Can Carbon Relay deliver AI efficiency like what Google's Data Center group uses?

The media covers Foxconn’s back of Carbon Relay. Here is tech crunch’s post.

Taiwanese technology giant Foxconn International is backing Carbon Relay, a Boston-based startup emerging from stealth today, that’s harnessing the algorithms used by companies like Facebook and Google for artificial intelligence to curb greenhouse gas emissions in the technology industry’s own backyard — the datacenter.

According to LinkedIn the founder Matt Provo has been with the company since Aug 2015.

Carbon Relay has on its web site a graph that shows how its model matches the actual PUE

Screen Shot 2019-01-30 at 14.01.40.png

Which looks a lot like Google’s predictive accuracy which is in this paper.

Screen Shot 2019-01-30 at 14.02.33.png

I don’t know Matt Provo or anyone at Carbon Fiber. You can see the team on this page which has their LinkedIn profiles. From taking a quick look I don’t see any mechanical engineers or data center operations people.

Google’s AI/ML energy efficiency project was headed up by Jim Gao who i do know. Jim is a mechanical engineer from UC Berkeley. Go Bears! I also have my engineering degree from Cal, but long before Jim went there. Jim had years of working in Google’s data center group and started down the path of machine learning and he had one of the biggest sources of training data, Google’s data centers. Which may explain why Jim’s predictive models look more accurate than Carbon Relay.

Jim published his latest findings as part of work in Alphabet’s Deepmind where is now a Team Lead.

Screen Shot 2019-01-30 at 14.09.40.png

So can Carbon Relay’s 14 people deliver a solution as good as Deepmind’s Jim Gao? Jim has gone through the painstaking efforts to get clean accurate data from systems. There are so many small details. I love the one example where Jim ran the model to be the most energy efficient, so it turned off all the systems bringing energy use to 0. And Jim has overcome the resistant to change from a well trained data center operations staff to trust a computer model.

Looking at the number of technical team on the Carbon Relay project I am reminded how the first models Jim ran could be performed on one PC. Time will tell if Carbon Relay can deliver on data center efficiency, but even if they have a technical solution getting clean data from all the BMS environments and executing a model that is used is so hard.

The paper that Jim published has Amanda Gasparik on the paper. Got curious looked her up on LinkedIn as she is senior data center mechanical engineer. Been at Google 5 years. 8 years as Nuclear Electronics Technician for US Navy. Masters System engineering and Bachelor’s mechanical engineer.

Add another DeepMind PhD Research Engineer and you have three people who have a broad range of skills that impress me much more than Carbon Relay.

Google's data center AI puts safety first just like airplanes fly-by-wire while saving 30% of cooling energy

Google has a post on its latest application of AI from the Deepmind group to its data center cooling systems. The tech media nicely covered the post. Here is a full coverage link.

Google choose to emphasize the system was designed for safety as indicated in the title of their post.

Safety-first AI for autonomous data center cooling and industrial control

The following graphic illustrates the safety principles used like a fly-by-wire system. 

While traditional mechanical or hydraulic control systems usually fail gradually, the loss of all flight control computers immediately renders the aircraft uncontrollable. For this reason, most fly-by-wire systems incorporate either redundant computers (triplex, quadruplex etc.), some kind of mechanical or hydraulic backup or a combination of both. A “mixed” control system with mechanical backup feedbacks any rudder elevation directly to the pilot and therefore makes closed loop (feedback) systems senseless.[1]

It has been a pleasure watching Jim Gao make progress with with AI in data center cooling and you can bet there will be much more coming.

Jeff Dean publishes Part 1 of Reflection of 2017 Google Brain results - ML, ML, and more ML

Jeff Dean posted part 1 of his reflection of Google Brain’s 2017 achievements.

If you don’t know what Google Brain is you can check out this wiki post.

When you read the post you can see the work is lots and lots of ML. Using the below infrastructure.  Well they are probably using some really amazing stuff that Google won’t share for a long time. This is the elite Google Brain team they can get anything they want.



As an example of some of their work Jeff references this text to speech work. 

Check out this graphc that shows where TensorFlow is used. 



Google shares its observations on Best Practices for AR

AR is a hot topic and Google has a post where they share their observations on best practices.

“From our own explorations, we’ve learned a few things about design patterns that may be useful for creators as they consider mobile AR platforms. For this post, we revisited our learnings from designing for head-mounted displays, mobile virtual reality experiences, and depth-sensing augmented reality applications. First-party apps such as Google Earth VR and Tilt Brush allow users to explore and create with two positionally-tracked controllers. Daydream helped us understand the opportunities and constraints for designing immersive experiences for mobile. Mobile AR introduces a new set of interaction challenges. Our explorations show how we’ve attempted to adapt emerging patterns to address different physical environments and the need to hold the phone throughout an entire application session.”

It’s a good summary of issues that are kind of obvious when you start down the path of building solutions.