Playing by the Rules-Based-Systems
summary
Rule-based patterns in spacey produce labeled examples that train a machine learning model and reveal concrete differences: the ML model shows higher precision but lower recall, while a simple matcher finds unseen classes like common lisp because there is no training data. An iterative loop labels disagreements, raising recall without losing precision. A hard safety rule in rasa guarantees “not a human,” and the human learn package turns functions and drawn polygons into scikit-learn-compatible classifiers for benchmarking and data-driven rule construction.
AI-written summary · about
chapters (9)
- 0:00Rule-based systems and AI
- 1:49Spacy and Rasa projects
- 7:45Rule-based NER for programming
- 15:55Combining rules and machine learning effectively
- 19:25Rules for non-text data
- 22:14HumanLearn package
- 27:52Models as benchmarks
- 30:01Why systems must follow rules
- 31:25Q&A
related talks
26mHow Machine Learning Can Help
15mMachine learning: Necessity vs obligation
32mThe Miseducation of This Machine
26mFew-Shot Learning in Conversational AI
15mAccelerating Rule-matching Systems with Learned Rankers
54mMachine Learning explained with Examples
09mMachine learning algorithms, choosing the correct algorithm for your problem
36mLearning about machine learning from kids