Google IO 2017 Sundar Pichai keynote highlights

Below are the highlights from Sundar Pichai’s keynote during the Google IO 2017 that was held on 17th May, 2017.

  1. YouTube: One billion users, who watch one billion hours of video daily.
  2. Google Maps: users navigate one billion kilometers every day.
  3. Google Drive: Launched in 2012, it has 800 Million+ users , there are three billion objects in google drive.
  4. Google Photos: Launched in 2015, has 500 Million+ users, every day users upload 1.2 billion photos.
  5. Android:  Two billion active devices.
  6. Mobile first to AI first.
  7. Smart Reply: Being released today to one billion Gmail users, machine learning systems learn to be conversational and automatically suggest response buttons.
  8. Voice and vision are two new modalities after keyboard and touch, to enable humans to interface with machine in natural in immersive ways.
  9. Speech Recognition word error rate is improving: July 2016: 8.5% , Oct 2016: 6.8%, Dec 2016: 6.1%, Current: 4.9%
  10. Two weeks ago Google Home started recognizing 6 individuals in a household, providing a personalized experience for each and everyone.
  11. Voice is becoming an important modality in Google products
  12. Google Photo can: Understand the attributes behind a picture, can improve quality of a low light, grainy picture, if there are obstructions in a picture, they can be automatically be removed.
  13. Google Lens: Being released today. Vision based computing capabilities, which will understand what you are looking at and help you take action based on that information. Will ship first in Google Assistant and Photos, and will come to other products later.
  14. Building AI first data center. Last year launched Tensor Processing Units(TPU), which are a custom hardware for machine learning .  They are 15 to 30 times faster and 30 to 80 times more power efficient than CPU’s and GPU’s at that time.  TPU’s are used across all Google products.
  15. Last years TPU’s optimized for inference.
  16. Cloud TPU’s are next generation TPU’s, each is optimized for training and inference.
  17. Each cloud TPU is capable of 180 Trillion FPU’s.
  18. A TPU pod is comprised of 64 Cloud TPU’s capable of 11.5 Petaflops.
  19. Cloud TPU’s are coming to Google Compute Engine as of today.
  20. Google is bringing all its AI efforts together in, which is a collection of efforts and teams across the company focused on bringing the benefits of AI to every one. Will focus on: Research, Tools, Applied AI

Link to the video: upto 56:44 mark.