amalie @amalie@scholar.social

hey can academics stop normalizing working on the weekends please? thanks.

ftr this is not a rag on my boss it's just amusing to me that someone leaves it until the last day of the semester in a modeling class before realizing they don't get that.

boss explaining to some students the difference between a variable and a parameter ... on the last day of the semester.

(How do you explain that, tbh?)

Also.

I suspect before I showed up, new lab had lab meetings with seriously detailed powerpoints.

Since I showed up, we've been small enough that stopped.

With several new grad students having accepted offers here, I'm worried that will start again.

BAN POWERPOINTS (they are not effective ways to convey information in a collaborative framework).

literally hitting the point where I'm sorting my "papers to read" list in my dreams.

This is not ideal.

anyway, ML is a hot topic in biology right now. Some cool regression tree/random forest projects are happening.

BUT in most public health cases we have

1) small data
2) human decision makers who make decisions without clearly stated objectives

so uh. ML isn't necessarily gonna get us anything.

More recently, we simulated a metapopulation with a gradient of coverage and two primary regions - one chunk of locales where disease had been eliminated and another where it was on the route to elimination.

We tried to classify those regions using a neural network.

That also failed (although I think a friend's still working on it).

At one point we tried to use reinforcement learning to discover a heuristic for when you should switch from your "achieving" strategy to your "maintaining" strategy along a gradient of vaccination coverage, using some immune summary statistics as state space.

In the end we got out some facts about our model for disease/immunity and no useful heuristic.

It turns out machine learning is not great for discovering heuristics.

A big take away from my PhD thesis was "the things you should do to keep an infectious disease at bay when the disease is locally eliminated differ from the things you should do to eliminate it".

I'm an applied theoretical disease ecologist (yes, applied and theoretical - empircal is the opposite of theoretical, not applied).

I model the distributions of disease and immunity and use those models to make management recommendations.

I like thinking about economic/decision theoretic tools and their applications to public health decision making.

(maybe I should tag this #introductions )

Went to a good talk on neural networks. More specifically, on the benefits/necessity of allowing local inhibition of neurons. It kinda had everything - humor, good pictures, a decent amount of detail but a decent amount of takeaways as well.

~buuuuuuuuut, it got me thinking about why every machine learning project I've been involved with is destined to fail~

(it's because we're asking the wrong questions)

Thinking about heuristics

Pet peeve: old emeritus white guys monopolizing the time of visiting speakers

Apparently some terrible things are being said about sexual harassment at the national academy of science meeting.

If you are afraid to collaborate with a scientist of another gender because you're afraid you might harass them, you're the problem.

metoo has only had a negative effect on the progress of science if you're a perpetrator.

Lab drinks are a blessing and a curse🍹🍺🥃

The option to sign my reviews makes me really uncomfortable.

I should sign them as my ex boss.

I feel like I may have understated at all previous points exactly how strongly I feel about the explicit inclusion of uncertainty in models that inform management.

You can't just not and say you did.

Thanks.

Maybe I'll just close my office door and do pushups until I feel like I can work better

(productivity tip: procrastinate by doing work)

aaaaaaaaaaaaaaahhhh I said I would review this paper today but my brain is doing cartwheels halp