15:31
This not really a statistical question, but it is a question about statistical communication. I've worked professionally as a data scientist for about 12 years, and something I've noticed is that there's often a pretty wide gap between what a business actually needs and how the scientists at the business think about their problem. I'm trying to figure out how to explain this problem, and its solution, in a way that will resonate with a C-suite audience.
In specific terms, I've worked in contexts where we have lots of positive examples (e.g. malware sample) and have even more unlabeled examples (e.g. a bunch of software that we know is a mix of benign and malware, but we don't know which is which).
The "textbook" approach is to say "We can't do much with only one class of samples, so let's just pay a bunch of experts to label the unlabeled data." Naturally, this is expensive.
A more innovative data science approach is to use weakly supervised/PU learning. So I was able to radically streamline and economize this workflow at my firm by using these methods.
I'm trying to figure out how to explain that this is a general phenomenon, not solely a success story about PU learning. In essense, explain that expertise means knowing a lot about many different tools and understanding which tool is right for what jobs. At the core, I think a lot of businesses are "stuck in a rut" with data science methods that are a poor fit for the real problem they need to solve.
2
In another setting, we encountered a problem where we would gain more information about a phenomenon over time, but we needed to make a binary decision. So if we wait too long, we lose value, but if we classify too early then we incur high costs. There are methods that accommodate this setting, and it blew the minds of the business leaders when I pointed it out.
But while these examples are true and specific, I'm not certain how to make the message -- methods are important & can solve problems you don't know about -- really resonate, and be understood as having general, far-reaching consequences to a C-suite/executive audience.
