ALERT: Machine Learning-Enhanced Risk Estimation for Databases Supporting...
summary
Transforms previous heuristic optimization into an ML multiclassification pipeline that trains on public auxiliary data to predict keywords online, uses dynamic clustering of queries to avoid huge class counts, and applies a co-occurrence optimization that lowers complexity from O(k n^2) to O(k n). The system runs inference in seconds, reports a 5.4 s median runtime with 86.3% median recovery (66.8% at 1.8 s), shows a 14.5× speedup with ~5.2% recovery loss versus Jigsaw, and drops only ~4.5% under padding.
AI-written summary · about
chapters (8)
- 0:00ALERT and DSSE basics
- 2:19Leakage abuse attacks motivation
- 4:05Risk assessment as ML
- 6:13Training data challenges
- 8:03Dynamic clustering for queries
- 9:51Optimizing co-occurrence matrix
- 11:19Evaluation results comparison
- 13:54Conclusion and resources
related talks
31mIn-database machine learning: Automate, scale, deploy
46mShorten the Path to Production with In-Database Machine Learning
26mKeeping Sensitive Data Safe Using Recommendation Systems
01h 01mD1T2 - Applying Machine Learning to User Behavior Anomaly Analysis
47mApproximate Lifted Inference with Probabilistic Databases
26mAI-Powered Predictive Analytics: Optimize Performance
31mLeverage AI to Protect Your Data Assets... from AI
39mDetecting Website Intrusion and Account Compromise with Machine Learning