Alex Korbonits, Data Scientist, Remitly
summary
Describes Remitly’s hybrid risk-management system that combines hand-crafted logical rules with machine-learning models, and a framework that learns concrete risk rules from data and ensembles rule sets to optimize a chosen objective. Covers how rules are generated from models, how multiple rule-producing models are combined into optimal rule sets, and the role of engineered input features derived from complex ML models.
AI-written summary · about
chapters (9)
- 0:00Start
- 6:42Risk management and risk rules
- 8:14Generating rules from machine learning models
- 9:42FOIL (First Order Inductive Learner)
- 12:18Decision Trees
- 13:19Synthesizing Production Rules
- 15:38Incremental Rule Ranking
- 20:10Rule inclusion/exclusion criteria
- 23:59Black box input features
related talks
12mInterview with – Kristina Reicher
34mSynergizing model and human decision-making using augmented machine learning
23mChallenges in using machine learning for financial monitoring
22mMachine Learning–Driven Third-Party Risk Management at Scale
37mThe Role of Machine Learning in Today’s Payment Ecosystems
18mAnti-Money Laundering ML Modelling Approach
37mFraud Prevention, Machine Learning, and Design Patterns: Keep Your Analysts in the Loop
13mMLOps in FinTech: Architecting Responsible AI