# FSDL 2022
## ML Lifecycle
- Problem formulation
- Data, training
## Data Management
- Feature stores, labeling
- DVC versioning
## Continual Learning
- Monitoring, drift detection
- Retraining triggers
## Infrastructure
- Testing, deployment
- Observability
Full Stack Deep Learning 2022
Lecture 01: When to Use ML and Course Vision
- Formulating problems and estimating project cost
- Sourcing, cleaning, processing, labeling data
- Picking the right framework and compute
- Troubleshooting training and ensuring reproducibility
- Deploying the model at scale
- Monitoring and continually improving
Lecture 02: Development Infrastructure & Tooling
Lecture 03: Testing
Lecture 04: Data Management
- Data sources: filesystem, object storage, database, data warehouse, data lake
- SQL and DataFrames (DASK, RAPIDS)
- Airflow, Prefect, Dagster
- Feature stores: Tecton, Feast, Featureform
- Labeling: Label Studio, Diffgram, Snorkel.ai
- Data versioning: DVC
Lecture 06: Continual Learning
- Monitoring metrics: data quality, distribution drift
- System monitoring: Datadog, Honeycomb, NewRelic
- Data curation: random, stratified, active learning
- Retraining triggers and dataset formation
- Online testing: shadow mode, A/B testing
#FSDL #DeepLearning #MLOps #FarshidPirahansiah