How to build personalization models using Amazon SageMaker
summary
Shows how to train a TensorFlow multitask recommender on MovieLens using Amazon SageMaker Studio, TensorFlow Recommenders and SageMaker hyperparameter tuning, optimizing root mean squared error by sweeping embedding dimension (k=32 vs 64), learning rate (.00001–.00005) and epochs (10) on p3 GPU instances. Demonstrates Bayesian tuning (max 6 jobs), finds k=64 best, deploys a TensorFlow Serving endpoint and returns sensible movie recommendations, and outlines scaling to one model per user with micro‑pipelines and Spark processing jobs.
AI-written summary · about
chapters (15)
- 0:00Introduction to AWS ML
- 0:21Importance of personalization
- 1:02Business impact and benefits
- 3:53Challenges in recommenders
- 6:13Popular algorithm techniques
- 8:16Matrix factorization deep dive
- 11:17Multi-task recommender models
- 14:37Building on AWS
- 17:06SageMaker features overview
- 19:06Studio demo and algorithms
- 22:29Hyperparameter tuning
- 24:13TensorFlow recommender demo
- 28:24Model deployment
- 29:50Scaling to millions of models
- 31:13Summary and resources
related talks
19mDeveloping large-scale machine learning algorithms on Amazon SageMaker
45mBuild NLP Models with Amazon SageMaker
21mAWS On Air ft. Amazon SageMaker Inference Recommender
01h 21mUnlocking Personalization: A Deep Dive into Modern Recommendation Algorithms
39mTrain ML models at scale with Amazon SageMaker Training
27mScaling ML to the next level: Hosting thousands of models on SageMaker
33mTune your ML models to the highest accuracy with automatic model tuning
43mDesigning and Building Complex Machine Learning Projects