Platform Intrusion Detection with Deep Learning
summary
A production pipeline for platform intrusion detection uses Google BigQuery for data ingestion, manual feature engineering to produce numpy-based quantifiers, and trains Keras/TensorFlow models. Models are served via Flask microservices and autoscaled using Kubernetes autoscaling predictors to classify and cut bad-bot traffic on shop.rewe.de (2.5M daily visitors); the talk reports deployment metrics, false-positive handling, lessons learned, and which components reduced fraudulent sessions.
AI-written summary · about
chapters (7)
- 0:00Platform intrusion detection intro
- 9:17Feature engineering
- 19:47Model design and training
- 27:42Model performance evaluation
- 35:38Deployment challenges
- 38:06Case studies and results
- 42:17False positives and tuning
related talks
25mReal-Time Fraud Detection at Scale
18mKubernetes to serve a fraud detection model
31mBuild threat detection models using ML
30mOwned by Statistics: Using Kubeflow to Defend vs Attacks on Your ML Models
26mEngineering Excellence: Platforms and Architectures for Real-Time AI Analytics
39mDetecting Website Intrusion and Account Compromise with Machine Learning
15mIP Detection at Scale: CI/CD Meets Cloud-Native Enforcement
17mDevOps for Private AI: A Scalable Platform Using Kubeflow