 Temperature phase and spoke with the high-temperature phase for example, right and this is super revolutionary thought that Juan Carrasquilla and Roger Melko had in had back in 2017 that in the end it can just be an image recognition problem So now instead of doing some clustering algorithm on our configurations We can just take the configurations Labeled them face a face be like we labeled star trek and star wars for the one zero Vectors like we just saw how we were updating the loss function on this minimal example That makes a zero difference for the algorithm and we can label We can label these things just as well Since then there was a huge progress in this field, but in the end things will always be looking like this that you have a That you have some kind of image and you can feed it into the network that will classify the image for you Again, let me then maybe at this point they ask you this question. I was asking you before So now you don't know yet if it works. We are going to persuade ourselves in a second But let's say that it does So this takes away this problem that was mentioned before from the audience that not every Not every Problem is simple enough to solve with the clustering algorithm Probably you can believe me that if I can distinguish really sophisticated differences between different breeds of dogs with the AI image classification for sure I can classify the two phases, right? So then again, like is the physics solved now or is there some remaining challenges with this Great one another one Yeah, that's a good question right so now I sort of and I will in a second serve you on a silver platter platter Monte Carlo configuration that someone generated for you But yeah, the question is what if you don't have this right? So there will be there are still some there are still some remaining challenges I made some Programming exercises for you. I hope that's fine with everyone. Did you bring your laptops or? Do we have like one laptop per three people at least? We do right So let me I will just explain a little bit and then I will then I will give you a link so classifying this icing configurations that we look at at the beginning is Is a Is actually in a machine learning language super super simple problem and the neural network that does that is Written in a programming package by torch Literally just like this you write your linear layer that takes 900 because the lattice is 30 times 30 then it feeds it into the 32 neurons in a second layer and then we map on this binomial distribution of of Face a and face B. So we go from 32 neurons to two neurons That's it super super simple. It's is this kind of you know Sophisticated artificial intelligence problem, but what you actually need is three lines of code You will have this written down from from us and Yeah, you can experiment you can experiment with all the different network architectures and What you are going to recreate at the end of this notebook is the plot from this from this Karaskea and Melco paper That tells you that if you plot the output layer of your network You put the probability of being in one phase and probability in the other phase They exactly intersect at the at the phase transition Temperature and this is the kind of plot that you can reproduce with this super super simple neural network that That just has a that just has the three lines of code that I showed you so Let's do that if you go to my web page either you can go alishka grapple Slash ICTP or in the news on the front of my website. I also have a link in The in the updates. Let me open it up so you arrive to the Website and then on top of the news There is a ICTP ML school and then you look at the notebook one and It will open in Google co-op if You are new to so if you want to work in the notebooks you want to save a copy in Drive or Save a copy in GitHub just you know duplicate the notebook because you are not allowed to make changes in mind so So you don't lose your changes Then I will super super quickly walk you through it and then for the remainder of the time I just let you experiment What you will realize when you look at this notebook there is just some introduction about this icing model face transition if you want to read it you will realize that a lot of it It's like data preparations and preparing labels and plotting the shapes and so on and so forth There is a lot of code that you don't have to worry about so much. You just have to press Shift enter to run these cells Um Basically when we were setting this code for you up we mirrored this like a super basic pytorch tutorial for a binary classification in a way that everything is written as a class that has a That you can just that it's like a data independent and you can just reuse it our goal being that You have a piece of code that you can then just stuff a new data set into for our own machine learning Application so there is sort of a lot of written down But you can just run through it for now and then you can save it and you have have it for later If you want to make your own neural network and then the main thing I was just showing you comes here in the step three where you set up your model so what you can do now is to is To you know, maybe change the number in the in the middle layer or add more layers Yeah, that's it then you compile the model you pick your optimizer here You need to remember this learning rate if you want to experiment with that and then again after that the training is a Predefined definition for you that is just like a yeah problem Independent so you can copy paste it to your next problem And when you are done training you will get some nice accuracy and loss function plots and Eventually you will be also able to plot this kind of plot and see the phase transition For this this is the first notebook in the school so everything is pre written down for you So if you are new to Python or new to machine learning, you can just you know read or read on the cells If you are already Profession then you know go to town and get a better accuracy than me because I didn't try very hard So also, let's see how many questions we have about the code now. My group has to raise their hands Ruvan you too They are here to help you with the with the coding problems so let's just let's just spend 15 minutes read through this and And yeah, see if you can make some changes make it work ask us questions Okay, so we are nearing your scheduled coffee breaks. I Walked around I feel like everyone Managed to open it and got it working if I if you did not and I missed you Come talk to me And I hope that like running this few cells and looking at how it actually looks like to build this kind of classifier that you got some That you got some idea how it works if you're new or you got to make some yeah better layers with a good accuracy feel free feel free to come talk to me about your code if you want and Yeah, let's take a 40 minute coffee break and then we will be back with some more complicated layers Thanks, Eliska for this very pedagogical lecture Yeah, so I just asked to stay to take a picture Maybe you can take the picture here in front Victoria. I don't know if you fit everyone here. I think so