Lat Designing and Building Complex Machine Learning....
summary
Shows a concrete stack: SageMaker Debugger rules trigger SNS alerts when training loss stalls, automated hyperparameter tuning runs many parallel jobs (and can span multiple model families) to choose the best model, and SHAP explainability jobs write feature attributions to S3 so regulators get interpretable results. Recommends combining serverless (API Gateway + Lambda) with containers, using lambda container images for lightweight deployments, running multiple models in one endpoint for A/B testing, and swapping long cloud steps with fast lambdas while iterating to save time and cost.
AI-written summary · about
chapters (25)
- 0:00Intro
- 1:07About Joshua
- 2:06What is Machine Learning
- 3:13The Simplified Process
- 6:03The Hammer
- 7:17Serverless
- 9:35Containers
- 10:35Build from scratch
- 13:13Flexibility
- 14:31Container Technologies
- 16:17Custom Flask API
- 17:18Hyperparameter Tuning
- 19:33SageMaker
- 21:04ML Explainability
- 22:13ML Example
- 23:53Platform Support
- 25:42Automation
- 26:24Workflow Tools
- 28:39Siegemaker Pipeline
- 29:01Sagemaker SDK
- 29:32Python SDK
- 29:50Testing
- 31:25Over Engineer
- 31:59Capability Experience
- 32:50Conclusion
related talks
43mDesigning and Building Complex Machine Learning Projects
33mMachine Learning Engineering Done Right
37mStreamlining Machine Learning Operations
37mStreamlining Machine Learning Operations
44mPragmatic Machine Learning in the Cloud
20mIndustrial-Scale Machine Learning
34mDeploying a Machine Learning System That Only Runs Locally
25mDesigning an AWS Well Architected Framework for AI/ML