Transcript for Vladimir Vapnik: Predicates, Invariants, and the Essence of Intelligence

It's completely different story. engineering, the imitation of human activity. You have to make a device which has a human decay. You have all the functions of human. It does not matter how you do it. But understand what is intelligence about is quite different problem. So I think I believe that it's somehow related to predicate, we talk yesterday about it. Because look at the Vladimir Propz idea. He just wants 31 he predicates. call it units, which can explain humor, behavior, at least in Russian tales here, local Russian tales, and derived from that, and then people realize that it may avoid in Russian tales. It is in TV, in movie series and so on and so on.

I think that it is different between you have different goals. your goal is to create something, something useful. And that is great. And you can see how much things was done and I believe that it will be done even more. It's self-driving cars and also business. It is great. And it was inspired by Turing vision. But understanding is very difficult. It's more or less philosophical category. what means and distance evolved. I believe in Him which starts from platter that's the exists vault of ideas. I believe that intelligence it is vault of ideas, but it is vault of pure ideas. And when you combines these reality things. It creates, as in my case, in variance, which is very specific. And that, I believe, the combination of ideas in way to construct and convert into intelligence. But first of all, predicate. If you know predicate and hopefully, Not too much predicate exist. For example, Sotyvan predicate for human behavior.

I think that's behalf, a lot of form of behavior, but I think the predicates is much less, because even in this examples, which I gave you yesterday, you saw that predicate can be one predicate can construct many different invariants depending on your data. They are applying to different data and they give different invariants. But pure ideas may be not so much. I don't know about that. But my guess I hope that's why challenge about digital recognition. How much you need to

Absolutely. You should start this simple story on the voice you're not able to do it.

I think that to beat records using 100 to 100 times, less examples, you need intelligence.

I don't know. I have a feeling formally as they exist. But I believe that predicate for 2D images. One of them is symmetry.

In my definitions, the simplest predicate is function function and you can use this function to my inner product, that is predicate.

input is x something which is input in reality. Say, if you consider digit recognition, it pixel space. Yes, input. But it is function which in pixel space. But it can be any function for no pixel space. And you choose the, and I believe that there are several functions, which is important to understanding of images. One of the most symmetry, it's not so simple construction, as I describe this literary, the results of stuff. But another, I believe I don't know how many. is how world structureized is picture.

Yeah.

It is formal definition say something happened heavy on the left corner, not so heavy, in the middle and so on. You describe in general concept of what you see.

Yeah. But I don't know how to formalize this.

You know, that is example. When critics in music trying to describe music, they use predicate. and not too many predicate, but in different combinations. But they have some special words for describing music and the same should be for images. But maybe Zara, critics who understand essence of what this image is about.

No, no, it's not on this level. It should not describe two, three, four. It describes some construction which allow you to create invariance.

It is property of your image. I can say looking at my image, it is more or less symmetric and I can give you value of symmetry. I say level of symmetry using this function which I gave yesterday. And you can describe that You image have these characteristics exactly in the way how musical critics describe music. So, but this is invariant applied. to specific data to specific music to something. I strongly believe in this plateau idea that there exists a world of predicate and wealth of reality, predicate and reality is somehow connected and you have to.

Yeah, you have projection. Oh, well, it was a deal. Yeah, right. In reality, you can realize this projection using the design variance because it is projection for on specific examples, which creates specific features of specific objects.

Exactly. Like in this music story, intelligent musical critics knows this all this world and they're really feeling about what they're like.

No, I don't want that too much.

to be concentrated on very simple story on digital recognition.

I don't know, because it's so complicated. It is involved a lot of stuff which I never considered. But I know about digital recognition. And I know that for digital recognition, to get a record from small number of observations, you need to predicate, but not special predicate for this problem, but universal predicate, which understand the world of images. But on the first step, they understand, say, the world of hundred and digits, or characters, or something simple. No, that's what I think one of the predicated related to symmetry.

Many kinds of symmetries. There is a symmetry, anti-symmetry, say letter S. So it has vertical anti-symmetry and it could be diagonal symmetry vertical symmetry. So when you cut vertically the letter S, then the upper part and low part in different directions.

Right, but there is a degree of symmetry. If you play all the slid-related stuff to do tangential distance, whatever I describe, you can have a degree of symmetry. And that is the describing reason of image. It is the same as you will describe this image. Same about digitists, it has anti-semitry, digital, symmetric, more or less look for symmetry.

And you can do this idea of symmetry, make very general, like the degree of symmetry. The degree of symmetry can be zero, no symmetry at all, or the degree of symmetry, say, more or less symmetrical. But you have one of these descriptions, and symmetry can be different. As I told, horizontal, vertical, diagonal, and anti-semitrys, also concept of symmetry.

I would like to concentrate on all I would like to know predicate for digit recognition.

It is not necessarily for digit recognition. It helps to create invariant which will be used when you have examples for digit recognition. You have regular problem of digit recognition. You have examples of the second class. Plus, you know that there exists a concept of symmetry. And you apply when you're looking for a decision rule, you'll apply a concept of symmetry of this level of symmetry which you estimate from me. So let's talk. Everything is continuity convergence.

You converging, you would like to have a function. The function which, say, indicator function which indicate your digit 5, for example.

Let's talk only about classification.

I would like to have these functions. Then I have some examples. I can consider property of this examples, say, symmetry. And I can measure level of symmetry for every digit. And then I can take average and I from my training data and I will consider only functions of conditional probability which I am looking for my decision rule which applying to digits will give me the same average as I absorb on training date. So actually this is different level of description of what you want. You want not just your show, not one digit. You show this predicate show general property of all digits which you have in mind. If you have in mind digits 3, it gives you property of digits 3 and you select as admissible set of function only function which keeps this property. You will not consider other functions. So you immediately looking for smaller subset of function.

It's a large number three large, but if you have one predicate, but according to the, there is a strong and weak convergence. Strong convergence is convergent and function. you're looking for the function on one function and you're looking for another function and square difference from them should be small. If you take difference in any points, make a square, make an integral and it should be small. That is convergence in function. Suppose you have some function, any function. So I would say, I say, some function converges to this function. If integral from square difference between them is small.

Yes. It is convergence in functions. But you have different. convergence in functionals. You take any function, you take some function phi, and take inner product, this function, this f function, f0 function, which you want to find. And that gives you some value. So you say that set of functions, converge in inner product to this function, if this value of inner product converge to value F0. That is for 1, phi, but we converge as liquid, it converge for any function of Hilbert's place. If it converge for any function of Hilbert's place, then you would say that this is a weak convergence. You can think that when you take integral, that is the integral property of function. For example, if you will take sine or cosine, it is coefficient of say Fourier expansion. So if it converges for all coefficients of free expansion, so under some condition it converges to function, you're looking for. But if convergence means any property, convergence not point wise, but integral property of function. So the convergence means integral property of functions. When I talking about predicate, I would like to formulate which integral properties I would like to have for convergence. And if I will take one predicate function which I measure property. If I will use one predicate and say, I will consider only function. which give me the same value as with this predicate. I selecting set of functions from functions which is admissible in the sense that function which I looking for in this set of functions. because I checking in training data, it gives the same.

Yeah, but, but property, you can know independent on training data. And this guy prop.

So the risk formal property. 31 property. You very tell Russian fairytale. But Russian fairytale is not so interesting. more interesting than people applied to movies, to theater, to different things and the same works, the universal.

It is the same story.

Yeah. But my point is why I hope that it possible to beat record using not 60,000, but say 100 times less, because instead you will give predicates. And you will select your decision not from why it's set of functions, but from set of functions which keeps us predicates. What's predicate is not related just to digit recognition. Right.

Because According to theory about weak convergence, any function from Hilbert space can be predicate. So you have infinite number of predicate and before you don't know which predicate is good and which. But whatever prop show and what people call it breaks through, that there is not too many predicate which cover most of the situation happened in the world.

I think that I would say only small part of predicate very useful, useful all of them.

Even in my talk, describe how to do predicates. How to find new predicates. I'm not sure that it is very good.

Well, in my talk, I gave example for diabetes. One, when we achieve some percent, so then we're looking for an area where some sort of predicate, which I formulate, does not, tips invariant. So if it doesn't keep, I retrain my data. I select only function which keeps this invariant. And when I did it, I improved my performance. I can look for this predicate. I know technically how to do that. And you can, of course, do it using machine, but I'm not sure that we will construct the smartest predicate.

Let's talk about deep learning. Sure. I know only Jan Slikun, convolutional network. And what else? I don't know any very simple convolution. There's not much else to do. I can do it like that. This one predicate.

It's single predicate. Yes, it is exactly, you know exactly, you take the derivative for translation and predicate this should be kept.

Note it, that is a risk, not too many predicate this. And that is big story because Jan did 25 years ago in nothing. So clear was added to deep network. And then I don't understand why we should talk about deep network instead of talking about piece-wise linear functions which keeps us predicate.

It's an engineering problem. I understand, but let's talk not on the motion level, but on the mathematical level. You have set of piecewise linear functions. It is all possible neural networks. It's just a piece of what's linear functions. This is many, many pieces.

Exactly, but very large. Very large. It's still simpler than, say, convolutional, then, reproduce and turn out Hilbert space, which have a Hilbert set of functions.

It's space with infinite number of coordinates, a function for expansion, something like that. So it's much richer. So when I talking about closed form solution, I talk about this set of function, not piece by linear set, which is particular case. It's a small part of the space you're talking about. A small set of functions.

But it is fine. It is fine. I don't want to discard the smaller beak if you take advantage. So you have some set of functions. So now, when you're trying to create architecture, you would like to create admissible set of functions, which all your tricks to use not all functions. But some subset of this set of functions. Say, when you're introducing convolutional net, It is way to make this subset, useful for you. But from my point of view, on the evolution, it is something you want to keep some invariants. It's a translation invariant. But now, if you understand this, and you cannot explain on the level of ideas what Neon Network does, He should agree that it is much better to have a set of functions. And they say, this set of functions should be admissible. It must keep the invariant and that invariant. You know that as soon as you incorporate new invariants, set of functions because small and small and small.

Yeah, but what I am hope that the resistant that predicate, like prop show, that what I want to find for digital recognition, if we start, it is completely new area, what is intelligence about, on the level starting from flat as a year, what is the world of ideas. So, and I believe that it is not too many. But you know, it is amusing that mathematician doing something in neural network, in general function, but people from literature, from art, they use this all the time. It is great how people describe music, we should learn from that. And something on this level, but so why Vladimir Propu was just theoretical, who studied theoretical literature, he found that.

Yeah, you know, just small or big. I know exactly. Then when I'm introducing some predicate, I decrease set of functions. But my goal to decrease set of function match. By the match is passed. By the match is possible. Good predict which does this. Then I should choose next predict which does this decrease set. As much as possible. So set of good predict. It is such that the decrease amount of admissible functions.

Yeah. But if you reduce very well the VC dimension of the function of admissible set of function is small and you need not too much training data to do well.

Right. How roughly speaking, how many functions in this set? So you're decreasing, decreasing, and it might easier for you to find function you're looking for. So the most important part to create good admissible set of functions. And it probably there are many ways, but the good predicates, such that they can do that. So that for this duck, you should know a little bit about duck because... What are the three fundamental laws of ducks? looks like a duck, seems like a duck, and quack, like a duck.

Not necessarily. Looks like the horse. It's also good. It's not. It generalizes from a duck. And talk like, like, it makes sound like horse. Yeah, something. And run like horse and move like horse. It is general. It is general predicate. That's the applied to duck. But for doc, you can say play chess like doc.

So you can, the story is formal story, with the magical story, is that you can use any function you won't let the predict it. But some of them are good, some of them are not because some of them reduce a lot of functions. So admissible set of some of them.

Yeah, I understand, it's like that. I understand, what do you find? What do you mean, I knew predicate? Like guy who understand music can say this is worth which he described when he listening to music. He understand music. He just not too many different, or you can do like prop. You can make collection. What he talking about music about this is about that. It's not too many different situations he described.

No, he just gave description of all suitations. He understands this vault of false tales. Yeah, not for not false stories. And this story is not in just four tales. The story is in the detective series as well.

At the end of the roads, this predicate is good for different situations. from movie, for movie, for theater.

I don't know. I'm not in this business.

It's always, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, It's not too many units that can describe, but they try to lead us to the units.

We need to set up predictions. It's not possible. But they exist, probably.

Let me ask you, do you understand? Did you dig it? No, I don't understand. No, no, no. When you can recognize this digital team, you just say, you understand. You understand. Caracas, you understand.

I'm not talking about, I'm talking about predicate. You understand that it involves symmetry, maybe structure, maybe something. I cannot form a way, I just was able to find symmetries.

No, I understand. Very frankly. You're using digits, you're using examples as well. So what it says is that if you will use all possible functions. From Hilderspace, all possible predicates. You don't need training data. You just will have admissible set of functions which contain one function.

Exactly. That is intelligent learning.

So, people who will solve this problem, they will answer. They will answer them.

Is it just neutralite function?

No, I'm not against neural network. I'm against admissible set of function which create neural network. You don't do it by invariance by predicate by by reason.

Are you on grad school listening? I give example. So find situation where predicate, which you're suggesting. Don't create invariant. It's like in physics, find situation where existing theory can not explain it.

Find contradiction. And then remove this contradiction. But in my case, what means contradiction? Do you find function? which, if you will use this function, you're not keeping converts.

Yeah. It is like in physics, find situation where you have contradiction for one of the property, for one of the predicate. Then include this predicate. making him variance. And so again, this problem now you don't have contradiction. But it is not the best way probably I don't know to looking for predicate. That's just one way. Okay. That, no, no, it is brute force way. The brute force way.

I don't think so. I think of just logic is not enough.

So everything is very simple. When you have infinite number of predicate, you can choose the function you want. You have invariance and you can use the function you want. But you have to have not too many invariance to solve the problem. So, and have from infinite number of functions to select finite number. and hopefully small phone number of functions, which is good enough to extract small set of admissible functions. So they will be admissible, it's for sure because every function just decrease set of function and leaving it admissible. But it will be small.

No, it is my decision. What means you know life?

No, no, you know something common sense, it is some rules.

I don't think it's so bright. I want to discuss, understand, dig it, dig it, dig it, dig it, dig it, dig it.

No, you know, it was doable because there is a challenge. Yeah, I see how to solve it. If I have a student, concentrating on this work, I will suggest something to solve.

I think that I know invariance which will solve this. You do. I think so. But it is not It is maybe, I want some universal invariance which are good, not only for digital condition, for image understanding.

Immediately, you will understand this. As soon as you will make a record, because things it is not for free, as soon as you will create several invariants which will help you to get the same performance that the best neural net did using hundred and maybe more than hundred times less examples you have to have something smart to do that and you're saying that that is invariant it is predicate because you should put somebody you have to do that but okay

And potentially much larger number. You know, maybe, but let's start from simple. Well, yeah, but you said that if you don't understand, I cannot think about things which I don't understand. This is I understand, but I'm sure that I don't understand everything there. That's a difference. I do as simple as possible, but not simpler. And that is exact case. With handwritten.

I start with very general problem. This plateau. So you understand and it comes from Plata to digital cognition.

I would say that I did not understand Plata. until recently, and until I consider the convergence and the predicate and the know, this is what Latter told.

No, it is metaphor. It is metaphor for sure.

But it is the way you should try to understand how attack a day is in the Lord. So from my point of view, it is very clear, but it is lying, all the time people looking for that. Say, Plato, Zen, Hegel, whatever reasonable exists, whatever existed as reasonable. I don't know what he had in mind reasonable.

No, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no

But what you should make a projection on reality. But understanding is it is abstract ideas. You have in your mind several abstract ideas, which you can apply to reality.

I think that it is, you know, my scheme is very simple. You have huge set of functions. If you will apply and you have not too many data. If you pick up function which describes this data, you will not worry well. I will randomly pick it up. Yeah, it will be overfitting. So you should decrease set of function from which you're picking up one. So you should go somehow to admissible set of functions. And this is what about weak conversions. But from another point of view, to make admissible set of function. You need just to do it. You just function which you will take in inner product which you will measure property of your function. And that is how it works.

But let's, let's think about examples. You have huge set of functions and you have several examples. If you just try to keep, put a function which satisfies this examples, you still view over fit, you need decrease, you need admissible set of functions.

But most of the function can be even bigger. I understand.

Creating admissible set of font municipal set of functions. No, you know, even from the classical history. from structure risk minimization. You should organize function in the way that they will be useful for you.

Yeah, but that is another story. I don't touch it because I believe that this small example is not too small. So you 60 per class. Low, flat, numbers, works. I don't need uniform law. The story is that in statistics, there are two laws. Love large numbers have uniform law of large numbers. So I want to be in situation where I use law of large numbers but not uniform law of large numbers.

I hope no, it still needs some evolutions, some bonds. But the deal is the following. If you trust that say, this average gives you something close to expectation. So you can talk about that about this predicate. And that is basis of human intelligence.

It is discovery of your understanding world, of your methodology, of understanding world. Because you have several functions which you will apply to reality. You have several functions, but they abstract. Then you will apply them to reality to your data and you will create in this way predicate, which is useful for your task. But predicate Not related specifically to your task, to this set of tasks. It is abstract functions, which being a line, applied to... When you task, you might be interested. It might be minute tasks, I don't know.

Yeah, there is like light in prop case. It was for fairy tales, but it's okay to interpret it.

I don't know him. Not torsional, I don't know. As ideas, these ideas.

For me, language and all the story of language is very complicated. I don't understand this. And I'm not, I thought about nobody. I'm not ready to work on that because it's so huge. It is not for me and I believe not for our century.

Not for 21st century. So, we should learn something a lot of stuff. from simple task like digital recognition.

You know, let me refer to each team. take the simplest problem as simple as possible, but not simpler. And this is a challenge, it's simple problem. But it's simple, by a day, but not simple, to get it. When you do this, you'll find some predictions, which helps it to do it.

Who's left? And I was a story, you don't have any universal instrument.

I don't know. No, I don't think in this direction. I just think in the direction that this is a problem which if you will, so it will create some abstract understanding of images. Maybe not all images. I would like to talk to guys who do encounter real images in Columbia University.

Yeah. What's their idea? Is there a predicate? What can be predicate? I still symmetry will play role in real life images in any real life images to images. Let's talk about to images. Because That's what we know, and narrow network was created for today images.

That's a, I'm like to discuss with them. Yeah. Because if we will be able to show that it is, what's working. And so it's because him, it's not so bad.

No, no, no letter. Langvich is much, much, much more complicated. It's a world, a lot of different stuff. It's not just understanding of very simple class of tasks. I would like to see list of tasks, where Langvich involved.

but even our records shows that we go in wrong direction because we need 60,000 digits.

No, no, in wrong direction because 60,000, this unacceptable.

If you can talk it's great, we have half percent of it.

If you will do from very small examples, you will find principles that we should do from what we're using now. And so basically it's more or less clear. That means that you will use the converges, not just strong converges.

Oh, yeah.

For example, I talk here today about symmetry. Yes. And it gave a very simple example. So the same will be.

For symmetry.

For degree of symmetry. Exactly. This is important. Not just symmetry. Existent doesn't exist. The degree of symmetry.

No, it's not for Henry, it's for him, he images. But I would like to apply to Henry.

Probably the uniform convergence, which we did, was Alexey Chelvin-Empis.

You have large, large numbers. So for any function, expectation of function, average of function, conversation. But if you have set of functions, for any function, it is true. But it should converge simultaneously for all set of functions. And for learning, you need uniform converges. Just convergence is not enough. because when you pick up one which gives minima, you can pick up one function which does not converge and it will give you the best answer for this function. So you need to uniform convergent to guarantee learning. So learning does not really want to reveal long, large numbers. But a deal of the convergence exists in statistics for a long time. But it is interesting that As I think about myself, how stupid I was 50 years, I didn't see V convergence. I worked on loan strong convergence. But now I think that most powerful is V convergence, because it makes admissible set of functions. And even in all products, when people try to understand recognition about dark law, looks like a duck and so on. They use weak conversions. People in language they understand this. But when we try to create artificial intelligence, we want to invent in a different way. You just consider strong conversions. Are we not?

You know, in classical mathematics, in Gilder's place, there are only two ways, two form of convergence, strong and weak. Now we can use balls. That means that we did everything. And it's so happened. then when we use Hilbert space, which is very rich space, space of continuous functions. Each has an integral and square. So we can apply weak and strong convergence for learning and have closed form solution. So for computationally simple, for me, it is sign that it is right way. because you don't need any of these techniques, yes, whatever you want. But now, the only word left is this concept of what is predicate. But it is not statistics.

No, it's so happen. Then when you're using Right, instrument, you have close to one solution.

You know, I know, I'm looking for bonds which I gave bonds for convergence. And when I look for bones, I think what is the most appropriate kernel of this bone would be. So we know the thin-saished, allow businesses to use radial-based function. But looking at the bone, I think that I start to understand that maybe we need to make corrections to radial-based function. to be closer to what better for this balance. So I'm again trying to understand what type of kernel have best approximation, no approximation, best

It still comes from, you're looking to mass and trying to understand what?

Then I try to understand what, what will be good for that?

Absolutely. When you're doing, say, self-driving cars, the great guy who will do this, it doesn't matter what theory behind that. who has a better feeling of to apply it. But by the way, it is the same story about predicate because you cannot create rule for a situation as much more than you have ruled for that. But maybe you can have more abstract rule than it will be less than the rule. It is the same story about a geosund and a geosupply to specific cases.

Well, you're suggesting that let us use this type of algorithm. When they start thinking, okay, so it's the full starting to understand what I want. Can they write down what I want? And then they try to formalize. And when they do that, I think you have to solve this problem. Till now, I did not see a situation where you need.

You know, most of the things that you don't understand in reasoning, human, it is for me to complicated. For me, the most difficult part is those questions, good questions, how it works, how people are asking questions. I don't know.

So, studies on the implementation. It's like predicate, like, say, admissible set of functions. It comes together, everything, because The first iteration of the theory was done 50 years ago, it will get this theory. So everything is there. If you have data, you can, and your set of function is not, has not be capacity. So obviously the mention, you can do that. You can make structural risk minimization control capacity. He was not able to make a possible set of functions. Now, when suddenly realized that we did not use another idea of convergence, which we can. Everything comes together.

Let's talk about philosophy. Philosophy means understanding of life. So understanding of life, so people like Plata, then understand and very high abstract level of life. So in whatever I do in just implementation of my understanding of life. But every new step, it is very difficult. For example, to find the idea that we need weak convergence was not simple for me.

You know, I've worked in my thinking about the same problem for 60 years somehow. Again, again, again, again. I try to be honest and that is very important, not to be very enthusiastic, but concentrate on whatever was not able to achieve, for example, and understand why. And now I understand that because I believe in mass, I believe that in ignorance, but now, when I see that there are only two ways of convergence. and we're using boss that means that we must as well as people doing. But now exactly in philosophy and what we know about predicate, what we understand life, can be described as a predicate. I thought about that and that is more or less obvious level of symmetry. But next, I have a feeling, it's something about structures. But I don't know how to formulate, how to measure, measure, structure and all this stuff. And guy who will solve this challenge problem, then when we will look in, how he did it, probably just only three not enough.

Oh, I'm sure it will be there. Oh, I'm sure it will be there. Level of symmetry will be there. And level of symmetry, anti-semitry, yoga, no vertical. And I, I even don't know how you can use in different direction the day of symmetry is very general. But it will be there. I think that people vary sensitive to the day of symmetry. But as a overall ideas like symmetry. As I would like to learn, but you cannot learn just thinking about that. You should do challenging problems and then analyzing why it was able to solve them and then you will see. Very simple things, it's not easy to find. Even with talking about this, every time. I also try to understand. These people describe in language, strong convergence, mechanism for learning. I didn't see, I don't know. But with convergence, this dark story, and story like that, when you will explain to you, you will use the convergence argument. It looks like it does like it is it. But when you try to formalize, it's just ignoring this. Why? Why 50 years? From start of machine learning.

I think that might be I don't know. Maybe this is your role, so we should blame for that, because empirical risk minimization don't stop. And if you read now text books, they just about bound about empirical risk minimizations, they don't look for another problem, like admissible set.

Music. Not below, what you've done, what you've done.

At first it was an idea that it was possible. And then when I was in the Bahia, I was like, well, it's a shame. By the way, I don't think that this is a predicate of the structure in the Bahia, but of course.

I don't think that Now, they were talking about a bag.

Chopin is another music story.

You know, it is very interesting to read what critics writing about Bach, which words they using, they trying to describe predicates and then shopping. It is very different, vocabulary, very different predicates. And I think that if you will make a collection of that. So maybe from this you can describe predicate for digital mission. Well, from Buckling Chopin. No, no, not from Buckling Chopin. From the creation of the music. They try to explain music what they use. As a use, they describe high level ideas of plot of ideas, what behind this music.

It is useless problems when you go from ideas to the representation. It is easy way, but when you're trying to go back, it is you'll post problems, but nevertheless, I believe that when you're looking from that, even from art, you will be able to find predicate for digital recognition.

No, yeah.

Not too much. Not too much. It is pity that you will not be able to do something cliché. I have a feeling to do that. For example, I didn't be very happy to work this nice tradition from music. to write this collection of description, what have they described in the music, how they use it and from art as well, then take what is in common and try to understand predicate, which is absolutely everything and try to stand for visual recognition.

You've got time. It's take years and years and years.

Yeah, but also some of many years. Yes, you know, the most time, it is not to make this collection. Don't understand. What is the common to think about that once again and again and again?

I even have some example. So I have friend who was specialist in Russian poetry. She is a professor of Russian poetry. He did not write points, but she know a lot of stuff, she make book several books in one of them, is collection of Russian poetry. She has images of Russian poetry, she collect all images of Russian poetry. And I ask her to do following You have nips, digit recognition. And we get 100 digits. I don't remember my 50 digits. And try from political point of view, describe every image you see using only words of images of Russian poetry. And she did it. And then, we tried to, I call it learning using privilege and formation. I call it privilege and formation. You have on two languages. One language is just image of digit and the null, the language poetic description of this image. And this is privilege and formation. And there is an algorithm when you're working using privileged information, you're doing well, better, much better.

And there is an NEC, she, unfortunately, direct the collection of digits in poetic descriptions of this digits.

But I think that the abstract idea is a plateau level of a need.

But as soon as we start this, this is a challenge problem.

You immediately connected to all this stuff.

I don't know. It's very interesting, huh? You have in Russia, I don't know, you know, the guy Strugatski. They are, I think, fictitious thinking about human, what's going on? And so you have a DM that's that developing to type of people, common people and various smart people, they just start it. And this two branches of people will go in different directions very soon. So that's what they're thinking about.

Yes, simple people and more complicated.

I can know that he is just his fantasy, but you know every view you have got, who is just writer and also so let's go to literature. And he explained, have he understand literature and human relationship, have he see life. And I understood that I am just small kids comparing to him. He is very smart guy in understanding He knows this predicament, he knows big blocks of life. I am used every time when I listen to him. And he just taught me about literature. And I think that I was surprised. So the managers in big companies. Most of them are guys who study English language and English literature. So why? Because they understand life. They understand models. And among them, maybe many talented critics, which just analyzing them. And this is big science, like, property. This is this blocks. Yes, there is.

I'm very modest about myself. I see so smart and grace around.