Transfer Learning for the Cold Start Problem
summary
A fraud-prevention pipeline for Zalando uses transfer learning and domain adaptation to solve a payment fraud cold-start when a new market has few labels. The approach trains models on labeled data from mature markets, aligns features across markets, fine-tunes on tiny samples from the target market, and combines models in an ensemble with cross-market validation; the ensemble is deployed and incrementally retrained as local labels arrive, producing an initial working fraud model that improves over time.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 3:14Cold start in e-commerce
- 10:01Black Friday campaign
- 17:43Transfer learning for cold start
- 25:55Key takeaways
- 29:46Q&A
related talks
36mContinuous Monitoring of Machine Learning Models
16mLeveraging ML to Create a Reliable Anti-Fraud System
25mUnsupervised shallow learning for fraud detection on marketplaces
47mTrain, Evaluate, Repeat: Building a Credit Card Fraud Detection System
45mAnti-Fraud in Reverse and the Use of ML Methods in It
45mTransfer Learning - Entering a new era in NLP
41mUsing Machine Learning to Fight Fraud
19mGenerating Adversarial Examples & Defense Methods For Online Fraud Detection