FOLD-SE: An Efficient Rule-based Machine Learning Algorithm with Scalable Explai...
summary
FOLD-SE is a rule learner that outputs default rules as an answer-set-style normal logic program, using a Magic Gini Impurity heuristic, a refined comparison operator for mixed numeric/categorical values, and a rule-pruning cutoff to remove long-tail rules. It matches XGBoost/MLP accuracy while running roughly an order of magnitude faster, fits a credit-card benchmark in ~53 MB versus ~38 GB for one-hot methods, and shrinks models from ~180 rules (RIPPER) to about 2.5 rules on the Rain in Australia dataset with higher accuracy.
AI-written summary · about
chapters (13)
- 0:00FOLD-SE introduction
- 1:24What is FOLD-SE
- 3:14Why explainability matters
- 5:21Default logic reasoning
- 8:35How FOLD learns rules
- 12:10Hybrid features handling
- 15:21Magic Genie impurity
- 18:12Rule pruning
- 21:29FOLD-SE vs others
- 24:01FOLD-SE efficiency
- 26:08Conclusion and key takeaways
- 26:52Inductive generalizations
- 29:04Q&A
related talks
38mPlaying by the Rules-Based-Systems
43mToward Human-Centered Machine Learning
15mMachine learning: Necessity vs obligation
42mJust-So Stories for AI: Explaining Black-Box Predictions
45mAGT and Machine Learning
36mEnhanced and Efficient Reasoning in Large Language Models
39mBeyond Bias: Crafting Fairness in Pythonic Machine Learning Systems
06mDecision rule elicitation for domain adaptation