Algorithm and prototype for machine-learning-based financial fraud detection using programmable network devices
Smart algorithm detects financial fraud using machine learning in programmable network devices, ensuring quick transaction fraud detection with low latency.
Applications: Financial fraud detection for banks, hedge funds, security providers and financial supervision organisations
| Features | Benefits |
|---|---|
| Algorithm conducts high-throughput and low-latency transaction fraud detection within programmable network devices (PNDs) | Computing tasks run in servers which are very resource intensive; Using PNDs (e.g., switch-ASICs, GPUs) has the advantage of lower latency, higher throughput and higher energy efficiency, compared with servers |
| The science behind the IP is in-network computing and in-network machine learning, where ML models are mapped into the data plane | Means transferring computing task such as machine learning inference into PNDs; Computing tasks are offloaded into the data plane, leading to reduced dependency on servers and lower costs, while also freeing up server resources |
| Prototype has shown the ability to process transactions 2-3 orders of magnitude faster than existing solutions (i.e. x100-x1000), along with 2-3orders of magnitude latency reduction compared with a server-based benchmark | Creates faster and more efficient fraud detection, highly relevant for financial transaction fraud detection in entities such as financial supervision organisations, security teams in banks and providers of security solutions to banks |
| The algorithm has been trained and evaluated on datasets from banks, and the academics are exploring other time-sensitive financial applications | The IP could be used in real-time anomaly detection in fields other than finance |
Available For
- Co-development
- Consulting
- Licensing