Canada STEM Award for Americans

This post is mainly intended for undergrads who are thinking about going to grad school.

I did my undergrad degree in Canada, and was subsequently very fortunate to receive one of the US Fulbright science and tech PhD fellowships to attend UC Berkeley. These are great fellowships, and if you are a non-american, and interested in coming to the US for PhD studies, I strongly encourage you to look into that program.

Recently, I became aware of a new program which is basically the inverse of the one I am currently a part of. This is a program run by Fulbright Canada to bring top US students to Canada's best universities to pursue PhD studies. The benefits are many, so I would encourage any potential PhD students to investigate more fully.

Even if you've never considered studying in Canada, I urge you to think about it. From my experiences in materials science, nuclear physics, astrophysics, and particle physics, the research facilities in Canada are top-notch, and Canada has some of the world's most liveable cities. Fortunately, our best universities also tend to be in our nicest cities!

Best of luck!

Uncertainty and decision making

So.... here is a post about my first biology publication: "how should prey animals respond to uncertain threats?".

I'll summarize very briefly some ideas about gambling, and the Kelly criterion, and then discuss what that has to do with prey animals.

Let's start our discussion by imagining that you and I are going to gamble on coin flips. We will flip a coin, and bet at even odds (so if it's heads, I pay you the amount of the bet, and it it's tails, you pay me that same amount). But, the coin is biased in your favor, so that it comes up heads 55% of the time, and tails 45% of the time. This means that you have a 10% edge on the bet: on average, you expect to get back 110% of the bet, for each bet you make.

If we only do one coin flip, and you want to maximize your expected profit, you would bet everything you have. You might lose, but your expected profit is positive.

Instead, let's consider the case where we keep flipping the coin over and over again, and you try to maximize your long-term profit. In that case, it would be silly to bet all of your money on the first coin flip because, if you lost that one, you would lose the ability to make money on future bets (because you would be broke and not able to keep betting).  Back in the 1950's, J Kelly demonstrated that the best possible strategy in this case is to bet 10% of your money on each coin flip. As your bankroll grows, you bet more. This strategy provides the best balance between betting big (since you expect to make money on each bet, and bigger bets mean more profit), and avoiding going bankrupt (which gets rid of any chance of future profit).

In my paper, I discuss a semi-related problem, which is as follows.

Imagine that you are a deer, in a forest. You spot movement out of the corner of your eye, but you don't know for sure what is causing it. If it's a wolf (or whatever predator), you should run away to avoid being killed, but if it's not a predator (say, just some leaves blowing in the wind), then running away would waste energy, and cost you whatever mating or foraging opportunities were presently available to you.

Now we want to figure out what the deer (you) should do in that situation.

Interestingly, much like in our gambling example, the "correct" decision (the one that would be favored by evolution; the one that allows the deer to have the most offspring in its lifetime) is very heavily influenced by the uncertainty of the outcome. So, even if it might be immediately (on average) advantageous to "risk it", and not flee, when you are uncertain about whether or not a predator is present, the fact that you lose all future mating chances if you are wrong makes the "correct" decision strategy more cautious.

This issue - the influence of uncertainty on prey escape decisions - was not previously understood in the behavioral ecology models, but I am hopeful that future work in this field will be influenced by my result.

criticality

After a long hiatus, I am back to blogging.

Yesterday's physics colloquium was given by Bill Bialek, physicist and theoretical biologist at Princeton (and the PhD thesis advisor of my PhD thesis advisor). His talk was based on a recent paper titled "Are biological systems poised at criticality?". In the context of neuroscience, Bialek's basic observation is that, yes, neural systems appear to have this special "critical" property.

In particular, the observed correlations between the activities of two neurons are such that, if they were any stronger, the brain would be epileptic (recall that, in epileptics, the activities of neurons are amplified such that you get huge cascades of activity, resulting in seizures), but if those correlations were any weaker, the brain would effectively be "dead" (there would be no significant collective behavior).

Now, Bialek's work also discusses criticality in protein sequences, and collective animal behavior, but my interest is mainly in the brain.

Now, from a purely functional standpoint, this "criticality" seems to be sensible, and I could imagine it arising as a product of evolution; animals with more strongly correlated neurons would be epileptic, and they would die off, but so would those with less strongly correlated neurons, as they might be unable to effectively process information.

However, the brain is not static over the lifetime of the animal. We learn and adapt, and as we do, the correlations between neurons in our brains change.

How, then, is this criticality maintained? In other words, is there some kind of homeostatic mechanism that adjusts the correlations (or synaptic connection strengths that, presumably, alter these correlations), to keep them at this critical point?

These are, admittedly, ill-formed ideas at present, but I may very well get back to them when I have a chance.

In other news, my first biology paper was just accepted for publication in "frontiers in computational neuroscience." I will post a link to the paper when it is all copy edited and ready for public consumption.