標籤:

Notes: Dr. Fei Fei Li on Innovate and Celebrate 2016

Dr. Fei Fei Li of Stanfords Artificial Intelligence Lab | Innovate and Celebrate 2016

https://www.youtube.com/watch?v=IXxh5C9iKFE

  • After 540 million years, intelligent animals like us use vision to survive, to navigate, to work, to entertain and to communicate. It has become the most important piece of our intelligence. In fact, our brain spends half of its neuronal process in vision processing. It is the most important sensory perception and cognitive system in our brain.
  • Summer Vision Program - 1966 MIT first attempt to solve the vision problem.
  • Goal: total scene understanding
  • There was enough communication going on in your brain that reconstructed the most plausible 3D scene based on these 2D images. Vision is beyond jus measuring pixels.
    • Plato described the problem of vision as the problem of the prisoners of allegory of the cave. He said that vision is fundamentally about reconstructing or re-interpreting what you see as if you are prisoners tied on the chairs and you are forced to look ahead on the blank wall in front of you. and the wall is projecting shadows on the back of the prisoners head, so the prisoners are forced to only look at the 2D projection of the 3D plate in the back of their head. Their job is to reconstruct whats happening in the back of their head. And that is what vision is like for all of us. And thats problem we need to solve.
  • Key developments in machine learning in 1950s (neuron network) - 1990s laid foundation for image recognition.
  • Object Recognition in 1990s and 2000s, first face recognition paper in 2000, first face recognition: 2006 Fujifilm
  • Babies take 5 pictures / second using their eyes, by age of 3 taking hundreds of millions of pictures - huge amount of data sets generated from human vision - need to marry machine learning with big data - ImageNet.
  • Image Captioning: 1/ how do we process images (CNN) 2/ how can we generate a story (RNN)
  • 500 million years later, computer vision will also be the enabling technology to give us a technological Cambrian explosion. And we are not far from that day.

推薦閱讀:

人工智慧的過去、現在、未來和幻想
人工智慧電視與普通智能電視有什麼區別?
[探科技]進擊的人工智慧
當機器學會畫畫
《數據科學家訪談錄》讀書筆記

TAG:人工智慧 |