How a Kaggle Grandmaster cheated in $25,000 AI contest with hidden code – and was fired from dream SV job
kagglecompetitionfrauddata-science-ethicsmachine-learning
Abstraction: Kaggle Grandmaster Pavel Pleskov cheated $25k PetFinder competition using scraped test data
Key points:
- Pavel Pleskov (ranked 3rd globally on Kaggle) led team Bestpetting to win the $10,000 first prize in PetFinder.my's pet adoption prediction contest
- Cheat mechanism: scraped and hashed ~3,500 PetFinder profiles from the test set; the
majority_votingfunction secretly looked up actual answers instead of predicting them - The cheat only activated 10% of the time to avoid suspicion; without it Pleskov's team would have ranked ~100th
- Discovered by 19-year-old sixth-place finisher Benjamin Minixhofer when integrating winning code into production
- Pleskov had bragged about similar data-scraping on a prior Quora competition; Kaggle banned his account
- Pleskov was fired from H2O.ai after ~4 months; acknowledged chasing Kaggle ranking points over money
Connections: Kaggle · H2o AI · Petfinder · ML Competition · Data Science Ethics
Source: https://www.theregister.co.uk/2020/01/21/ai_kaggle_contest_cheat/