Machine Learning is Boring

Little interest.

What? Isn't your major CS+bio?

There's so much more to computational biology than import tf.keras. Machine learning (ML) in biomedical research is not usually a matter of "is your model innovative?" but rather instead, just "do you have data?" The actual work entails:

• having one or more GPUs
• "ooo, that didn't work, leT'S add AnoThER lAYEr"

ML research speaks more to the resources the research institution has, rather than the researcher's ingenuity.

P.S.: An example of cool and creative computational biology, in my opinion, is network (neuro-)science. Unfortunately I am not smart enough for that.

ML In My Research

I work in neuroimaging and on the ChRIS project.

An issue I notice is, researchers simply move on after publishing their paper, sometimes leaving their software deliverables behind to collect dust.

We do research to change lives, right? Not just to score citations?

I am interested in ML research at the meta level:

• How can ML research be conducted openly and reproducibly?
• How can ML software be applied in practice, in the real world?

So don't just pump and dump a Jupyter notebook and call it a day. Github should not be the last stop for your code — your goal is to have your code running in a hospital's HPC center, solving problems and saving lives. First, let's try writing some quality, open-source software, and then figure out how to deploy it so that people can use it. Finally, for a software project to be truly complete, it needs users.

Software is beautiful because its development cycle from development to deployment can be much faster compared to traditional research. As software developers, we ought to have high standards to live up to the potential of our technology.