Shared publicly  - 
Taming Latency Variability and Scaling Deep Learning

Recently, Google Senior Fellow +Jeff Dean presented the talk Taming Latency Variability and Scaling Deep Learning to the San Francisco Bay Area Professional Chapter of the Association for Computing Machinery ( Given in two parts, Dean’s talk first covers achieving low latency in shared environments with the goal of improving the overall user experience, e.g. updating the Google search results page as the user is typing.  

In the second part of the talk, starting at the ~20 minute mark, Dean speaks about the efforts in constructing computing systems that are able to automatically generate "understanding" of the raw audio, image, and textual data that is openly available on the Internet, by building high levels of abstraction in an unsupervised manner.    

While human beings are incredibly good at building abstractions, such as the ability to identify an object in an image regardless of its perspective, background, or context in which the image was taken, teaching computers how automatically identify objects in a similar manner is a challenging task.

Watch the video below to learn more about recent efforts in Machine Learning ( via Neural Networks (, which would allow computers to automatically understand data in a fashion similar to humans.
Chad DeChant's profile photoJim Rosswog's profile photoMatt Trifiro's profile photoEli Viertel's profile photo
Why is this post about object recognition but caries a title that references latency in shared digital environments?
+Randall Lee Reetz the talk is categorized into two parts. In the first 20 minutes-ish, he talks about low latency strategies in shared environments.
Why is the video now marked private? Any chance it can be made public again? thanks!
Very inspiring talk Jeff!   The scale of the data (and the problem) is amazing.  Sounds like a fun work.
This is one of the best overviews of how Google manages its servers. Thank you.
Hearing about this talk right now. Must remember to watch
Add a comment...