Remember What You Learn (Spaced Repetition Model)
summary
Claims that Half-Life Regression fits the Ebbinghaus forgetting curve to real learner logs and that Duolingo uses it to schedule reviews. Implements a simple exponential retention model, trains with gradient descent on a regularized squared‑loss objective, and evaluates with MAE, Spearman and ROC. Uses Python stack (pandas, numpy, scikit‑learn, matplotlib) on ~100k+ students and ~2k words with hundreds of millions of traces to estimate half‑life parameters and reduce forgetting in practice.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:29Forgetting curve and spaced repetition
- 5:02Duolingo's spaced repetition
- 7:13ML model components
- 9:27Python for data science
- 13:12Half-life regression model
- 17:37Model performance metrics
- 20:50ML model best practices
- 23:03Key takeaways
- 24:51Python libraries for ML
- 25:57Feature engineering
- 27:43Spaced repetition in quant fields
- 28:26Optimizing online learning
related talks
32mThe Importance of Forgetting for Both Humans & Machines
40mHuman Learning
36mLanguage Model Post-Training in 2025: an Overview of Customization Options Today
01h 17mOnline Reinforcement Learning and Regret
18mMaximizing Luck in Reinforcement Learning
01h 00mOn continual learning with gradient descent for neural networks
13mTowards Lifecycle Unlearning Commitment Management: Measuring Sample-level...
35mJe ne regrette rien: Teaching Machine Learning Models Regret Avoidance