Database Reference
In-Depth Information
Gutierrez:
What do you look for in other people's work?
Shellman:
Presentation. What can I say? I'm shallow. I don't just mean visual
presentation (though it's important), but the ability to convey results both
technically and non-technically. The work needs to communicate the point
clearly and coherently. I look for whether they did something fancy but
without rigor. Sloppy complexity is rampant because so many methods,
particularly in machine learning, have been commoditized with external librar-
ies. Presentation is the ability to craft a story, from the reason that you did
the research, why I should be interested, what you did, and justifications for
your methods.
Unfortunately, I think presentation skills are undervalued, but is actually one
of the most important factors contributing to personal success and creating
successful projects.
Gutierrez:
What do you think the future of data science looks like?
Shellman:
If you look at trends in data science-y start-ups, we appear to be
moving in the direction of push-button data science. I hear a lot of marketing
about “freeing data scientists from having to program” or “freeing data scientists
from the technical overhead so that they can get back to the data.” I think
these products are a response to a lack of supply of people with data science
skills. In small companies, who are lucky enough to have one data scientist, these
tools promise to make them more efficient. It's not an easy job though, so I'm
pretty skeptical of these products and their longevity.
What I think we'll see in the future is an evolution of what a data scien-
tist is. Right now it's typical that a data scientist is an ex-academic with a
masters or PhD, but there just aren't enough of those people out there
to meet the demand, so I expect we'll see a lot of retraining of software
engineers and migration into the data science role. You can see some of
that already happening with Coursera and Udacity offering data science
courses and certifications.
Gutierrez:
What is something a small number of people know about that
you think everyone will know about in five years?
Shellman:
Ad tech doesn't work, but I think we all know that now … just
kidding. Tying it back to the work of the Covert lab I mentioned earlier, I hope
that in the next five years we'll see a greater appreciation for predictive mod-
eling in cell science. I think the ability to run experiments computationally and
make predictions at the whole-cell system level is immeasurably valuable. The
cycle of learning by modeling, testing model predictions with experiments, and
updating the model with the results is so obvious to me, but academia as a
whole is extremely risk averse and hasn't effectively made use of these models
as a tool for experimental design.
