Ashrith Barthur, Security Scientist, H2o.ai at MLconf Seattle 2017
summary
Claims a production ML pipeline at H2O that uses time-bound features and heavy feature engineering (90% of effort) to spot AML behavior, trains interpretable tree models like a random forest (25 trees, depth 5), and labels alerts via diverse investigators. Tests report cutting false positives by about 60%, with best-case identification of false positives at ~95% and worst at ~79%, and a goal to move toward data-driven deep models plus surrogate explanation (LIME/SHAP) for regulators.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:42What is Money Laundering
- 3:52Money Laundering Impacts Society
- 5:22Solution for Money Laundering
- 6:10What is the next solution
- 6:29What does H2oai do
- 6:46Regulation
- 8:53Current Day Design
- 10:10Limitations
- 12:44Featurebased approach
- 13:52Sources of data
- 15:00Features
- 18:36Implementation
- 19:49Advantages
- 21:56Investigator
- 23:22Datadriven model
- 23:40Pure datadriven model
- 25:01Black box model
related talks
29mThresholds Are for Old Threats: Demystifying AI and Machine Learning to Enhance SOC Detection
04mInterview
49mEnhanced AML fraud detection solutions with Azure Machine Learning
10mML-based Log Analysis for Faster Debugging
25mHow AI Transforms Enterprise Data Pipelines with 85% Gains
23mChallenges in using machine learning for financial monitoring
55mApisecure, 14 & 15 March 2023 - Sagar Bhure, Security Engineer at F5.
27mDeploying distributed AI and machine learning in financial services