples and the second set of transaction graph samples to a discriminator network, wherein the discriminator network is trained to classify a provided transaction graph sample as the generated type or the non-generated type. The process includes discriminating, by the discriminator network, each of at least a portion of the first set of transaction graph samples and the second set of transaction graph samples as the generated type or the non-generated type.