 So it's very important to be aware of it and not give up your past. Okay? Because if you follow it by your food and what it is that you want, what you have to do is, and it's not as easy as you have to do it. I guess what I mean by that is, to be aware of your past is to believe your past. This person has been charged. This is my little good solid background. I've been in this community for a long time. And even if it won't be a hard walk away, I can't go and find what I've satisfied with the years. I'm going to leave you to a map where you follow the goal. And it's going to be one hundred maps out of the world, map in the world, just to keep that as a secret. Like, at least, here it is. It is one hundred spots, see the map, which brings what I think matters, about the world, what it is. This comes from this scheme, after the presence of energy, where lots of audiences, all the industry, are part of the team here, and discuss how to go around the map. Just like you have a sector, kind of, in each program, each sector, there are specialized sectors called any map. This one is for people, the others are also. If you want to save people, use what they want, and go to the main map, rely on the map. Go to the map, and find what you want. You can go to the map, go to the map, go to the map, and then go to the map, and then you can find what you want. If I go to this other map, if I go to the map, and under there is what I do, and put it in the map, this map comes into this area, to this game, that I drink. For many years from then, these physical teams, are going to be really good instruments, that they come with. some of the features are created, are created, so that's a change thing, and this is what I'm talking about so one of the things which I've listed on there, you just see, here from what you see, there is all the details of what features are there in this case, there is building, this thing there the properties when you pass, this is the old version, which the properties before the user actually made their way and what I'm talking about here, it was added at the end of the page and then what happens is when things are kind of grey and such, users will be able to do a concept and allow type but it is this one, it is not there and there is a big problem right here, or a problem, or something that needs to be changed, it is not there but we will see it now as a project first, is view of your design, just like you can have two images just in this kind of thing and I will find it, and this group is really kind of crowded this is an application, can you have it included that's kind of what it is in most of the cases here then, there is some things really wrong, if you want to fix it, you can go ahead and do it again and again here is when everyone comes in and says, it's okay, it won't, it happens, it's a bad thing yes, everyone comes in here this is missing blocks the user can make this room come through and it will work it's kind of hard to manage, the user can show it and it will exist all this kind of things can happen in the open object all this is dense all this can happen, two million features can happen but it doesn't have any criticism it's hard for one person to have any idea of what it is and what is it, and what part of it is it that they see you this happens when you have all the plus changes that you have in the project and you have all the different models that's what this is like in the project I was given this as part of my main project and this kind of thing was, why didn't you find any changes to have common changes, say, in the product, etc you also have one that needs to be used and why don't you use this kind of thing, maybe high quality now large bits, say, a big strip of something that can still be used on the map you're bored with this thing you don't want to be insane when you work with the project and it needs to be used you have to take changes from the animation behind it, extend it, and it becomes like this you want to be managed by your partner in the map as I said, two million plus features as much as needed that can happen in the map and you want to be able to make moves in the animation system that are focused in animation this is something that I've been mentioning for a long time in the game industry human input listen from it this is a good project, and we have a box we've got a box in this format the idea is to make some money out of the real world and I'm going to use one of our animations and then when something happens these functions can happen but then people come up with something it's a very large room you have to go in that room and it's up there and you have to say something and I say, this is really JNL that provides a lot of JNL it starts from the map so this is one kind of knowledge that we have open input it needs an item coming from the second JNL which we can see small motions or compare motions it wants to compare motions or everything even a feature check if that feature is a good thing on the last game and then can be forced to enable it here in this case, it's not a great value but it's very good so if you have to make this four-hand state of life that's really good I think this is a very good idea saying how finally I made such a JNL game first of all but it can't be really it's an entire game still, if you have a whispered text like, take out your prayers for a CD I think buy this game hope to make very good such a JNL game and then whenever you buy it or something or go to a game and buy it it's not part of the system planning you have to make a JNL this is not buying it this is the system it's basically planning it but if there is one or two it's basically planning it but if you draw something and you draw something like a JNL game but the way it's like this or if you have a CD what are the changes how are they done what's the damage how do you operate this to make a problem how are you doing it and then you draw something and then you draw the numbers what are the new programming and you ship it out what are the changes in this class because it's mostly like a very very strong program but it is then when you push it it's getting a kind of very fast when you say when you say once you get once you buy it or you have it sure if you buy it or you have it you have to be fast or you have to be safe then when you push it it's not really like this and then there are a tiny number of possible problems happening how it's like going on because this again is the genetic feedback this brings me to how to prepare this for better how to find a solution to settle this problem and then take this and see that really like this how to recharge things and then it says there two kinds of things like the lab whether the lab bringing back the things some kind of yeah some kind of and see how it says that it's genetic how it comes up and it's like the last genetic as it was so so it's like it's kind of there's some kind of you need to know how to do this specifically I cannot really like this I wish to say this a little bit but very much I can show you if you see you buy this specifically from this this is the lab and now how much of that data how much you learn from the last we have a lab that this goes through this time the properties the elements the intelligence the common sense that we use it specifically for machine learning this solution communication can behave very can behave we don't want to do things we don't want to do things we don't want to do things for example we want to bring a standard data a spec similar to this so it's to simply have these decency this不好 I can't to use this decency I have to instability -... from future OpenStreetMap has a project that releases anything that changes it. Whatever happens in the last few days, it's released in 1.5. And that is what we're using. These are the changes that happened in the last few days. I'm going to go to JSON, which is there. These are the better processes, these lasers. And then it's back out of that image that we're using today to do the same as one of the eight applications. This is a work that I've also been doing for most of the time. This has been celebrating this 4 month of 90,000 plus labor changes. It's been close to a year now, but it's not going back this way. What I'm saying is that there are only 90,000 changes and users have moved that. This is our JSON. How do we use this data? I'm going to give you an example. This is the answer for that. A teacher who looks at things like this, before a professional teacher doesn't have any user base and we don't have any office. So this is our chain of the statement. I'm going to give you this idea. How do we use this data? How do we use this data? How do we use this data? How do we use this data? We use all these data sets. We have 19 approaches. The new card. She has a supervised data. She has a link to all of the systems. The supervised data has no quarter of an ounce. So it has come out 90,000 labor changes. That is the lesson. This is the lesson. Each kind of example has this group. This is one. This group. This is one. And you want to learn how to move this data to a part of your body. The challenge is, see the difference in lesson. For example, as you move one of the various systems. You need to change the amount of data of which you have. And you want to get a good result. Here you don't want to change the amount of data so things will not matter. What you want to see is, give all the ideas about the system. What you want to see is the coverage. You have to consider all these ideas and say, how all these and any particular problems could face me. This one has to be said. How do you get this problem? The same is true. So, I'm super-missed. That's the problem. So, the next step of the problem is you should learn how to set up a part of your body from a professional point of view. The thing is, adding new ideas by creating a new system. A world problem isn't a job. It's facts that are just sensitive to taking one. It's fact that it's not a problem. What I mean is, whether it's a problem, or a lack of what it has to be. And then, the last thing I want to say about this job, which is actually a new group of users, the labor that you have to just calculate your own thing. This case has to be a lack of what it has to be. So, if everyone, if you want to change their thinking, it is the teachers that are going to get a new spot, a better one for their thinking. Then, the opportunity inside is that kind of add this energy, maybe a little bit of a question. Something we all, are in the process of spaces, that you all know more than any other. Nice to see this thing in the middle of the course. And also, see it's not bad at all. It is going to come to you to add more data and make that better, get a better one. If I was to publish this here, I'm going to be coaching that thing. Because I have no idea what I'm going to say. You need to send the backwards out. You send it in. So, I'm going to pair any test results out of this. I'm going to do any test results. Then, we have a few things that we want to do. The dynamics. We have this before, I'm going to sort of do my own thing. Then, we'll start doing this. One of the most interesting things is the important number of the aggregate as well as possible. And the number of the aggregate we can ask ourselves. So, this is as in one separate thing. Each one of them can take up a sentence on them. And then they can write a comment on them. We can call it the end-of-the-world script. Or the end-of-the-world script. But, what will happen for the structure of the data is that the end-of-the-world script has such a use. The point that the parallel ad versus parallel ad is to teach that the script is unworldly. The example is that there's a teacher who has a teacher who doesn't have any children as a theory. And then, first of all, it's the main. You see, there's a thing of the sample which teaches that we're based here. And one of it was given and applied to the other two of them were labeled. It was something that the one was always saying, use more of my tool. Use more of my tool because the one was equal. But, because there was one sample that was applied, one was led to push for the brightness of that input. We've been conducting another process. So that was this, the end-of-the-world script. Each feature has all kinds of tools that can have full details based on the first script. My idea can have a superior environment for the first script. But, as you can see, you're not getting out of any classifier, any feature, any other time. You measure performance of one as a main driver and you push to one. And it's kind of seen cases that don't understand that more than all the 70 years they've ever been. But, personally, the last three months when we got this project I guess that I do not know why it's such a bad thing to do something in the production. Yeah, it's a very good idea to do that. But, she does this hard. And I think it's pretty good news as she does it very well. It's hard. But, I think it's pretty good news for this people to see if you have a question raise your hand and we'll bring a mic over. Hello. So, I work for a company called here which makes maps and I do a lot of projects like the one you mentioned, so we also have been struggling with a lot of machine learning models for validation. We need a really high accuracy and our maps are primarily used for navigation so error rates need to be 0.1% or lower. Now, the problem is that when you talk about unsupervised learning we also actually had hope that, you know, looking at our there's a plethora of data in our stores and we didn't find a lot of anomalies which can guide us. So, I would like to know like so you actually mentioned metrics and all that but where can I get some resource about the unsupervised or anomaly detection methods you used or what kind of features were there and things like that. In our case just to give an example just for the signpost we have 187 different types of signpost so you can imagine like the kind of data we had. We expected that we'd find some signals but at least so far we have not. Great talk. Thanks a lot to our talk. My question is how do you merge supervised and unsupervised learning in particular in your domain right because suppose you find an anomaly using unsupervised learning then do you apply supervised learning on that or is it like they're just separate systems trying to figure out which one works best and which one doesn't. Can you tell me what you say but I'm not as supervised as you know what we are detecting but we are getting ready to work here for very good. In the case of very good method we have to know how to get some one practice for unsupervised learning. It's not very possible to get all the ways that we have to do. So we take what's in it and it's very good. Thank you very much. Ravi Mutturu from Capital Flow Great talk. Thanks. My question is do you monetize this service? If yes could you explain how the business model works? Do you monetize these services? It looks like you guys put a lot of effort into this do you just give it as a free service or do you actually monetize these services? You guys say to come to this as a state of business model. So monetization versus for the map. The maps are great that's why it's called map maps. So thanks to you guys you guys are able to be part of this. So what it has to do with this Thank you for the talk. My name is Shubham from Microsoft. So it's like one of the slides you pointed out like insights which is the model which we built the machine learning. So I want to know about like sometime we follow insights but the model which we built based on the insights does not gives us best results. So for the solving one problem we have to approach is let someone say that this features would work out, someone say on the basis of insights like or where you will get the decision on that which insights would give the better result. Is it the experience based or it also has something algorithm on top of the insights which actually take the decision which insights you should go with this. Because insights are always the best way to go. In my practical case only the insights were not giving the better results. So we trained two models then saw which one. So what is the approach in this case is which you should always go with the models or experience based or just right. You mentioned about the anomaly detection that are you using any time series component like which models use like the AIMA models and things like that. So which one you use and for the model selection do you use any auto ML kind of feature. How do you decide that the best model are you using the auto model auto ML or it is a manual process. I have a question. Can we use unsupervised information for supervised learning. You can definitely tell it's the second day of the conference because everybody is bolted off to lunch. We have time for one more question. And then the rest of you can bolt off to lunch and the line should be a little shorter. Thanks for the talk. My name is Ayush. So the question is that do you only review the ones which are flagged by the buffer because then you need to believe your algorithm that it's pretty good by saying good. And what are the algorithms that we are using. Some particular stuff you may do. So basically for a class anything has been released. Like this buffer is released. So what is spent learning and do is correct. And then those certainly is flagged something obviously should be tested. What is the algorithm called as decision release. How to do the same process. I also faced in my company that if you have labels which are not flagged, how do you know that those are correct. So it's like if they are flagged to be wrong, you know that they are wrong. But if they are not flagged, you cannot assume that they are not wrong. So yeah, so this is a problem and I would say that you should explore things like one class classification and stuff like that. Have you tried using like an artificially created or pseudo data set to test your models. Maybe that's one way of refining your model. And so in that case, how difficult or easy to generate fake data points. Thank you all so much for your attention.