Ben Hamner, Co-founder and CTO, Kaggle
summary
Kaggle reports a 410,000-strong community making about 100,000 submissions per month and shows concrete wins: a diabetic retinopathy model at ~85% agreement with human graders, a toxicity predictor with ~26% improvement over prior benchmarks, and frequent use of XGBoost and Caris as winning tools. Competitors iterate rapid prototyping and feature engineering, almost always ensemble models for marginal gains, and repeatedly expose pitfalls like data leakage (stratified IDs, time mishandling, duplicate rows) that skew leaderboards and real-world deployment.
AI-written summary · about
chapters (9)
- 0:00Start
- 1:05Automatically Grading Student Written Essays
- 1:33Predicting a Compounds Toxicity Given Its Molecular Structure
- 2:13Automatically Detecting Diabetic Retinopathy
- 4:33Leaderboard Mechanic
- 5:00The Roger Bannister Effect
- 12:53The Brute of Feature Selection Algorithm
- 14:05Data Leakage
- 15:15Computationally Efficient Solutions
related talks
21mBen Hamner, CTO, Kaggle, NYC 2017
41mWhat Kaggle has learned from 2MM machine learning models
12mCommunity matters: 8 reasons why you should be involved with
03mDBTB INT Anthony Goldbloom r
50mInnovation and data analytics: Solving real-world challenges
24mJosh Wills, Head of Data Engineering, Slack
06mTop 5 takeaways
58mImproving disaster response with machine learning