Open Source Powered Real-Time Fraud Analysis in a Hybrid Cloud Environment
Talkat Open Source in Finance Forum New York 2022 · on Thursday, 2022-12-08· at 16:15· HUB 2, New York, NY, USA·
summary
Demonstrates a pipeline that trains LSTM/Gru models in TensorFlow on Cloud Pak/Watson Studio using open source sample data, converts models with a TensorFlow→ONNX conversion step, and serves either via TensorFlow Serving or Watson Machine Learning on IBM Z with the Telum AI accelerator. Latency drops from ~50–100 ms to under 2 ms, raising scoring from ~20% to nearly 100% of transactions and delivering an estimated $20M risk reduction; security uses Cloud Object Storage + hyper protect keys.
AI-written summary · about
chapters (27)
- 0:00Introduction
- 0:39Agenda
- 1:19AI Impacts Every Industry
- 2:53Challenges in AI Adoption
- 3:32Ideal AI Stage
- 4:34Real World Use Cases
- 5:08Other Use Cases
- 6:30Customer Use Case
- 12:32How Open Source Helps
- 13:15AI Life Cycle
- 13:32AI Modeling
- 16:32Data Source
- 18:32Deploy
- 20:32TensorFlow Model
- 21:32Deep Learning Compiler
- 22:32Swagger API
- 23:32Data Security
- 24:32Data Protection
- 25:32IBM Cloud
- 26:32Model Deployment
- 27:32Machine Learning Models
- 28:32Open Source Models
- 29:32Cloud Object Storage
- 30:32Data Mapping
- 32:32Testing
- 33:32Summary
- 34:32Open Source in 10 years
related talks
13mTransforming Financial Services with Real-Time Analytics
38mReal-world Machine Learning with TensorFlow and Cloud AI
29mBuilding Resilient AI Platforms: Financial Fraud Detection
12mRevolutionizing Predictive Analytics
25mReal-Time Fraud Detection at Scale
51mReal-time Fraud Detection: Challenges and Solutions
55mLarge Scale Distributed AI on the JVM
47mReal-time Fraud Detection Challenges and Solutions