Loading slide

ImageNet and the Turning Point

  1. 01ImageNet and the Turning Point
  2. 02Fei-Fei Li and ImageNet
  3. 03The traditional approach
  4. 04AlexNet
  5. 05What it felt like
  6. 06What made it work
  7. 07How it taught itself
  8. 08Take a breath
  9. 09What did it actually learn?
  10. 10The field responds
  11. 11Why this was a hinge
  12. 12Memory card
  13. 13Reinforce your understanding
  14. 14Question: Data vs. algorithm
  15. 15Question: Hand-engineered vs. learned features
  16. 16Question: What does a turning point feel like from inside?
  17. 17Quiz: answer
  18. 18What you now know
  19. 19The machine could now see. The next question was whether it could read.
  20. 20Want to go deeper?
8 / 19
BackNext

Take a breath

If that last slide felt like a lot, let it go. You don't need to hold the mechanics of every nudge in your head.

Here's the only thing worth keeping: nobody taught AlexNet what a dog looks like. It taught itself, just by being wrong over and over and correcting a little each time.

That is the part to keep. Everything else is detail.

Which leaves one irresistible question. If the network worked all this out on its own, what did it actually figure out? What is hiding inside those 60 million numbers?

Citations