Welcome to Day 1 of the Secure AI Competition. You are part of a consortium of five banks aiming to catch fraud and network-intrusion patterns without ever pooling customer data, which is legally forbidden. The solution lies in federated learning (FL): each bank trains a model locally on its own data, sharing only the trained model updates (never raw data) to average them into one global model.
Your task is to improve a working but weak federated intrusion-detection system currently trained on NSL-KDD. Full technical details and a working baseline are provided, allowing you to enhance a real system rather than building from scratch. This page covers Day 1 specifically; the full hackathon includes two additional tracks that unlock tomorrow via the main CAIRLab Secure AI Hackathon page.
To get started, copy the Day 1 starter notebook into your Kaggle Notebook or Colab, run every cell top to bottom to see the working baseline, and make your improvements in the section marked 🔧 YOUR TURN. Once complete, run the submission cell to generate submission.csv and upload it here.
Scoring is fully automated with no writeup or video required for this leaderboard. Your model's predictions are scored against a held-out test set using precision, recall, and F1 metrics (not accuracy). You may submit as many times as the daily limit allows, with your best score counting unless you select a different final submission. Your Day 1 result also contributes to your overall hackathon score.