The links of August

Like every blogger, I probably should start collecting links to interesting stuff. Posting them on FB is nice, but FB has a rather short half-life of content.

Science and semi-science

Financial Crisis of 2015: written in 2011.

Predictability of Android lock patterns: Not as secure as you’d imagine.

Sex and cognition: It’s complicated. Very, very complicated. Seems like jumping to any conclusion whatsoever would be wrong.

Art

Sabaton – Poltava

“I love you” – “Shiny”: Princess Leia x Mal Reynolds

Random things

Tesla automated car charger: another toy for rich kids. (All other comments by me would reference a certain kind of animation.)

Antifragility: Nassim Taleb furiously promoting buzzwords.

Military spending is like Christmas: An interesting detail to the Silicon Valley story.

Games, rules, cooperation

Disclaimer: This is to be read as a personal position, not a hardcore philosophic work (For hardcore philosophic works, read Kant, he seems to say the same, but in a different way). Thus, the text may contain simplifications, logical shortcuts and things derived from personal or second-hand experiences and may not be generalizable to everyone.

Let’s talk about rules. Even in the age of Enlightenment, it is customary to consider rules as something holy and unalterable, like the Ten Commandments. Examples include the infamous “dating rules”, laws, and the (un)written social code. However, it is important to remember that rules have a purpose: they constrain everyone’s actions and thus impose a bound on entropy. Rules are a good thing, because they enable you to limit the diverse possibilities and allow you to concentrate more on the “allowed” alternatives; rules allow to expect behavior.

Continue reading “Games, rules, cooperation”

Living with Markov chains

It has been nearly two months since I started in grad school, and I have not told you what I am working on. (Well, some of you know, but I still have not announced it.) The project is called “Bounded-Parameter Markov Decision Processes” and it deals with decisions under certain uncertainty conditions. The broad context is optimization of technical systems under uncertainty, and this is one approach to the general problem.

So, what is it all about? It begins with Markov chains, which are a fairly old formalism that describes a system that has some states, like “lights on” and “lights off”. The formalism contains probabilities of the event that the system changes its state after one unit of time. With some linear algebra and probability theory, it is possible to compute the behavior of such system. Then, we go on and extend this formalism: now, in each state, we can make a decision and execute an action from some given action set; in our case these actions might be “flip switch” and “do nothing” (as you can see even in this simple model, inaction is also a decision and implies consequences). Each action has some costs attached to it (costs may be also negative, thus, being rewards) and alters the state in a different way: in our example, flipping the switch would mean a state transition and doing nothing would mean staying in the previous state. This leads to the obvious question of finding a strategy that minimizes costs. And for this case, too, one can employ Mighty Linear Algebra and find several algorithms that deliver optimal strategies.

Now, my problem is a little more complicated. In my case, the transition probabilities are not certain; one only knows lower and upper bounds, which in the most general case can mean arbitrary transition probabilities. In this case, it is far less trivial what the optimal strategy can be.

But the thing that bothers me most is that, actually, I’d like to get a joint Math/CS degree afterwards, since what I am doing is basically lots of (applied) math 🙂