 Restart. I am a master's student. I'm working on a radio telescope project here in Brazil. And my task is to reconstruct a red shift of far away objects. And I will tell you how I do this. First, you have to know what is redshift. To understand what is redshift, you have to record what is the Doppler effect. The Doppler effect is the changing frequency of a wave in relation to an observer who is moving relative to the wave source. One example of Doppler effect shift is the change of pitch heard when a bus sounding a horn approach and recedes from an observer. Compared to the input frequency, the recieved frequency is higher during the approach. And it's identical when it's passing by husk. And it is lower during the recession. Redshift is the light version of this phenomenon. We see it all the time. And we can see, if you look at the sky, we know that there is blue stars and red stars. The red stars are far away from us than the blue ones. And because we have this in the color spectrum, this frequency is higher than this. And when we see a celestial object that there is red, it means that it is moving far away from us in the expanding universe. This is why we this one main reasons we study this thing, because we want to know that if the universe is expanding or if it is contracting or if it is stationary. When the rebel constant is a number that measures how the universe is expanding, the rebel number is calculated measuring the redshift. There are two main approaches to know the redshift of a star. We can do data driving analysis. Is that what I do? And we can do a template 50 analysis using physics theory. I don't know physics theory because I am a mathematician. In my research, I use that data and use two machine learning methods to study this phenomenon. If you use machine learning methods, we have to have a training set. We have two kinds of data. We use our training set. We use spectroscopic redshift and photosy redshift. Spectroscopic redshift refers to the measurement of radiation intensity intensity as a function of our wavelength. We use it to describe the intensity of all electromagnetic events of our stars. We have infrared information. We have all the colors information and a bunch of gamma ray information. It is a more complete analysis of an object. It can be an atom or it can be a star. But photosy redshift, it's just some colors. We can just, how is the intensity of red, yellow, blue and infrared in the stars? There are many systems, there are many systems of photosy redshift. It depends on the survey and the data you analyze. For example, I use pen stars data, which have five colors. We call these colors magnitudes. We compare these photosy redshifts. We have with five colors of the beams of the spectroscopic data we have in our training set. Frequently, we have many stars of uncertainty and some of them are there. We have a lack of inputs to training because when we do this analysis, we have the map of skies and each is divided in pixels. We can have, but we are comparing spectroscopic data to redshift to one. There are some missing galaxies in one or another. We compare to surveys to match our training set. We have the uncertainty of the different algorithms we use and we have completed training sets because we are dealing with a very sensitive data that could be incomplete and you depend on the survey you are working on. Here in my work, I use two of the most popular astronomical softwares. For doing this, I use GPZ and ANNZ2. The first is an approach that uses Gaussian processes to estimate a photometric redshift. We are focusing on the variance of the set because our data has a bunch of errors. We have to use statistical analysis because the source of your research is far, far away and we don't have a good precision. The second we use is ANNZ2. It is an updated version of the original algorithm which uses neural networks and boosts decision and regression trees. Does anyone have a question? Hello, I'm Francesca Deschi and I have just come. I was wondering if you could resume briefly the core of your work. I work with reconstruction of a photometric redshift. Do you know what is redshift? Yes, in general, but if you can explain something else. Do you? Are you an astronomer or something? No, I work in the field of computational fluid dynamics, so totally another field. I'm a beginner in Python, so very basic notion in astronomy. I will try to explain what I do because I'm still learning about it. I am a mathematician and I've been working in a cosmology group. I'm doing this in my master thesis. To understand what redshift is, you have to record what is the Doppler effect. When there is a bus near to you, there is a change of pitch when this bus sounds in a hard approach and recedes from you. Compared to the empty frequency, the empty frequency of the wave is higher during the approach and it's lower during the recession. Redshift is the light version of this phenomenon and we experience it all the time with sound. You have, you can see this with this. We have here blue and we have here red. The frequency is high of the light when we are here in the blue. As we go, the frequency of the light is lower reaching red. So when you see ice star that is red, it's bluest than water. It's more near than that. Why we why we study this thing? Because we want to know if the universe is expanding or if it is contracting or if it is doing nothing. We have some of this that the universe is expanding and the measure of redshift allows us to measure this using a thing called rubble constant. And this is one main reason we study this thing. There are two main approaches we use when we measure redshift. We can do a data analysis. That is what I do because I don't know enough physics to do a template fit analysis. They use healthy physics theory. And in my research I use machine learning techniques. There are many machine learning software that we can use for it. But I use specifically two sources. I use GPZ. They use Gaussian process. And I use ANZ2. They use neural networks and boosted decision trees. But to do this we have to have our training set. Because of this we have the spectroscopy redshift and photosurgery redshift. What is the difference between the two? The difference is that photosurgery redshift has just five colors or six colors. It's colors that we actually see except the infrared. But spectroscopy redshift is all the bands of the spectroscopy, all the bands of the spectrum. It includes x-rays, gamma rays and a variety of different bands of the spectrum of the atom. It's used in atom analysis and it's used in stellar analysis. So we have a lot of difficulties when we are doing this because we have a very uncertain source. We have frequently, we have incomplete training sets. We have the uncertainty in the algorithm we use. And we have a lot of uncertainty in our imputes of training. Because of this we do not use a package like learning TensorFlow directly. We use it there, but we use it there in some way. We can get it's variance. Because we work with some uncertainty of the redshift because we cannot, because we have this uncertain source and we cannot have the exact result. This is why we work with specific software which are described by academic papers. In my work I use these two, GPZ and ANN2. Do you have some questions? Yeah, I have no experience about these two softwares. But I was wondering, because my work is about reconstruction in medical imaging. So do you do a sort of reconstruction process? My data is, it's not an image. It's a number. It's the measure intensity of some color. It's not the image of the stars. Do you know SSD? I work with two services. They are available on the internet. I cannot find the chat. But if you search, you can you can find them. It's 10 stars with P, A, N, N, stars and S, D, S, S.