OBSOLETE! Page is now in Diff-Segm

data and process

DP

What I find always very interesting is when I try to explain to students floating points numbers vs integers numbers, how they are constructed. In some way, the fact that with the same number of bits you can store much bigger ranges of numbers only depends on the fact that you have established a very tight interpretation of that information. The standard of how you read these bits AND the bits, together, makes more than just the bits. In a 32 bits world, you can store a number of states.

 

HHR

2^32 values

 

DP

Exactly. But the fact that then you have a specific reading of the same bits, that allows you to have much more dynamics, a bigger range and precision, that comes only because there is a very strong interpretation of how you have to read this. So the standard AND this bit region together makes more.

 

HHR

Yeah, of course, is a convention. Like in zip compression, you store a dictionary and then you can use the bits effectively. All sorts of compressions, in theory. It is entropy coding, so to speak. 

 

DP

Of course, but I mean, for the floating point number this is very clear. But I feel that this is almost always the case for any kind of data. The data is never just the data itself, it has always this part somewhere, that is sometimes not so explicit, which tells you how to read these data. It's really very very tightly related and I think that this is sometimes very implicit, this aspect. That you cannot have one without the other, that the interpretation is not part of the data sometimes.

 

HHR

Yeah, you have two bytes of unicode and you need to know that it's a smiley puking in the street, you know.

 

DP

But that's not written in it, that's something external.

POZ 191114

maybe we could also consider one compression algorithm  as a candidate for our installation?

DP

I like somehow the idea that the input is part of the program. To make in some way the input signal part of the program. I don't know how to do this, but I think it's very interesting.

 

HHR

Well, some sort of memory basically. That is kind of crucial for the continuation of the system somehow. 

 

DP

Yeah, that first of all. But also that it's not just data that an algorithm uses to compute something, but it is already a function somehow. It is a function that computes something, or something like that. I don't know how to do this, but ..

 

HHR

This would point to learning algorithms and neural networks a bit. Because it's kind of their way to say that the data is slowly kind of engraved into the system.

 

DP

Well, it depends how you see it how course. This would be one way, but one other possibility is to say that still there you have quite some steps that extract informations from the signal and then applied. So the signal itself is not an instruction. 

 

HHR

Yeah, but the question is what is the signal itself. 

 

DP

Yes, that is the question, but we just take the samples.

 

HHR

But that also equates to say: this is the signal. The samples are the signal. Not necessarily, you could also say the frequency in the signal is the signal. 

 

DP

Yeah, but it's another thing. I think it's still different than the neural network or this learning algorithm. Because you still have this step in which you condensate something from this data, you still consider the input as data somehow. I don't really know how to do it, but I'm just thinking. 

 

HHR

I mean, learning basically means reacting more slow to change than the rate of the change. Condensation is this idea that there comes something and it leaves a trace. You don't replace everything that has been there by the new data, but it aggregates with what's the history of before. This is what learning is. 

 

DP

But I'm arguing that maybe this kind of learning is not the only method, there is probably another way. This is of course one possibility, to integrate the input - call it data, or signal or sound - into a running algorithm, that's one possibility. I'm asking if there's a more direct way to integrate the signal or that incoming data as part of the program, without this learning, this step that extracts something. I'm just asking myself if this is even possible. Maybe not, but it's an interesting question to pose.