 So throughout my scientific career, I have been interested in working with pattern recognition tools and pattern recognition is the old fashioned name for what we now call machine learning. And indeed this project is about discovering extra terrestrial intelligence with artificial intelligence. Now, one of the something is going wrong with the control. So one of the interesting things about our universe is that we have over indivisible universe we have over 125 billion galaxies, like the Milky Way is our own galaxy but we have 125 billion of them. And that's only in the visible universe and of course the rest we cannot see. Then in each galaxy there are about 100 to 400,000 million stars, and that's also true for the Milky Way. And each of these stars hosts at least one planet, and a considerable fraction of these planets is in the so called habitable zone, which means that it's not too cold or too hot for life as we know it to emerge. It's a long life that we don't know, but like that we don't know we don't know anything about. Okay, so one of the things next slide please. One of the things that we can see in recent years is an enormous upsurge in the collection of data about so called exoplanets, these are planets that are moving around stars other than the sun. All these stars are far away so it's hard to see them through a telescope, but as you can see from this picture this is showing the number of detected actual planets over every the past years. You see this green bars the green parts of these bars reflect the recognition or detection of these actual planets by means of the so called transit method, and I will explain you what this method is. Next slide please. So with this methods. The data for these methods is collected by so called satellites and one of them is the transiting actual planets survey satellite tests. Please click it yes this is an animation showing tests, and it's currently in the sky out there collecting data for us it just had two years of data collection of a lot of bright brightness variations in stars. You should imagine that it hosts several cameras that are looking at sky at the sky and each pixel is illuminated by one or more stars and it's tracking the brightness variations over time. So next slide please. This is a kind of animation that shows the transit methods so you have an artist impression of a star and you see a planet moving in front of it and a nice dip in brightness. And the idea is if you use this camera of tests or other surveys you can detect the passing of an exoplanet or something else in front of the star between actually the star and the camera of the satellite. Next slide please. What you see here is what a light curve. This is this brightness tip is looking like in reality it's quite noisy, and this light curve shows the intensity or brightness variations over time astronomers refer to this as the flux that just that's what you see on the vertical axis. And the horizontal axis is time. So the point is it's very hard to recognize patterns here and you have two options to find these patterns. One is to use humans. And actually with a previous survey the so called capital survey there was a planet hunters project as a citizen science project, where people all across the world could analyze these kind of curves by using a simple web interface. And, but as you can see it's hard to see anything here, but the next slide shows you what a more a better light curve looks like and this is one that is the trend and some other signal processing tricks have been applied. And what you see is a regular pattern of dips, and these dips correspond to a planet moving in front of a star. Now to enhance the signal to noise ratio and the visibility you can superimpose these dips and then you get the graph shown on the bottom. And this pattern that you see here reveals something about the shape and the size of the object moving in front of the star. And this shows that this is a planet, but it could also be another star if you have an eclipsing binary as it's called that that's just two stars that move in front of each other, it might look very similar so it's a subtle pattern recognition and toss. Next slide please. Now the predecessor of test was Kepler. And in AI research this Kepler data is used for the simple reason that it's not only providing light curves but also so called labels so that you know for each light curve, whether it's corresponding to a an actual planet or not. Next slide please. So I was looking at a fixed position fixed region in the sky, and this animation if you click you see the animation is the collection of all the data of Kepler over different quarters. And what you see it's comparing the, the length of a year that's the horizontal axis, the earth is of course around 365 days. There are a lot of exoplanets that have a shorter year, and on the vertical axis you see the size so many of the planets were larger, probably this an artifact of the way that data is collected. Next slide please. So all this data this data release was used in 2018 by an engineer from Google and an astronomer who used a deep convolutional neural network, which is at the heart of the current AI revolution and is used for image recognition and many other tasks. So they used it to train the system on recognizing light curves, and they used in this network which is quite custom, an enormous amount of number of free parameters these are values that have to be set by a learning algorithm to 29.8 million parameters. So they achieved an accuracy on this data set by Kepler of 96% correct, and they achieve that, amongst others by using the engines of Google, of course, to automatically set the structure of the network so called hyper parameter optimization. Now together with my colleagues Coco Fisher and boss, boss ma next slide please. We try to simplify this network by reducing the number of parameters. This simplified architecture and this is ongoing work now has a 70 fold reduction with respect to astronaut in the number of parameters as you can see it's much simpler, and it suffers in terms of performance only 0.7%. And of course we try to improve that but the hyper parameter optimization was not done with Google, but was done by Coco Fisher, and he's a great researcher but he can of course not compete with all these servers by Google. So another line of research is anomaly detection and for anomaly detection we use all kinds of techniques that take a part of the light curve and try to predict the next samples of the light curve. The traditional techniques to do that for instance in economics are a Rima and related models linear models, but we also experimented with transformer network which is a kind of novel deep learning approach. And we were able to predict the future occurrences of the light curve and also to detect the deviations from that anomalies with a 40% improvement in root mean squared error. And we applied it to the capital data and we found of the all the day we found about 40 anomalies, 39 of them were related to machine failures of the satellite itself or rotation movements, or some sunlight what was that was reflected in the camera. And one was a rediscovery of tabby star and tabby star is named after an astronomer that studied a particular star, but actually the light curve the anomaly of the light curve was detected by planet hunters the citizen scientists. So next slide please. They discovered this light grid. And as you can see, you see these dips but they're not regular and actually the depth of the dips is much larger than you would expect for a transiting planet. So, there was all kinds of speculation that this might maybe relate to an idea of freeman Dyson once proposed that alien civilizations will try to build a sphere surrounding a star to collect energy. So maybe we had detected an alien civilization. And, as you can see, this is something that does not look like the regular pattern that I showed before. Next slide please. But after some follow up studies. Now that astronomers discovered that the most likely explanation for this strange light curve is a couple of asteroids or a damaged planet that is associated with a lot of dust. So, unfortunately, no alien civilizations detected yet. Next slide please. What this shows is that with these deep learning methods, you can contribute to science by helping to shift to the ever growing scientific data volume in astronomy cosmology, and in the search for extraterrestrial intelligence. For instance, there's now this breakthrough lesson project which is generating lots of data, and also other astronomy projects that are creating an exponential increase in the amount of data it's undoable for humans to to shift through all this data. So this means that AI is changing the way that science is being performed, not only in these fields but also in other fields. If you think about the recent breakthrough of alpha fold for the big mind breakthrough for protein field protein fielding. That's one of the many examples that you can read about in the papers every week. And I think for all branches of science is essential that you understand the strength minutes left Eric of AI and that it's essential to all scientific disciplines. Next slide please. So that brings us to the question, is there intelligence out there. And that's of course a question that we cannot answer yet. And you might be familiar with the Fermi's paradox. So, if the universe is teaming with back so planets and by the way by the time of Enrico Fermi it was not known that there were so many actual plants, but he knew about the enormous amount of animals that are out there. So why haven't we seen them yet. Now there are many possible explanations one popular explanation which is a bit aligned with our side guys is that I might be these filters and of course we know about these filters or meteor meteorites that can damage complete civil or eradicate a complete civilization what is dinosaurs or more advanced civilizations, pandemics, we're witnessing one now, or nuclear war or anything that can wipe out the civilization. And it would be very sad that there might be other civilizations out there but we can't see them because they already are gone they're already gone. And I would like to end with a nice quote by Max Tecmar is researcher in physics but also an AI. And we should be open to the possibility that the destiny of life in the cosmos is up on us, in other words that we are alone. And in that case we should be better stewards of the earth we have. Thank you very much.