Quality Assurance in Projects with Machine Learning
summary
A concrete Python pipeline for a Cyrillic-letter recognition ML project is described: data collection (~450 images for each of 33 letters), cleaning and augmentation, neural-network training, and deployment via a server API plus a mobile app that sends drawn letters for prediction. The project sets per-letter accuracy targets (95–98%) and highlights measurable pitfalls—overfitting (e.g., 99% train vs 20% validation), missing classes, adversarial pixel masks—and recommends ongoing monitoring and periodic re‑training.
AI-written summary · about
chapters (9)
- 0:00ML project testing intro
- 3:02Why QA needs ML
- 8:18Requirements and datasets
- 16:30Data quality checks
- 23:40Model training accuracy
- 29:49Model deployment testing
- 31:55Mobile app and UX
- 32:40Model monitoring updates
- 34:05Conclusion
related talks
26mTips on Quality Assurance process in Machine Learning projects
32mTesting Machine Learning Systems: What it is and why you should care
43mDesigning for Guidance in Machine Learning
33mRunning Machine Learning in Production - a Journey to Success
54mMachine Learning explained with Examples
42mMachine Learning Crash Course
34mML from a software engineer's perspective
40mBuilding a Handwriting Recognition System