Overview
This course covers how to implement various flavors of production ML systems, including:
- Static, dynamic, and continuous training
- Static and dynamic inference
- Batch and online processing
You will delve into TensorFlow abstraction levels, explore options for distributed training, and learn how to write distributed training models using custom estimators.
Why Choose KORNERSTONE
- Accredited, practitioner-led, expert certificate trainers to provide high quality training
- Official training material included
- Practical, scenario-based learning – focus on application, not just theory
- Excellent Passing Rate
Skills Covered
Upon completion of this course, learners will be able to:
- Differentiate between static, dynamic, and continuous training pipelines.
- Implement static and dynamic inference for production models.
- Choose between batch and online processing based on use case requirements.
- Navigate TensorFlow abstraction levels (from high-level Keras to low-level custom ops).
- Set up and manage distributed training jobs on Google Cloud.
- Write distributed training models using custom estimators.
- Apply best practices for productionizing ML systems.
Prerequisites
- Completion of Course 1 in the Advanced Machine Learning on Google Cloud series (recommended)
- Working knowledge of TensorFlow (including Keras)
- Familiarity with Python and basic cloud concepts
Target Audience
- Cloud Architects designing ML pipelines
- Intermediate Machine Learning Engineers
- Data Scientists moving models to production
- Learners who have completed the first course in the Advanced ML on Google Cloud series
Course Curriculum
Module 1: Introduction to Production ML Systems
- Production challenges
- Static vs. dynamic vs. continuous training
- Batch vs. online processing
Module 2: Inference in Production
- Static inference (precomputed)
- Dynamic inference (on-demand)
- Latency and throughput considerations
Module 3: TensorFlow Abstraction Layers
- TF 2.x ecosystem
- Estimators, Keras, and custom loops
- When to use which abstraction
Module 4: Distributed Training Fundamentals
- Why distribute training
- Data parallelism vs. model parallelism
- MirroredStrategy, TPUStrategy, MultiWorkerMirroredStrategy
Module 5: Custom Estimators for Distributed Training
- Writing custom estimators
- Model functions and input functions
- Lab: Distributed training with custom estimators
Module 6: Production Pipeline Architecture
- Continuous training pipelines
- Model versioning and rollback
- Monitoring and alerting
Module 7: Challenge Lab (Skills Badge)
- Jump directly to a challenge lab
- Demonstrate production ML skills without completing all modules
Dates & Locations
September 8, 2026 - September 9, 2026
Location:
Online
Format:
Live Virtual
Availability:
TBC
November 3, 2026 - November 4, 2026
Location:
Online
Format:
Live Virtual
Availability:
TBC
