How to efficiently model learner’s knowledge with...
summary
Recreates Deep Knowledge Tracing with recurrent nets and entity embeddings and reports concrete results and engineering lessons the model uses an embedding size of 50 and an LSTM with 200 hidden units trained with Adam and a weighted cross entropy loss on sequences up to thousands long on AWS single GPU the pipeline initially yields AUC ~0.9 but suffers class imbalance and reproducibility differences between CPU LSTM and cuDNN LSTM implementations experiment tracking with MLflow and switching to the TensorFlow Dataset API cuts training from 7 hours → 7 minutes
AI-written summary · about
chapters (7)
- 0:00Knowledge tracing intro
- 3:34Adaptive learning importance
- 7:10Established vs deep tracing
- 14:26Deep tracing implementation
- 19:36Model training challenges
- 26:19Future improvements
- 27:45Team recruitment
related talks
34mHow Deep is Deep Learning?
37mCode First, Math Later: Learning Neural Nets Through Implementation and Examples
40mHuman Learning
35mDeconstructing Deep Learning
01h 30mIntroduction to Deep Learning & Natural Language Processing
39mDeep Learning vs. Conventional Machine Learning
51mTensorFlow Wide & Deep: Advanced Classification the easy way
35mCompress language models to effective & resource-saving models with knowledge distillation