Tips on Quality Assurance process in Machine Learning projects
summary
Claims practical QA focus: most effort is in data collection and assessment, not model code; build an application expert when no human expert exists; use variance-based feature removal to cut irrelevant inputs (example reduces five features to three with identical scores); visualize tokens to discover composite features (airport = person+location); impute missing values with mean imputation or multiple imputation; use pretrained models for rare cases (transfer learning); monitor feature drift in production and retrain regularly.
AI-written summary · about
chapters (9)
- 0:00Introduction to QA in ML
- 3:00Assessing data in ML
- 10:09Visualizing data for insights
- 13:00Handling missing data
- 15:29Training models with limited data
- 17:02Transfer learning approach
- 21:46Production data mutation
- 23:11Long-term ML project support
- 24:39Prototyping and flexibility
related talks
14mEarly Test Strategies in ML Model Development Phase
43mQuality Of Machine Learning Models
42mQA for ML: How we can trust AI with Food Sustainability
53mUnleashing the Black Box: Testing Machine Learning Systems
30mFrom Black Box to a Known Quantity: How to Build Predictable, Reliable ML-based
32mTesting Machine Learning Systems: What it is and why you should care
32mAn Only One Step Ahead Guide for Machine Learning Projects
30mHow to Properly Test ML Models & Data