Benchmarking data agents for verifiable analytics over heterogeneous workspaces.
1st place (Phase 1) / 4th place + Merit Award (Phase 2 final) solution for the KDD Cup 2026 Data Agents Challenge
Industrial CVR recommendation case study: unified-token backbone, pairwise ranking, dense-only EMA, and noisy-label ablation discipline.
KDD Cup 2026 DataAgent-Bench: ReAct agent for heterogeneous data analysis with fault-tolerant JSON parsing and SQL verification
Are your model's probabilities trustworthy enough to spend $2 a mailer on? Calibrates logistic regression, random forest, and XGBoost with Platt/isotonic/temperature scaling and checks if the fix is worth the compute (tracked with CodeCarbon).
Competition-first traffic forecasting system for KDD Cup 2017 with strict anti-leakage guarantees and reproducible experiments.
Data analysis and machine learning on the KDD Cup 1999 network intrusion detection dataset
ML benchmarking of 15+ algorithms on KDD Cup 1999 dataset achieving >99.9% accuracy for intrusion detection
Execution-aware NL2SQL, document-to-database conversion, and DataAgent augmentation research prototype
To associate your repository with the kdd-cup topic, visit your repo's landing page and select "manage topics."