The Missing Step in Moving from Data to Models
summary
Argues that the real leverage sits in feature engineering not newer models and shows concrete transforms that let older models work with less data while improving stability and interpretability. Demonstrates binarization and adaptive binning, log transforms on long tails, and min-max scaling and L2 normalization to bound and speed training. Covers text pipelines from bag-of-words to n-grams and parts-of-speech, warns about embedding bias, emphasizes human-in-the-loop accountability and privacy-by-design, and points to an accompanying GitHub repo with runnable examples.
AI-written summary · about
chapters (14)
- 0:00Introduction to Concur Labs
- 2:00Speaker background and journey
- 4:05The reality of data science
- 6:12Importance of feature engineering
- 7:51Defining key terms and concepts
- 9:46Data ethics and accountability
- 12:26Framework for feature engineering
- 16:29Mathematics of feature space
- 20:07Binarization techniques
- 22:23Binning and quantization
- 26:54Scaling and normalization
- 31:40Text processing and analysis
- 39:17Advanced text techniques
- 41:37Q&A and conclusion
related talks
31mLife cycle of data becoming a model
37mMachine Learning, The Right Way
46mWhy data-driven methods will shape the future (...)
39mDoing Data: The Critical Process of Data Preparation (...)
41mModels Schm-odels: Why you Should Care about Data-Centric AI
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
04mThe age of machine learning
07mPredictive Modeling and Machine Learning