Can Decision Theory Be Used to Understand Complex Adaptive Systems?

[hepiaaro]

Hi, I am a mathematics student. I have recently become interested in Systems Theory and Complex Adaptive Systems Theory. In reading through a textbook on the subject I came across a small section on decision theory.

Is anyone familiar with Decision Theory? Or more importantly, familiar with its relevance to complex adaptive systems? What can you use the language to reason about? How "versatile" is it?

Thanks!

 

[Physics-UG]

I'm not super knowledgeable in this topic, so someone who is can hopefully chime in with more detail, but, I'll add what I can. A friend of mine is doing research in computer science and neural networks. What they are using is algorithms to solve puzzles, and so what they do is code whatever it is (I'm not exactly sure, unfortunately), and run the system. It will try and find a way to solve the system it's given, while simultaneously remembering which ways worked best and it remembers it for future trials. The more trials they run, the more that it finds the better, and faster, route to take.

Like I said, I'm not super knowledgeable, but this more or less gives you an idea of some applications.

 

[hepiaaro]

What they are using is algorithms to solve puzzles, and so what they do is code whatever it is (I'm not exactly sure, unfortunately), and run the system. It will try and find a way to solve the system it's given, while simultaneously remembering which ways worked best and it remembers it for future trials.

Did your friend tell you whether this "memory" can be understood by humans? And whether the algorithm explores the reasons a particular solution did not work? What I'm trying to get at is whether we can learn anything about how to analyze systems ourselves? You can imagine a kind of feedback loop between the algorithm and optimization by human intervention?

 

[Physics-UG]

Did your friend tell you whether this "memory" can be understood by humans? And whether the algorithm explores the reasons a particular solution did not work? What I'm trying to get at is whether we can learn anything about how to analyze systems ourselves? You can imagine a kind of feedback loop between the algorithm and optimization by human intervention?

He hasn't. However, although his responding can be hit or miss I did text him and ask and whenever I do get a reply I'll let you know. I would imagine that the system must use some form of analysis in order to understand why path A didn't work, path B did, and find out what made path A more susceptible to failure, but I may be wrong.

 

[hepiaaro]

He hasn't. However, although his responding can be hit or miss I did text him and ask and whenever I do get a reply I'll let you know. I would imagine that the system must use some form of analysis in order to understand why path A didn't work, path B did, and find out what made path A more susceptible to failure, but I may be wrong.

Thank you stranger!

 

[Physics-UG]

Thank you stranger!

So this is what he said. I was a little off since it was a while since him and I had talked about it. It doesn't really memorize paths, it learns parameter weights. So, basically what it does is learn a function of the input, usually by mapping to a probability of some sort; so it can't really be understood by humans.

It's kind of what deep dream does, it would take a trained NN, give it some input, and enhance the features that it detects in the image so you see what it "sees".

 

[hepiaaro]

So this is what he said. I was a little off since it was a while since him and I had talked about it. It doesn't really memorize paths, it learns parameter weights. So, basically what it does is learn a function of the input, usually by mapping to a probability of some sort; so it can't really be understood by humans.

It's kind of what deep dream does, it would take a trained NN, give it some input, and enhance the features that it detects in the image so you see what it "sees".

Ah yes, I thought as much. I wander if anyone has tried to create an algorithm that can analyze its own functioning? Thanks again!