Genetic Regex
summary
An end-to-end system genetic programming to build regexes combines evolved patterns with a classifier that uses regex matches as binary features, trains and evaluates each candidate via cross-validation, and parallelizes training with a distributed island model. It reports an 85% F1 baseline on an SMS spam corpus and reaches 90% with stacking on character 3-grams, demonstrates evolved patterns for acronyms, legal forms, URLs and digit sequences, and implements crossover/mutation over regex trees.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:46Text classification at ING
- 2:33Spam detection with GP
- 3:56Evaluating regex features
- 4:39Genetic Programming fundamentals
- 6:26Crossover and mutation
- 8:19Algorithm results on spam data
- 9:24Optimizing GA performance
- 10:51Island model for diversity
- 11:12Stacking regex with n-grams
- 12:18Final results
related talks
15mAccelerating Rule-matching Systems with Learned Rankers
30mOutwitting Deep Learning Models
16mScalable Synthesis of Regular Expressions From Only Positive Examples
29mSurvival of the Safest: Automating LLM Defense with Genetic Programming
32mApisecure, 14 & 15 March 2023 - Meenakshi Kaushik, Leader - Product Management at Cisco.
46mTight Genes: Intro to Genetic Algorithms
57mSolutions That Evolve: Building Self-Improving Systems with Genetic Algorithms
01h 19mAnalyzing Genetic Networks Using Neural Networks