Outwitting Deep Learning Models
summary
Shows concrete deployments where data-driven regular expressions power real-time COVID-19 detection in 64 languages, run from SQL/Python/PHP, handle misspellings and emerging terms, yield tier-one precision ~98.99% in English and reduce manual checks to ~3–15% per language; a distilled regex pipeline is far faster and more explainable than DNNs. For 20+ languages of clickbait detection a regularized Logistic Regression on word and character n-grams beats a BiGRU by AUROC and pointwise precision/recall, and simple ensembling with the BiGRU improves results further.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 3:50Font Configurations
- 4:30Practical Considerations
- 6:50Examples
- 8:05Integrity applications
- 10:10Challenges
- 11:43Dealbreaker
- 13:20Regular Expressions
- 15:22Keywords
- 20:06Keyword Discovery
- 21:31Manual Evaluation
- 22:32Automation
- 23:34Summary
- 25:05Comparison
- 26:32Problem Analysis
- 27:48Generic Solution
- 28:20Conclusion
related talks
13mGenetic Regex
37mRecent Solutions Using Deep Learning Models/ The How and Why
15mAccelerating Rule-matching Systems with Learned Rankers
46mWhy data-driven methods will shape the future (...)
31mDeep Learning Techniques for Natural Language Processing
39mDeep Learning vs. Conventional Machine Learning
35mIntelligent Ruby: Getting Started with Machine Learning
31mRemaining Challenges in Deep Learning based NLP