Rail, Aquariums, and Data

Author: Josh Patterson

Date: May 25th, 2018

This is the 5th article in a 5-part series supporting the Smart Communications and Analysis Lab (SCAL) and the 2018 Chattanooga Deep Learning Conference.

Making the Case for Regional Graduate Research Investment

We'll start out by re-stating our opening thesis:

"you don't have to found Google in your city to get key companies to locate there, but you DO have to invest in your local graduate programs in a significant way."

We showed how research output in a region correlates to economic output, along with the effect of attracting top-tier labs (Google, Facebook, Amazon, etc). These areas, when the effects have enough years to compound over time, tend to create hub-economy-class companues (Google, Facebook, Uber, etc) that participate in value creation and disruption at a rate much higher than normal. We've also seen how top-tier tech companies tend to either have a connection (founder graduated there) to certain clustered universities in these areas or they were founded directly as a result of research produced there. We then listed the key factors from the Nature article:

  • Freedom to work on their own ideas
  • Tools and infrastructure to do so[footnote: pub-private]
  • Attractive lifestyle[footnote: lifestyle]
While we noted the success of cities like Silicon Valley and Boston, MA in the context of this model, we go on to point out how Austin, TX is a better exemplar for the City of Chattanooga to emulate in its success. We've noted that while Austin tends to not found some of the key "hub" companies, Austin attracts their secondary labs which employ thousands of people:
  • Amazon - 900 Austin employees, office opened in 2015
  • Facebook - 700 Austin employees, opened in 2010
  • Apple
  • Oracle
  • Blizzard
  • eBay

Still we see a gap between Chattanooga has accomplished and what similar, yet economically more successful, cities such as Austin, TX, have accomplished in terms of attracting high-growth employers in the software space. If we had to point out the one major difference it would be the lack of compsci research output (typically correlated with lack of funding + support) for UTC's computer science program. It may not be simple to get foundational tech companies that act as economic hubs to move to Chattanooga, but becoming a place that attracts their secondary offices ("spokes of the hub") connects to Chattanooga's history as a connector in the region (river, rail, interstate). If we put the graduate school investment in the light of the rail investment or the sacrifices made to get the interstate through Chattanooga, it fits in line with those historical narratives. Hopefully this time it won't cost us any more elevation off of Cameron Hill, though.

The easiest counter points to our argument here would be:

  1. We already have the Sim Center and ORNL (Knoxville area) here; Why hasn't that produced more commercial output from their research?
  2. We already have incubators and startup competitions locally; Why isn't that enough?
For #1, part of the answer is hard to define here, but I'll point back to the Nature article that mentions how regions such as New Mexico have a lot of research locally (Los Alamos and Sandia national labs) but do not always translate it into commercial success. Simply put, some types of scientific research do not translate well to the industrial or consumer space easily, so even with ancilliary support structures such as incubators or funding around them, they just don't produce more output beyond the papers they produce.

For #2, while venture firms, incubators, and competitions are all helpful, they tend to not produce the same class of original technical IP as years of pure computer science research does. It's hard to replicate years of relational database research in a 6 week startup competition (for example), and it correlates that you see the industry-defining database companies coming from areas (Berkeley, Boston) that have made significant investment in their graduate research producing papers in that sector.

Another factor at play is that hard questions such as the ones we face in data science about "what can I do with this sensor data?" typically have the initial answer of "we don't know".

Getting Comfortable with "We Dont Know"

Why do graduate programs make the best places to develop answers for hard problems? Because its where the "We dont know" questions end up getting solved. Start-ups are better at commercializing compelling ideas when they have a basic initial answer, but they are beholden to finding "product market fit" (and from Seed to Series A funding) as soon as possible, not exploring the problem space.

Corporations and private entities tend to have lots of data, but many times lack the appetite for discovering compelling use cases with it because many times the answer to "what good is the data?" is "we dont know." --- these answers tend to make managers find reasons to change the subject quickly and find other things to do with their engineering time, so many times companies simply report that they still aren't sure how to proceed with data science initiatives.

This article was meant to support the nacent UTC deep learning phd program work in the Smart Communications and Analysis Lab (SCAL), (which is showcased at the 2018 Chattanooga DL Conference if you have not yet signed up). This research group is focused on machine learning and data applications in:

  • Healthcare
  • Connected car / city
  • Smartgrid
Which have underpinnings in computer science areas such as:
  • advanced analytics
  • machine learning
  • deep learning
  • storage and database technology

One of the more compelling stories coming out of this group is how the city is letting them use the resurgent MLK Street area as a next-generation test-bed to further develop their smart city initiatives. For a city that wants to create a true "innovation district", its an interesting story to see one of the streets that 20 years ago most had written-off become a next-generation smart city test bed. Multiple universities (including GaTech and Vanderbilt) have committed to doing research on the data being collecte by the sensor arrays being installed on the street.

The projects we'll see at this conference will display next-generation answers to hard problems in their respective spaces. It's research like these projects that will eventually bring the major labs to town (for talent, as in other cities such as Austin). In the context of Chattanooga's lineage of public+private success there are two near-term things that we suggest happen to drive graduate school investment for economic growth:

  1. get regional private organizations to donate data and collaborate on research with students to create compelling publishable results to drive graduate school research output
  2. raise funds to support more graduate students and pay more new researchers for the comp science program
These types of activities will take time to show results, but they have shown to consistently deliver long-term benefits to a region.

What is Patterson Consulting Doing to Help?

Obviously Patterson Consulting cannot move the needle across all the facets discussed above by itself. To support the Smart Communications and Analysis Lab (SCAL) and their research efforts, we've set the following goals for the 2018 and 2019 academic years:

  1. Be on the committee for the (hopefully annual) Chattanooga Deep Learning Conference
  2. Help mentor deep learning and distributed systems graduate students on research projects for alignment with industry
  3. Purposely hire and recruit from the UTC graduate program
  4. Provide guest lectures on deep learning
  5. Help build and deliver a graduate class on advanced distributed systems (Hadoop, Kafka, Information Retrieval)

To close, if we step back and take a look at where Chattanooga has been (rail), how it changed its narrative (the Tennessee Aquarium, parternships, planning), and then look at where the hub economy is going (data), we can summarize and title our whole thesis with the moniker:

"Rail, Aquariums, and Data"