exciting practical applications of scalable deep learning
play

Exciting practical applications of scalable deep learning and image - PowerPoint PPT Presentation

Exciting practical applications of scalable deep learning and image recognition in the cloud Georgi Kadrev @georgikadrev Imagga Technologies @imagga GTC16, April 5th 150+ companies 3,700+ developers all around the world 3Y 150Y 2B+ NEW


  1. Exciting practical applications of scalable deep learning and image recognition in the cloud Georgi Kadrev @georgikadrev Imagga Technologies @imagga GTC’16, April 5th

  2. 150+ companies 3,700+ developers all around the world

  3. 3Y 150Y 2B+ NEW PHOTOS SHARED EVERY DAY (just to be forgotten in 30 minutes!)

  4. All the buzz words… • Artificial Intelligence • Machine Learning • Deep Learning • DNN • CNN • …

  5. Why not solving the problem using these via an auto-tagging API ?!

  6. As simple as: 1. Submit an image 2. Get list of tags or categories 3. Do whatever you need with them

  7. The tags • physical objects: car, dog, computer, … • scenery and conceptual: office, job, family, … • related terms: car -> vehicle

  8. The categories • personal topics: events, beach, street view, … • safe/unsafe content • or customer defined:

  9. But who would need that… ?

  10. Unsplash: for search • Technologies used: auto-tagging • Impact: reduces/replaces manual tagging, enhances search 10

  11. KIA: for user profiling and advertising • Technologies used: auto-tagging, color extraction • Impact: very precise personalised targeting 11

  12. Tavisca: for sorting out 25M hotel photos • Technologies used: custom auto-categorization • Impact: automates classification, improves browsing experience 12

  13. Seoul National University: for waste sorting • Technologies used: custom auto-categorization • Impact: automates pre-sorting of waste Image source: Wikipedia

  14. Eden: for organizing personal photos http://edenphotos.io

  15. Are we close to human performance? 32.5 % 51.1 % https://demo.algolia.com/clashOfTags/

  16. Thank You! api@imagga.com twitter.com/imagga facebook.com/imagga

  17. Hipster Bar: to let only hipsters in his bar! :)

  18. How we’ve built it? • a lot of image data • precise model optimization • scalable infrastructure

  19. The data challenge and our feedback loop

  20. The specific content challenge and our customer-defined training

  21. The high throughput challenge and on demand infrastructure

Recommend


More recommend