Machine Learning: Beyond the Tutorial
summary
Goes beyond tutorials by showing concrete pitfalls and fixes: uses MNIST 99.78% to illustrate label errors; demonstrates normalization and feature scaling to avoid skew, and SMOTE to rebalance 99:1 fraud datasets. Recommends k‑fold cross‑validation and Google Colab for heavy training. Describes a Twitter + Google Cloud NLP trading prototype that shorts stocks after specific Trump tweets and reports high short‑term success, notes mobile limits (face‑ID glasses need ~20s) and stresses ethical accountability and regulatory risk.
AI-written summary · about
chapters (13)
- 0:00Introduction and welcome
- 1:47Limitations of tutorials
- 4:35The magic of data
- 9:10Data imbalance and bias
- 14:12Data validity over time
- 17:23Sources of data
- 20:34Beyond data collection
- 24:44Training and testing models
- 25:04Machine learning on mobile
- 26:48Hardware constraints
- 27:08Ethical accountability
- 28:09Unexpected outcomes
- 28:51Key takeaways summary
related talks
52mMachine Learning Crash Course
42mMachine Learning Crash Course
22mReal-world applications of machine learning: Beyond the hype
33mRunning Machine Learning in Production - a Journey to Success
36mApplied Machine Learning Playbook: From Idea to Production
06mInterview
48mLearning Machine Learning
30mMachine Learning Techniques for Class Imbalances & Adversaries