 And, well this shouldn't be a too big problem today, but if you're sitting too close to each other, please consider wearing a mask. We already now had one outbreak in a village just today, so it's actually around here, so take care and protect yourself. Yeah, with that I will turn it over to my first speaker, and yeah, please a big round of applause to El Carto, who will talk a little bit about making neural networks secure. Thank you. Test, test, hello. All right. Can you hear me well? All right, cool. So thanks for being here, I think this session is the only session where there are only speakers in the audience, more or less. This is my first talk at an event like this, so I thought I'd start with a lightning talk. The topic of the talk is how to make neural networks better with uncertainty, and we'll find out in a minute what that is all about. So if you're not working in machine learning, and you want to learn about it, you might Google for it, and you see basically images like this, there are different kinds of machine learning, supervised and supervised reinforcement, and then you think, okay, I'll start with the first one, supervised machine learning, then you see cartoons like this, and you wonder what the heck, just fruits and hats and everything, or pictures like this, again, fruits and machines and stuff, and to make this a little bit more mathematical, what supervised machine learning is, is basically finding some mapping from data X, which is the input, to data Y, which is the output, and this is all there is to it. We have some data, we have some inputs, we have some outputs, and we know that there is some relation between them, but we don't know which it is, and this is exactly what supervised machine learning is all about, finding this relation in some way, shape or form, and we call, in this context, lowercase x and input point, which is an element of a set of points, capital X, and lowercase y is a label or target or output, and this is what we got, and this is what we have to work with, and what we do is we want to find a mathematical function, which maps from our X, from our inputs to the outputs Y, and our task is to find that function. That function takes the input X and has parameters theta that we need to find, so in mathematical terms, very loose, very floppy mathematical notation, we define what's called a loss function, which is the, which depends on the parameters of our model F, and we try to minimize that loss function in order to find the parameters that model the relation in our data best, so this is what supervised machine learning, apart from all the fruits and hats, this is what it's all about, it's very simple, and now to introduce the concept of uncertainty, we start with some one-dimensional toy data, so what you see here is a function g of x, which we usually don't know, we want to find it, but for the sake of this argument we pretend that we know, but in reality what we are giving is only the input points, the blue points, with some added noise usually, think of it as a measurement from a measurement device or something, and our task is to find that orange function that we don't know, and notice that there are some gaps in the data, so we have two gaps where we know what the data looks like, but we don't have, for example, because our measurement device broke down while taking that data, and what we want our function or our machine learning model to do is to tell us that in the regions where we don't have data, that the prediction of the model should tell us that it's very uncertain about this prediction, and we can do this without neural networks in a classical statistical way, using what's called a Gaussian process regression method, this is well-established textbook stuff, and what it gives us is the red curve, you can see it, yeah, pretty much, and this is the fitted model to the data, and you see that in the middle region where we don't have data, the true function and the model don't quite match up, and in this region, the gray area, which is our uncertainty, is pretty high, and this is what you want, and this is what normally nobody does now, or not many people do in machine learning to fit a model, but also to get some information about where the model is good and where the model is bad, where basically what we want is a model that tells us, I can give you a prediction for this particular input point, but I did not see data at that region, and for that reason, my uncertainty is high or my confidence in my prediction is low, and the good news is we can do the same thing with neural networks, which don't have that capability built in, and the most simple way to do this is instead of training one neural network to fit that model, to fit that data, we use 50 different neural networks, and we exploit the fact that neural networks are super bad at extrapolating in data regions where they did not see data before, so you see in the top plot, you see barely these gray, bushy curves, so each of them is one neural network fitted to the same data, but in between the regions, in between the data regions, the models basically do what they want, and we exploit this to get some uncertainty information, and the even better news is, instead of using 50 neural networks, we can often do quite well just with five, so this means our computational load only increases by a factor of five, which for some workloads might just be good enough. There are other methods, I won't bore you with this because of time and math, there are other methods that scale computationally much better, approximating these kinds of things, the one white well-established and fast method is called the Laplace approximation, yeah, math, which basically gives us a picture pretty similar to the one we saw before, that in the regions where we did not see data, we have a higher uncertainty, and this is what we want, so the message here is we can make neural networks give us uncertainty information, which is not a standard thing to do yet in machine learning pipelines, and finally after some boring 1D toy data, let's use some 3D data, so this is 3D data, what it is, it doesn't matter, consider it's a temperature field in this room, it's very hot, so you have regions where it's hot, regions where it's cold, and when we use a machine learning model and we use only 10% of the data to train it, which I did to make the model deliberately worse, we see that there are some differences between the ground truth and the prediction, and now we check if we can quantify these differences with uncertainty, so what we first do is we subtract the right from the left and see the difference between the ground truth and what the model predicts, and this is what we get, and now we look at our uncertainty where we hope that in regions where we have a high difference between data and model prediction, we also have a high uncertainty, and yeah, so message is we're still working on it, but so far it looks quite good, so this is very recent data from my research project where I'm working on that I obtained in the train when coming here, so it's not perfect yet, but we see that we have these four kind of blobs where the difference between the model and the data is high, and we see some of our four blobs in the uncertainty, so yeah, and with that I'm done, thanks. Yeah, thank you very much for a very interesting presentation and probably one of the most mathematical presentations that we are seeing on the camp, and well, if there are any questions now, there are microphones in the middle of the room, please step forward and ask, and of course I hope the speaker will be available for direct questions afterwards as well, so with no questions right now, I would again thank you very much for this very interesting talk, please give him a round of applause for that, and with that we're coming over to the next presentation, which will take a couple of seconds to connect, etc., and our next speaker is Martin Hamilton, who is doing a quick and dirty batch-based CO2 monitoring because COVID is airborne, it will be interesting to see what this talk is actually about, and well, with that I would open it to our speaker whenever you're ready, thank you. 50 seconds to switch from one speaker to another instead of 15 minutes as you usually have between talks, so that can happen. You might be able to see a little loading progress meter here, so hopefully, boom, we're up. Hi everyone, I'm Martin, and I'm going to talk to you about monitoring for CO2, and particularly in the context of COVID, COVID's not the only thing we might be interested in, but kind of been a little bit of a big deal recently. I've been interested in air quality monitoring for a few years now, so here's some old tweets. We've got on the left, this is the clean space tag, this is a little gadget which monitors air quality, and intriguingly, you don't plug it in anywhere, it's powered by wireless energy harvesting, so it feeds off your Wi-Fi and your GSM signals, and it uses them to charge a little tiny battery, and it can do this because it's very low power, just takes a few small readings infrequently, but just enough to kind of build up a map of what's going on around you. But this is old stuff, so let me see, that's 2018, so if you think about this, this is a few years ago. The one in the middle is the plume lab's flow, and the flow is another one of these kind of personal, not DIY, these aren't DIY, these are actual products you could buy. The plume, unfortunately, this device is now only sold corporately in batches, so you can't buy it individually. If anyone wants to club together and buy a few, hit me up afterwards and we can maybe do a shared order. The one on the right is one I personally invested in, so this is something called Learnometer, and Learnometer is designed for schools, primarily, but really any kind of place where lots of people might gather together for extended periods and close proximity. We weren't thinking about COVID, COVID was just coming, that picture is from January 2020 when Learnometer actually launched, so Learnometer was just hitting the shelves at the same time as COVID started spreading in towns and cities. The thing about Learnometer is it's going to cost you a little bit of money, it's about 500 euros, and actually it tests for a whole bunch of things because we discovered children's learning is affected by noise, light, lots of little things, air quality is one of them, but there's these other things too, and Learnometer tests for a range of different possible problems with your air quality, different pollutants, and things like that. So, okay, people mostly weren't talking about air quality until, would you say maybe January 2020? Actually, maybe a little bit later than that. I've got this graphic here, which is actually from July 2020, and this is one of many explainers that you will remember from the news media. Oh, hey, we just kind of figured out this COVID stuff, maybe it spreads through the air or something, and many of you here will know, scientists have been saying for some time, virologists have been saying for some time, hey, do you know what, there's these things called aerosols, and maybe you want to have a little think about those. So, it took a while for that message to get across, and the picture on the right there, you maybe can't see it too well in this environment, the T-shirt probably is quite legible, and the skeletons saying, I can't wait to die for the economy, and of course that was the reality for a lot of people, there's this killer disease going around, but you've got to get to work, you've got to pay your bills. As we heard from the previous speaker, if you don't do all of these things, the wheels that turn the economy seize up, and everyone starves, which is not so good. So, okay, I talked about some air quality monitoring tools which have been created as products, but what I found increasingly interesting during the COVID era is how people who realize there's a problem to solve and are inventive, which actually is an awful lot of people, can put things together that are quite astounding. So, the picture on the left, or the composite of four pictures, is something called a Corsi Rosenthal box, and this is what happens when you take a box shaped fan and a bunch of HEPA filters, so you can make your own air filtration system, and this will suck in and trap a lot of the bad stuff that's in the air. And remember, this isn't just COVID. If you think about a typical school, where lots of small children running around, COVID is not the only thing they're spreading, by any means. So, this also true in workplaces, of course. And we had the beginnings of what I think we probably mostly come to call the culture war, particularly around masks. So, the image on the right is one of many, again, pieces of scientific advice, saying, hey, do you know what? If you wear a mask, they're a little bit less likely to be infected with this stuff. If the person you're talking to wears a mask, well, do you know what, that risk goes down dramatically. So, we shouldn't have to say this, right? But, of course, there are a lot of people who, for one reason or another, would like us to think masks there for wimps or masks, there for pinko, commies, real people don't wear masks. So, we've had all this stuff going on, annoyingly around what is perhaps the most useful non-pharmaceutical intervention against COVID and other airborne diseases, which is a bit rubbish when you think about it. So, what am I really here to talk about? I tried building a few cheap and cheerful DIY CO2 monitors. Remember the Learnometer? It's like 500 euros. Could we make something which does just a little subset of what Learnometer does and does it for a lot less? So, this is taking you through what I've done. I haven't invented any of these. These are recipes that people posted online. I've got the links to them. You'll see the link at the bottom of the slide. These slides are online. So, the throbbing QR code you saw at the beginning will take you to my website where there's a link to them. You can follow all the links. You can make the recipes yourself if you want to. So, this first one, this was good fun. This is quite a nice little recipe. A guy, I think, was in Canada, made this for his daughter so she could understand how safe she was at school. So, she would take it to school, a little cardboard box, and very quickly, everyone in her class wanted to know what the numbers on the box said. The picture on the right is me taking it shopping. But how I mainly used it. I wanted to kind of know if it's okay in the supermarket. So, what's the verdict? Well, it's a lot cheaper than the lernometer, but it's still a little bit expensive. And it's quite big and heavy. I called it a workout for your neck. If you're wearing this thing, you're gonna know about it. The sensor that we use for this, the SCD 30 from Sensoron, needs a lot of calibration. You've got to keep calibrating every time you power the thing up to be sure that the readings are reliable. And yeah, you've got to do a lot of soldering. So, when I was on this journey, I developed my soldering technique from zero. So, I ended up doing several hundred solder joints by the end of it. And this one probably most likely to be mistaken for a bomb. I was slightly worried about that, you know, being taken to one side by the security in the Superstore or whatever, you know, hey, we need to have a word. Two more. So, the one on your left is using the BBC micro bit. If you're not familiar with the micro bit, they kind of mass produced them and handed them out to kids in schools in the UK. A lot of people have built interesting things that the micro bit powers, it's just a little micro controller ultimately. This particular product is an air quality monitoring board that happened to have a CO2 monitor. I didn't really explain why CO2. And the answer is as CO2 falls to the ground and builds up in poorly ventilated areas. So, if you got a high CO2 reading, you've got problems. It's time to open some windows, open some doors, turn that fan on or just get out of here. So, the micro bit, that was an interesting solution. Another one is using the DIY Octopus, which is like a baseboard for DIY IoT projects. And again, the one on your right, the Octopus is using that SCD 30 sensor. How do we get on? Well, the micro bit, it's super cheap by comparison. This is still a lot of money for people, 50 euros, but it's a lot less than 500. It's not really bad shaped. And I was kind of hoping to get something that people really could carry around on their neck. Didn't work so well for that. Also, it's estimated CO2. This is not the real CO2. The sensor looks at a mixture of gases and says, well, the CO2 is probably about this much on the whole. So, you know, it's probably this much CO2 in this room. The Octopus with Node MCU as the micro controller. That's very nice. A little bit expensive. Lots of soldering. Get that soldering technique going. But ultimately, that particular kit gives you traffic lights. It just gives you red amber green traffic lights. And I came to realize, this does not work so well. You want more information. You want situational awareness. When did the CO2 levels start to increase? At what point? Where was I? What was I doing? So, this is why I was super excited. This is actually just a month ago to discover this really fantastic solution, which is based on a board called the Badger 2040. So, if you saw my tweets, I said there would be badgers. This is the badger. Or that's the badger. If that's the name of the code repository on GitHub that you can get this from. Why do we like the badger? Well, it's very simple. Raspberry Pi micro controller. Very low power. Integrated eInk screen. So, this really is. And I'm wearing it now. This thing is so light and portable. You can just carry it around with you everywhere. It's got the Stemmer QT quick connectors. So, you can plug all kinds of stuff into it. And the badger itself is only like 17 euros. So, you might not want to plug an expensive CO2 sensor into it. This is the problem still. The CO2 sensors, even this newer generation you'll see here, the SCD 40, the one I'm using and there's now an SCD 41. They're a little bit expensive. So, maybe 50, 60 euros. So, we'd like this to be something that really could be mass produced so that everybody could have one. This is still a little bit expensive. But maybe if you order 10,000 of them, we could get that priced down a little bit. It's great. Like I said, it's great size, shape, weight. And crucially, because it's an eInk screen, it will run for days and days on that one battery. There is a flaw with it. You'll notice with the other ones I made them red. I said, don't recommend. This one I'd say is AMBA. This definitely works. It does the trick. With the SCD 41, you don't have a ton of calibration to do, which is good. But unfortunately, you need to unplug the battery to charge it so there isn't a handy USB connector or something you could charge it from. So, for a regular person, you can make this. This does not require soldering. You do not need electronics experience. Do not need a PhD. You can actually just go and make this. However, you've still got that annoying problem of having to unplug the battery to charge it. So, you can't have everything. But I think give it a little bit more time and that will probably be nailed. Maybe there'll be a second edition. So, that's kind of me wrapping up. But before I stop, there is one thing that I wanted to remind people about. So, in the West, we tend to look around and say, okay, things are looking a lot better for us now. There are a whole ton of people who have not had a COVID vaccination. These people are as vulnerable as we were before we had our vaccinations, before we had our boosters. The countries in red on the map, our countries where 20% or less of the population has had a COVID vaccine. These people need things like this. They need the vaccines too. We've got to help them any way we can. And without wanting to sound too doomy about this, the image on the right, there's a screen scrape from advice from the UK government to health and social care providers, which is about, I'm going to say it now, it's about monkeypox. And yeah, it says much the same as the COVID guidance that they were given a couple of years ago. Okay, you want to mask up and get those FFP3 masks on if you're dealing with these people. There's a common misconception that monkeypox is only spread through sexual contact. Actually, it does spread through aerosols, but you have to be really close to the other person. So this is a real danger. It's really coming for us right now. We're better prepared than we ever have been. We do have a vaccine for monkeypox too, which is good. But we shouldn't let our guard down. We should think about the people who this could help. I'd love to talk to you some more about if anybody's interested. So come and find me afterwards. Do scan the QR code or jump on my website, martinh.net, got links to the slides, which take you to the software that I've been talking about. Thanks very much. Thank you very much for another very interesting presentation that we just got to see. And thanks for keeping our eyes open that there is not only we were just one problem named COVID, but tons of different things that are using aerosols to spread. Well, if there are any quick questions right now, there are microphones in the middle. There may be time for one question, maybe two. And otherwise, please directly contact the speaker for a short discussion if you're interested in that subject. Well, with that, I'm looking forward to my next speaker, Victor. Yeah. Hello and welcome. Welcome to the stage. And Victor is going to give a talk about the document documenting for my future self, well, for his future self, actually, his memo system since 2015. And I'm kind of anxious to learn what that is actually about. So while this may still take a couple of seconds, I will remind you again to please remember to drink enough that actually is important in here. You're losing lots and lots of water just by sitting here and then you're stepping out in the sun that I don't want to see you falling down. Then another thing as well, lightning talks are also here tomorrow and on Tuesday. So if you have a talk that you want to present, maybe a short idea or maybe some unfinished research, please check the wiki for the for the email. I was just send an email to mch lightning at gmail.com. There are still slots available tomorrow and on Tuesday. Actually currently I have no submissions. So I hope that there will be some more during the day. Yeah, how are the technical things looking? Yeah, like always, they figured it out. Yeah, maybe another side note, I already asked for participant as being an agent. So please do consider that. And also I know that the security team is looking for some people being available today in the late hours. So if you if you think that may be something for you, please go and see the security team. Yeah, that looks good. Yeah. Okay, thank you. So thanks. Stuff like this happens all the time. Messing up, making mistakes. And I tried to learn from that. So probably this event is going to be another memo. Because I've been documenting myself and for myself and for others since 2015. And I'm using a memo system for this. And it started thanks to Pete Lammertse, who is an engineer at the company I worked with. I once did something with a motor and modeling friction and using an algorithm for that. And I let him feel what I did. So I gave him the motor and the friction compensated and you felt it and said, Ah, do you use this in this algorithm? And I said, No, I don't. It's an interesting. Yes, it's really interesting. I'll send it to you tomorrow. And I thought I would just get some link to code or whatever. The funny thing is, like the evening I met him, I got this email with two Word documents describing another friction algorithm plus how he implemented plus some other things. And I was like, Whoa. So I replied like, this is amazing, you know, that you can share this with me. And he said, Yeah, it's because of my memo system. And jokingly said, So you probably also have a memo about your memo system. And he said, Yes, I do have a memo about my memo system. So he sent me that and I was I was sold. The funny part is everybody who worked with this guy now has his own memo system because they think it's so great. As I mentioned, he uses Word. I don't like Word. So I'm using something else with which I'll explain. Something else is that they're in the meantime, there are many, many programs that kind of do this for you. So like Evernote has a lot of note taking stuff. There's a set of custom for in many implementations. But I still think my way is still the best way. Although I think if you start with anything, you'll be helped greatly. So why do I think what was my purpose when I wanted to make my memos? I want my memos to be mine. So I really want to defend them. I'm a gnarly dog and they're my memos. I write them for me, maybe also for sharing, but the primary audience is me. So I don't want to give them to Evernote, write all stuff in an online program and find out when I have 10 years of invested knowledge that they said, yeah, we're going to switch to some kind of subscription model. And if you have more than 100 pages, then something something. So I want it to be mine. So what am I using now? I'm using markdown for note taking so I can insert in my notes, I can insert links images, links to other memos, which is really handy because you start reading somewhere and somewhere you find references to stuff you did before and you actually forgot you did it before. And I have a folder structure, which I will show in a minute. So if you want to copy it, you can copy it for you use it for yourself. And that's meant for organizing stuff that belongs to those notes. So if I write a memo, I can include pictures. But sometimes I find something really interesting on a website, which looks like it was made in the 1990s or something. And I think like, maybe this website could go offline shortly, or it's really interesting. I just make a copy of the website. And also put that in the folder to which the belongs to the memo. So I'm sharing this with several PCs I use. And I think one of the nice things is that I already switched from using mkdocs to also serve the memos to sublime as my primary editor to add them to VS code and now VS code plus obsidian to use all my to view my view and edit my memos. And I see that as a strong point, because the memos themselves have not changed. So what does it look like? Here's a screenshot of official studio code. You see all the memos on this side. So they're all just an MD file with a number. Then here you see all my talking to myself. And here on the other end, you see the rendered view. So I have a preview of whether my inclusion of image actually worked out. So other stuff I did is with obsidian. And the nice part about obsidian is that if you make you can easily get a network of your notes. So if you create references as I did here. So I created references to other memos I made, you can see how my memos connect to each other. So the funny part is that you kind of feel like, Hey, wait a minute, I'm now working on whether I should change my organize my family in an agile way, which could be a completely an under talk to how that relates to scrum or getting things done or whatever. Sharing thoughts is of course what made me brought me into this. Very nice thing is if you have markdown documents, you can use a tool called pandoc to easily create word PDF or lathech output. And for those of you who are writing code, the very nice part is that in markdown, you can specify what languages is. And when you output the render the file through pandoc, the word file has actually the correct syntax highlighting. So I'm really happy with that. Same for lathech. But then those of you who use lathech will probably start crying because the way they do it is apparently very ugly. But it works for me. That's the first way to share it. So how to do this for yourself. Create a folder, the memo folder. Every create the index markdown file. And then I just create files with have a year, year month month and index number. So it looks like this, my index. And say I want to have one for today. I create I put the images and PDFs or whatever belonging to that. And if I ever find something and I don't have time to make a memo for it. I put it in the top in the top folder. And whenever I have spare time, I tried to sort it back to create new memos. To me, this really helps because I find that whenever I'm debugging or writing something new, I write down what happened. When I'm reinstalling something, I write down like, oh, I went to this page to find this quirky little thing. Recently reinstalled my laptop. So wrote down where I found the tricky parts for this laptop to get the brightness keys actually working. So next time I do it, I already know where to find it. That's it. I wish you all very happy note-taking. And a very nice conference. Also again, thank you very much for the very interesting talk. And it hopefully gave some of us some idea how to actually like keep sort of a diary in a useful way that you can actually use for something later and not just put it up somewhere under your bed or so. Yeah, so thank you very much. If there are any questions, the microphones are in the middle. Just step up if you do have a question. Thank you very much. Do you do any kind of key wording? So you have some idea which belongs to, I don't know, writing web pages or whatever it is you do or reading a book. So do you have any approach to that? So sometimes if I don't have a key word to start with, I just start. I think that's also the most important part. So somewhere I'm doing research on systems engineering. I like model or I heard stuff about model-based systems engineering. I'm making a memo about that and start my first stuff. On the go, I maybe split my memo. It's a bit of work because I have to sort all the files again, which are in the folder. But that's the way I do it. But like I made also make made it like a books to read list. And then if I read it, there's a new note with like my lessons learned from it. And because of the links, it really gets interesting because you like somewhere like you get emergent patterns. So just start. Hi. So I've also used the note-taking systems like this for a long time. I would be in fact like interested in organizing some kind of get-together with other people to talk about patterns in their stuff and whatnot. I have used Rome research, Athens research, Trillium note is currently the thing that I'm on. I've previously had like a flat text file, markdown files. What are some like other tools that you tried or like has anything else worked out better for you than just markdown files in Visual Studio Code? No. For now. So I looked at a lot of tools. And for me, one of the most important parts is that you have to be able to use the standard markdown because that's going to stay for ages. And it's so widespread. So I'm now using Obsidian. And Obsidian has some really nice extensions to markdown, but I'm not using them because if I start using those, I'll be tied to markdown or to Obsidian. Same goes for other tools that are very specific in how you put your files or how you do. And one of the things that this system tackles, that no other systems I find tackle is that I can store a lot of PDFs that belong to a memo in my file system. So if I find a lot of data sheets or background information or just PDFs of like web pages, I can store them with my memos. Yeah, Trillium note has this, but it makes sense. Thanks. Yeah, you're welcome. Hello. Thank you for the presentation. It's a very interesting topic. My question was, I have the requirement that I need to access and possibly modify the notes from mobile. And I did not find a good solution for this. No, I'm I'm struggling with the same thing. And especially, and now I'm, this is the bad part about my system, that if you store all these large files, you don't want to sync them all to your mobile. At least I don't want to do that because sometimes the PDFs get really large or whatever. So one way I looked at it is to use the paid Obsidian service. It's kind of possible because then they do the synchronization to a mobile app. And they also have a way to do it with your own tools, but it's quite a hassle to set it up. I also looked at other systems to use note taking app on a hosted next cloud system. Then you can log in from the next cloud system and edit your notes, but it just doesn't work. It's it's it's open for improvement. So basically, no good solution up to now. Correct. Thank you. Yeah. Okay. So thank you very much again. And Victor will still be around. Hopefully, if you've got any more questions or want to set up meetings with them, feel free to do so. So thank you very much again. And the next talk is actually going to be a little bit of a hardware talk. So a ramp on this talking about do it yourself solar bike caravan. And he should be on the stage anytime soon. Thank you. Need help? Hi, my name is Rapone. I'm from the C4, the Chaos Computer Cup Cologne. And yeah, during COVID, I was sitting at home. And then I refound my love for biking. And then I saw new two people were building caravans. And yeah, I started building it. And the also the bike is converted. And so yeah, it's really hot today. And, you know, enforced talk. So the bike is actually a normal bike. You can buy it used for like 1000 and you get like really good hardware. And the conversion was another 1000 euros. And you have a motor here. You take out the stuff and put in the whole motor. Then you crunch it together. And it's actually done. It's the conversion takes about two hours. So if you want to convert your own bike in this style, maybe you do it a little bit more tidied up. But yeah, two hours and you're done. And your bike is now an e-bike. And it runs well. You have to look. I don't know about the Netherlands and Germany. It's okay to do your own thing because the law is actually pretty not lush. It says it only may be the motor may only run while you pedal. And after 25 kilometers, it may not push anymore. And so it's okay if you put in 25 kilometers and you have your own motor. And also, oh yeah, that's my best friend. I need to put it nearer. So yeah, you can actually put it in different modes. But this one is now legal. It can actually go up to 40 easily. And it's a 1.4 kilowatt motor. But I only use it with around 300, 400 watts. So it is not so it lasts longer instead of like going really crazy. Yeah, and this is the caravan. It's actually built out of, yeah, I'm just going to open it because I was not finished. So there's a lot of duct tape, but duct tape really works well. So this is going to be my window. You see this is the bed and there's like storage room down there. Because I need was no time anymore. I just put screws inside and made it like really so it's tight. And I want to like have it open because the problem is you cannot really reach the downside and you're always going inside. It's better. This is the inside. So water does not get inside. So on the second day here, it actually broke and there was a little bit of wetness on my bed. But yeah, that's okay. It's all dry again. Essentially, it's mostly built out of those aluminum fear can't bear you can see. It's it's easily viable and then you a lot of how's it called? Neaton rivets. Yeah, the things are all with rivets and only the plastic is made inside with screws. And yeah, also I have a battery and it's in the back. So it always charges the solar will always charge inside. And so it provides extra energy in front. And yeah, that is actually we are the better thing is somebody has questions because I'm sweating a lot right now and I need more water at some point. Yeah, I will do a little self organized session. And if somebody wants to like look closer, it's tomorrow at three o'clock near cows west. So cows west, I think it's if you come from backbone, and you go past the heaven, then on the left side cows west. Yeah, okay. Thank you very much. Just two words for that. Cool stuff. And some water. Yeah, I'm not quite sure if you actually managed to get that one down step down the stage. So I think I just keep it. Yeah. All right. If there are any questions at right at the moment. If not, and or if you would like to inspect the caravan in more details, I'm really sure he will be hanging right around outside for a while. Okay, now switching speech speakers becomes a little bit more interesting than usual. Okay, so now we I'm happy to announce Martin Richter, who is giving us a very short introduction into to chaotic dynamics. I'm really interested to see how that introduction is possible in 10 minutes when it takes other people years to even understand the basics. So I'm very looking looking very much forward to this talk. And with that, I can hopefully, yes, send it over to our next speaker. Okay, thank you very much. So my name is Martin. I use at the, I work at the University of Nottingham. I am a physicist, but I'm working as an applied mathematician. And one, one things I one of the things I care about is chaotic dynamics, almost to a fault. And that's why I thought maybe to give you a short introduction. So I hope by the end of these 10 minutes, you have a vague idea of chaos is and that it's actually quite all around us. And I want to give you an introduction in terms of a small and simple example. So let's assume we have a some form of landscape. So imagine a hill and a valley and another hill in the middle and a valley and then this goes off to infinity. And you just put a beat or say a ball of some kind here. And then you just let it go. And as you might, as you might think, it just starts moving, it accelerates, decelerates, gets slower again, and then accelerates again, has some higher velocity here, goes all the way up here, decelerates, turns around and goes back. So nothing to see here really. And that's not at all chaotic by any stretch of the imagination. But this is the kind of system I want to talk to you about. It's an asymmetric double well. So if you focus a bit on the depth of these troughs, this one is a little less shallow than this one is. Unfortunately, I will bore you with a little bit of maths formulas on the next few slides. So here we are. What we just saw interactively was a particle and a hilly landscape of sorts. And physicists like to model that by so-called potential functions. And this is the one I used. You can see that here. So it's x to the power of four. That's basically what makes it go to infinity at the very end, to the left and to the right. And then there's a minus x squared. That's basically what does the bulge in the middle. And then this mu, because we all love Greek letters, times x is what gives it the asymmetry. So if mu is larger and larger and larger, the asymmetry would be much larger, but it's not too big here. One property of this system is that its energy is conserved. So you can write down this equation here. It's called the Hamiltonian functions. And it's a sum of two contributions, the so-called kinetic energy and the potential energy. And if you put all the equations in, you'll see that this number is a constant. So what do you do with that? I mean, you just saw this ball rolling down and up again and down and up again. Let's assume you want to do this yourself. Now, solving such a problem is already quite complicated. So you usually do this numerically. And the way you do this is, for example, physicists, we love Python, so we use Python for that. There's a routine called SciPy Integrate Quot, or Ode Inde, actually. And then you write the system in a different form. So what you can see here is Newton's equations you might remember from school in a slightly different form. So if you implement that, what you will see is that there is a link between what the particle is doing and this potential. And I would like to draw your attention to the right-hand side of these slides. So at the bottom again, here is our double well. This well is slightly deeper than this one. And on the picture above, what you can see is a so-called phase space. So instead of plotting what the particle is doing while it moves along, you register its position and velocity or momentum at the same time, and you plot that in a two-dimensional plot. So each of these lines here is actually what one particle is doing. I'll give you an example. If you start, for example, here, it would just close this line here, and that would mean it would oscillate around the left-hand side minimum. There's the same thing over here. So you start it here, and it would go around and this circular motion of sorts is what corresponds to the oscillation back and forth. If your initial position is further up on the left-hand side, say, you're on one of those curves, and instead of going back immediately, you go over to the other one, all the way down over here, and then you have this almost eight-like shape. And this is basically how we prefer to look at these things because they allow us to see everything at once. So just watching this ball going back and forth is nice to look at, but it's actually very hard to understand things. So therefore, we prefer this phase space picture. But again, there is still no chaos. In fact, if you look at that, it couldn't look any more regular. So how do we get chaos into the game? We get chaos into the game by adding a so-called external driving. So instead of having this double well, we will have this double well, and we will now start to rocket gently from the left to the right. So you might think, OK, that's all made up nonsense, but in fact, if you do experiments with, say, lasers or atoms in laser fields, that is some of the approaches you can use to model that. So I'll show you again what happens. We have again our ball here, and now at this time, the potential will move up and down on the left and on the right, and we will be watching what this particle is doing. I'll also plot in a faint color where this potential was originally because the rocking motion is periodic, so at the end we will come to that state again. So here we go. So you see this potential goes down here, and then it goes high up again. It comes down here. So this was one period of the driving, and it does a few of them, and you can see that now, of course, the motion of the particle is way more complicated because it now is an interplay between what the particle is doing and what the potential is doing. But again, we have the problem that we can't really learn much from that. It's maybe nice to look at, or it would be nice to look at if I wouldn't be that bad at programming, but we can't really learn much from it. So we'll just leave it at that and think, is there anything we can do and plot it in a form like that? So we just follow the particle again, and instead of watching where it goes, we just put the x and the velocity component into the same coordinate system, and we'll just have a look at how that looks like. So that is here. That's an interactive plot. Again, in the background are these lines we saw for the case of no driving. So when the potential is static, these would be our solutions. Right? Remember, the static case was your particle started here, and it just went around like this, or it went around like this, like this eight, or it was over here when it was on the left-hand side. Now, how do these orbits look like, or these trajectories look like when there is driving? So let me just start one. So if you can see the mouse pointer, it's currently over here, I'm going to click, I'm going to start a trajectory there, and just integrate what happens in the background and give you all the results here. And that's what you get. So because of the driving, it doesn't really lie onto these one-dimensional curves anymore, and it gets worse. It does look similar on the other side, but if we say go here, there's actually not much resemblance anymore. So for the time being, we might just take these lines in the background away, because they are not of any help. And I just want to show you what happens by starting a few more trajectories. So all the old trajectories eventually they are just colored gray, and only the newest trajectory is in color. But as you can see, this plot is not as helpful as you might think anymore. It's a bit like a drawing of a three-year-old, really. So there was a famous mathematician called Henri Pancaré, and he came up with a thing called the Pancaré section. So this is one of those cases where less is actually more. So what I'm going to do now is I'm going to show you exactly the same data as you can see here, but a very specific subset of it. So I'm not going to show you all the orbit, x and momentum, or x and velocity changing over time. I only give you the points at integer multiples of the external driving. So remember, the potential does something like this, and every time it's here, I'm going to give you these two values. And just this again. Snap, I'll give you these two values. Snap, and I'll give you these two values. So it's a very tiny subset of all these drawings you can see here. Before I do that, let me start a few more just to convince you that looking at all the data is actually a big mess. Okay? And now I'm going to remove most of them and just leave these two points I was talking about. And then it looks like that. And it's a bit like being out in a hut in the woods in the night and you go out and you look at the stars and suddenly everything is clear. So by removing plenty, or the majority of these points, we actually saw that we can see and understand way more. So there are a few initial conditions like these here which would fill two dimensional regions densely. That's what mathematicians call them. And then there are others which form these one-dimensional curves. And it turns out that they are what we call integrable and they are what we call chaotic. So if you have two points close by and you start them, they essentially diverge exponentially fast. But now you can actually see way more things than you could before because everything is so much clearer. So for example, we could now start trajectory up here and we see this blue thing. It pops up all over the place but you can actually see way clearer what's going on. If we put the whole trajectory back in, I bet you would not have been able to spot this in all the chaos which happens there. So that was what I wanted to show you. That if you start with a simple potential, start rocking it that you can introduce chaotic dynamics but that sometimes less is more, namely by removing most of these points you actually arrive at much clearer pictures and that is actually something you can then use to describe for example molecular dynamics in external laser fields and that's what I'll do for living. Okay, thank you very much. Hello, Mike. Yeah, thank you very much for showing us to create chaos and then if you have chaos, how to still be able to see stuff inside that chaos. That was very interesting to note. Thank you for that. And yeah, we do have one more talk coming so please don't run off immediately. And now if you've got any questions on this talk, there are microphones in the middle. Please step forward and just ask your questions. And if there aren't any immediate questions or more complicated ones, I hope he will still be around a bit for more detailed talks. Okay, yeah. Why? Why doing things like that? You mean? Yeah, what's the use of it? So the main problem you have or you're facing as a physicist that your task is to describe nature. And you can't, it's impossible to do this with all the details included. So somehow you need to find models. And depending on what kind of physical aspects you want to model, you'll use different languages from your mathematical toolbox. And this is a very prominent one. Now chaotic dynamics are very, very complicated. So in order to understand how they, how they work in principle, it's much better to use simpler models first and examine what kind of generic behavior can show up in them. And that's one of the simplest cases to look at. Hence, that is one thing you can use to, to get, to, to understand it. Thanks for the talk. Much appreciate it. You're way more educated than I am, but I actually tried looking at the, so you were talking about the constant that you have in the velocity, right? And when you start shaking it, velocity is actually well obviously changing as well, right? Would you agree with me that what we're looking at is basically going from a linear regression model to a nonlinear regression model here? No, that would be the wrong mathematical tool to describe that. So the field we are looking at here is called dynamical systems, more specifically ordinary differential equations, while these regression models, that's more statistics and things like that. So here's no uncertainty. This is, it is chaos, but it's completely deterministic. If I, if I choose this initial condition here, it's completely clear what the next points will be. It's just that if I'm an epsilon off, so 10 to the minus 15 or whatever, then after a few iterations you'll see something completely different, but it has nothing to do with, with any of that. Yeah, okay. Thank you very much again and a big round of applause for teaching us chaos in just a few minutes. Okay, and then we are coming to our last speaker for today. Petros Ilios, who is going to give us a short presentation about his research project, which is, if I understood it correctly about drones and using them in SAR operations. Yep, Ben. Okay. Hello. Yeah, my name is Petros Ilios and I'm going to show you some short intro on the project. We as a group called Searchwing are working on. As you can see here, it's about a drone and yeah, we use it for Mario team search and rescue missions for and we help NGOs like Seawatch and all the others who work in this field to find people in distress and rescue them. So just give you a brief overview of what it is about. So we build these drones mainly for search and rescue in the Mediterranean Sea area. So as you might know, they are still people fleeing for various reasons from global south, mostly to the global north. And often they do this by sea and I put here a slide or a screenshot of the recent data from the UNHCR, where you can see how many people flee in 2020. It was 52,000 and 800 people died while doing this journey from the south to the north. And this is still not nice that this is happening. It's basically unacceptable. So we try to help the NGOs to help find the people before they drown. So in the NGOs who work in this field, I put just a few of them here. It's SeaWatch, for example, or CI. Rescue ship is another one. So there's a bunch of them. And from now and then there are new ones also emerging and they go to the sea. They search for people in distress and mostly in the Mediterranean, but also in the Atlantic. And if they find some, they rescue them and bring then hopefully to a part of safety and where they can go and land. But it's not only the NGOs who try to find the people. It's also governmental actors like Frontex and they also use drones. You can see them up here. They are way bigger than the ones we have. And they fly around and you can see it here in this kind of patterns over the sea and search for the people. And if they find them, they tell these to the Libyan Coast Guard and then they catch these people up and bring them back to Libya, which is not legal by international law, but they do it anyway. And when they catch them, they don't treat them very nice of human. You can see here that they basically beat them with ropes and stuff. So it's not very nice what they do. But yeah, we as this group of drone enthusiasts and programmers and mechanical engineers, we try to build these drones. And yeah, I can show you how this works, actually. This is a screenshot from a movie I want to show you now. This gives a nice intro and how the system works. OK. So. We. Start from the ships. And yeah, we not only starting from the big ones I showed you in the slides. We also start from small sailing boats like you can see here. And this is a video material where we were on a mission in Mediterranean and on the sailing boat. And yeah, here is such as some, yeah, how it works. So we have a tablet. We program in the tablet the trajectory the plane has to fly. It's quite small, the drone. It's quite cheap, actually, as we use a hobby RC parts, which are mostly available from off the shelf. And yeah, it's we have to the thing is we land at the end in the water. So we need to make this whole thing waterproof, which makes it a bit. It's a bit unsoftly related problems we had to solve. And it was very hard for us. It took us basically four years now to find a working systems to do this. And it flies quite far. 100 kilometers. It's it's not that much if you consider the whole Mediterranean, but it's something at least, which can help the ships to find the people. We fly at 550 meters altitude and we have like roughly 20 centimeters per pixel resolution, which is enough to find these yeah, two by 10 meter boats. And we also fly this rescue or this square pattern you just saw. And yeah, and then we take pictures after we took the pictures where we land, then we get the drone on board the ship again. And there we analyze the images. And I think this will also come in a second that you can see this in a picture. Yeah, you make four first forward. Yeah, as I told you, it's waterproof. And yeah, we retrieve the drone with a small boat and then we get it on board and we do the analysis, which is assisted by a computer vision algorithm which tries to find the boats and then starts the images. So the human still needs to check every image, but we just saw them because we cannot trust those algorithms. And therefore we always tell the people you have still have to watch every image. Okay. So I hope this gave you a small idea how the system works in general. Okay. So I told you we were working hard on the hardware issues we were facing with the waterproofness. And in the end, we settled down with a 3D print, but instead of doing it on our own, which wasn't working very well, there were always holes in the 3D print even if you took so much effort to make it waterproof. It's not really waterproof if you put it under water for some time, it's still water coming in. So we settled down with a laser centering box which was then finally waterproof. And we put there in all the components like GPS, Raspberry Pi you can see it on the bottom right, the green one. There the two cameras are connected which are on the bottom middle side, mid area. And we have a telemetry on board. There's the antenna which is coming out of the box. And we also have a wifi connection maybe soon but that's still in development. We have on the right-hand side, you can see a small board which is the autopilot. So the drone is flying fully automatically running autopilot, autopilot, autopilot. Yeah, I told you we have this boat detection algorithm and to give you an idea how hard it is as a human to find all the boats, maybe you can now try to actually find all the boats in this image because it's not so easy. I mean, one is quite obvious, but do you find any other boats? Or how many boats are in it actually? Any bets? I hear 10, four, 14, all wrong. It's only two, but as you can see it's very, very hard to distinguish all these small boats from waves and stuff. It's not a really easy task, especially if you're on board a rescue ship for two weeks and there are high waves all day, then you're super stressed and you cannot really focus on finding all the boats in these roughly 3,600 inches per flight. So we need some computer to help us on this task. Yeah, I'm also coming to an end now. This was just a small intro and we might do another self-organized session today where I can say more details and stuff which went wrong and which works quite well. We are always trying to find more people to help us on our stuff. It's mostly open source or yeah, it's almost everything is open source what we do. We search people for software, for hardware and yeah, on the right hand side I put some tasks we recently have, currently have regarding the image recognition task or if you find anything which appeals to you or you want to learn maybe also then just hit me up and would be very happy to talk to you about this stuff. So yeah, thank you. Maybe one more info regarding this self-organized session so it will be maybe because my laptop has just died so I need to check if I still can do it but it should be at seven in the 1,001 nights village just to the left to you. Yeah, thank you very much for a very interesting overview and if there's, well, it's about saving people's lives so if there's anything that is worth your support that's probably their work. Yeah, for any questions, microphones as usual are in the middle. So if you've got questions, just about please. Hi, I was, thank you for your talk. I was wondering who's funding all this? Currently or until two months ago it was basically our own money from the pockets and now we get some funding from Stiftung Synodretton so they fund us with some money to pay.