Machines and the Magic of Fast Learning - Strata Keynote
Download as PPTX, PDF1 like2,396 views
The document discusses the challenges and statistical data related to child sexual abuse imagery and online trafficking, highlighting a significant increase in such abuses since 2007, with 42% of underage victims being advertised online. It emphasizes the importance of data science and machine learning in identifying and combating these issues through a feedback loop for precision tuning. Additionally, it notes the vast amounts of escort ads and the efforts of organizations like Thorn in processing this data to protect children.
Machines and the Magic of Fast Learning - Strata Keynote
- 2. DATA CORPUS
- 5. DATA CORPUS ACTORS Applications OPERATORS Data Science ACTORS + OPERATORS Real-Time Machine Learning
- 6. ACTORS Applications OPERATORS Data Science ACTORS + OPERATORS Real-Time Machine Learning FEEDBACK LOOP
- 7. PRECISION
- 8. TUNING
- 9. DISCOVERY
- 11. 63% 5,000% 42% Of underage sex trafficking victims are advertised online Increase in child sexual abuse imagery since 2007 Of victims met perpetrators online
- 12. DIGITAL DEFENSE IS A BIG AND FAST DATA CHALLENGE
- 13. 100k escort ads posted every day in this country THORN SIFTS THROUGH MASSIVE AMOUNTS OF DATA
- 16. 4,996 POINTS CLASSIFICATION DE-DUPLICATION MATCHING
- 19. select id, description, vector_dot(images.val, UNHEX("XXXXXXXX")) as distance from images having distance > CONSTANT; • SIMD Euclidean Measurement Between Points • 1,000x match improvement time THE MISSING DATABASE SYNTAX
- 20. 2000+children saved in 2016 REAL-TIME HAS REAL WORLD IMPACT
- 21. 21 Putting Fast Learning Practices To Work
- 22. MAPPING SOCIAL IMAGERY HANDWRITING
- 23. Application Model Evaluation and Regression TensorFlow Reference Image Store New Image Stream MACHINE LEARNING FOR IMAGE RECOGNITION
- 24. Real-Time Application New Image Stream Real-Time Logistic Regression Reference Image Store High Speed Vector Algebra TensorFlow BUILDING FAST LEARNING FOR IMAGE RECOGNITION
- 25. REAL-TIME ANALYTICS AT UBER-SCALE 11:00am LL20 B BUILDING REAL-TIME DASHBOARDS 1:50pm 210 D/H LATER TODAY
- 26. THANK YOU