 And the previews of it now, I think it's amazing, in fact, it's amazing, it's the first time I've seen it. So apparently, there's supposed to be a snowstorm here tomorrow, but right now, it's blue skies. We'll see. So, a neural network in the framework we just talked about is a model. Okay, so what is a neural network? This is my one-sentence description. So that's what pooling is, it just means that I'm going to cut down on the number of the size of my input data, basically. Then, typically, people pair that with a convolutional layer. Hey, Geert. Hey. How you doing, man? This is your vlog, right? It is. That's awesome, I love your vlog. Thanks, man. I think people should subscribe below. Hit that subscribe button. Click on Geert's face to subscribe. Click here. Is that how you vlog it? You're just holding it like that. Yeah, man. It's good. It's good. It's good. Shoulder worked out. I bet it is, holy shit. Here I am on Davenport's vlog. Yeah. I mean, vlog. Vlog. Vlog. Yeah. I can't do that. Well, you can do that. Nice. After a double S, my shoulder was really sore. Well, it's huge to be making. I've been seeing you making your tea. Genmaicha. Genmaicha. I'm all about genmaicha. I love toasted rice. Yeah. Genmaicha. That's my go-to. I had this problem that I drink 19 cups of coffee a day. Yeah. What's the problem? You run out of coffee or? I start to feel like I'm going to kill everybody I know. So I've tried to take it down to 14 cups a day and go to genmaicha. Genmaicha. Yeah. Oh, my god. All right. So day two. We'll do the same as yesterday. Elliott's been looking at using F3, my F5 parameter package, an animation showing the position of the star. Is it off by, like, one? No. So you can see very clearly what's going on in the log scale version. At some level, we don't care because we just need housekeeping data about what the star looks like. All right. So I was able yesterday to resurrect some code to just make fake flares. Top left is just an example of one day of light curve where we added some white noise to make this look Kepler-ish. So if we find a bunch of big flares, we should be able to predict the color of the noise. So we should have a prediction of what that red noise looks like based on the flares we've detected. Yeah. That noise should be skewed as well. So you could also look at higher order statistics. So this is the real version of the fake data. So these are powered density plots for two different Kepler stars. So I would like to more rigorously compare these models. Oh, my God. That is a sweet comparison. That's incredible. That is so clear. So we've just wrapped up day two of the test sprint. And it was pretty successful. I've been working on flares for the last couple days with a big team of people. I think I'm getting tired of working on that. And I'm going to move towards working on something to do with Tabby's start tomorrow. I think it's going to be a little more exciting. But we've learned things. We've made some cool graphs. We've done a terrible job of taking video for the last couple of days because we've been too busy actually making code, which has been better, but it's less fun to watch. Good morning. Day three of the test sprint starts today. And it's supposed to be a snow day. So I spent the last couple of days working on flares, which is what my main research topic is. I also wrote a paper this last year about Boyajin's star. This is a mysterious star. This star has a bunch of dimming events up to 20%, which have no definitive explanation. There have been explanations. Some of them are more believable than others. So one thing I'm interested in with tests is just how many of these stars should we expect to see? And also, it might snow. It's not snowing now. It looks like it's been raining, but that could be fun. All right, so far no snow. Oh, yeah. Flatiron building. Cool. Two comp, please. Much two comp. TensorFlow has modern implementations of all of these things and sort of built into it. One of the hot questions they're getting into this. One of the reasons TensorFlow talks about graphs is that the deep learning world describes their network architectures as graphs. OK, there's a little bit of snow now. Join us. Yes. Talk about citizen scientists. Yes. If you want to turn as many times as you need. Every time you turn, I don't wish you'd not turn. MIT is really excited about helping in any way with a citizen science project. And so being able to field all of those questions is a resource that we need to have. It's at least one person to staff this full time. The scope of the project has not been planned. That's exactly true, yeah. And so one of the things we have to decide is which of these two things or whether it's something else we want to do. Started today eager to think about this mysterious object that was found in Kepler. And we ended up talking about it a little bit. This is one of the biggest successes, I think, in citizen science. And it was driven by Kepler, driven by these really amazing light curves. Unfortunately, I didn't think of anything novel that we could add with tests other than we need to look at every single one. So the big problem I'm having today is I've hit this wall. I've hit this sort of energy wall but also just sort of like a planning wall. For me, one of the challenges, before these kinds of hack weeks or hack days, I can have a lot of anxiety personally because I know that if I don't prepare for them, if I don't have a sort of strict preparation for them, I can walk away feeling very disheartened or very burned out, you know. One of the big reasons I come to these events is to try to work on new projects. I had some cool results, I think, on the flare side. Most of that was good discussions with people. Things for me to chew on. It's nothing that I can act on yet. And so after a couple days of working on that, I'm left without anything to work on. And this is one of the big challenges of these kinds of events is that if you sort of fall off the productivity bandwagon, it can be very hard to feel like you can get back on it. Now, we do a lot of things to try to catch people, to try to pick them back up and put them back in projects. But even for an experienced hacker or hack week attendee like me, it can be easy to fall off that wagon. So I had to have sort of this reckoning this afternoon, paper and pen and just figure out what are the things I wrote down. Test related projects that I can hack on. And so I'm gonna circle back on this flare thing which I got burned out on after a couple of days and now I'm gonna start working on it again. And I think I'll be able to produce something really cool. So that's how the day is going. In weather news, it's snowing but there's no snow on the ground basically. This winter storm seems to, this nor'easter, this winter storm seems to kind of be a dot here in New York. It's probably an okay thing.