DBTB INT Omar Alonso
summary
Principal data scientist at Microsoft describing how to improve model outcomes by focusing on good quality labels rather than cutting-edge models, outlining a workflow that combines machines and humans for labeling, reuses classical algorithms from older textbooks, and treats daily data exploration as a debugging habit. Claims that labeling is the “boring” but critical part of ML, emphasizes practical tool mixing (manual review, algorithmic pre-labeling, iterative correction) and argues this approach materially raises model reliability.
AI-written summary · about
related talks
21mDataEngBytes 2023 - SYD-T2-04 - Suneeta Mall
38mAutomatic DB annotation with ML tools and its problems
40mData labelling to scale your ML and AI pipelines
42mClean Labels for ML
20mGreenfield vs. Brownfield Data Labeling
04mManaging Machines
31mReliable Data for Large ML Models: Principles and Practices
38mBuilding robust machine learning systems