Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Scientists meet Entrepreneurs - AI & Machine Learning, Tambet Matiisen, University of Tartu

65 views

Published on

Scientists meet Entrepreneurs - AI & Machine Learning, Tambet Matiisen, University of Tartu

Published in: Data & Analytics
  • Be the first to comment

  • Be the first to like this

Scientists meet Entrepreneurs - AI & Machine Learning, Tambet Matiisen, University of Tartu

  1. 1. Software is eating the world. AI is going to eat software. Tambet Matiisen Mobile Monday 14.01.2019
  2. 2. AI success stories Image recognition Speech recognition Machine translation But what about? Algorithms? Databases? UI design?
  3. 3. Beyond perception: algorithms
  4. 4. Learning to imitate Photoshop filters Xu et al. Deep Edge-Aware Filters (2015)
  5. 5. Learning to imitate any existing costly algorithm ● No shortage of training data - you can always generate more! ● The algorithm now runs in constant time and memory. ● Easy to make tradeoffs between accuracy and speed: ○ Want more accuracy? Add more layers! ○ Want faster running time? Remove some layers! ● Easy to port to new platforms or even bake into silicon! ● Examples: weather prediction, fluid simulation, ... ● Caveat: may need network architecture tuning. Andrej Karpathy. Software 2.0. https://medium.com/@karpathy/software-2-0-a64152b37c35
  6. 6. Beyond perception: databases
  7. 7. Learned indexes f(x) = p x1 x2 ... xp-1 xp xp+1 ... xn p - position of the record in sorted array x > xpx < xp Kraska et al. The Case for Learned Index Structures (2017)
  8. 8. Learned indexes Kraska et al. The Case for Learned Index Structures (2017) ● Increased lookup speed. ● Vastly less storage needed. ● From O(logn) to O(1)? Actually more like O(sqrt(n)), which is > O(logn).
  9. 9. Beyond perception: UI
  10. 10. Automatic generation of UI elements https://twitter.com/TeleportHQio/status/1043245039261044736
  11. 11. What you can do about it?
  12. 12. Learn AI! Stanford course “Convolutional Neural Networks for Visual Recognition” http://cs231n.github.io/ fast.ai course “Practical Deep Learning For Coders” https://course.fast.ai/ Andrew Ng deep learning courses https://www.deeplearning.ai/
  13. 13. Thank you! tambet@ut.ee

×