Day 2 - ML Basics and Video Recommendation System

[henrywu]

ML Basics Topics:

• Handle imbalance class distribution
• Choose the right loss function
• Retraining requirements
• Non-stationary problem: Bayesian Logistic Regression (paper: https://quinonero.net/Publications/AdPr ... -final.pdf)
• Exploration vs. exploitation: Thompson Sampling
• Offline/Online metrics & AB Testing

Video Recommendation System Topics:

1. Problem statement
2. Metrics design and requirements
3. Metrics
   • Offline metrics
   • Online metrics
     Requirements
   • Training
   • Inference
     Summary
4. Multi-stage models
   Candidate generation model
   Feature engineering
   Training data
   Model
   Ranking model
   Feature engineering
   Training data
   Model
5. Calculation & estimation
   Assumptions
   Data size
   Bandwidth
   Scale
6. System design
   High-level system design
   Challenges
   Huge data size
   Imbalance data
   High availability
7. Scale the design
8. Follow up questions
9. Summary

Reference:

https://en.wikipedia.org/wiki/Matrix_fa ... 20matrices
https://en.wikipedia.org/wiki/Collaborative_filtering
https://ai.googleblog.com/2020/07/annou ... ector.html
https://engineering.fb.com/2017/03/29/d ... ty-search/

[mldev]

https://keras.io/examples/structured_da ... nsformers/