Systems and Software for Machine Learning at Scale with Jeff Dean - TWiML Talk #121


:star: :star: :star: :star:

Notes

  • Background:
    • Moved around a ton. 11 schools as a kid
    • Undergrad research on neural networks. Not enough compute
    • Joined Google in 1999
    • Worked on MapReduce. Reliably at scale
  • Beginning of resurrection of neural networks around 2008
  • Next 5 years (from 2018): automated machine learning. E.g. maximize throughput through intersections in cities. Should be universally available to all cities, not just those with deep learning specialists

  • Differential privacy is a mathematical definition for the privacy loss that results to individuals when their private information is used in the creation of a data product.