 Test, test, test, does it work? Which microphone is in here mute? The camera is working, right? Hello, hi. Oh, wow, OK. This is the microphone, right? But it's just, how do I figure out the microphone? Where are the settings? Oh, my God, edit. Let's see. Customization, webcam, public, education, customization, slow mode, redirection, no. OK, I guess the microphone is correct. And the camera is this one. Yes, very good. All right, so why we are here? Because we are saying goodbye to this nice place. Yeah, OK. So yeah, as you figure from, I guess, Twitter, this was going to be the last time we are casting from this apartment, because the apartment goes back to wherever whoever had the apartment did. The school is leaving the apartment from the leasing company. The leasing company is going to be destroying the apartment, so there is no more apartment. Therefore, I will no longer be able to stay here. And I will have to find a different place. So well, I just already found a different place. I will be moving in a week time. But many things happened just in just the last minute, and that was a bit hard to deal with, because finding housing in one day in New York City is not a joke, I guess. But yeah, the new place has been found, so next week I'll move there, and then you see where I am. I guess I'll take some pictures or something. Do you have questions or anything? I know we are a small group here in 26 people. That's so nice, at least there is someone saying hi. Because also over these past days, who do I talk to? Everyone seems not to give a fuck, especially at the school. Everyone is like, oh, please, you have to leave. Where am I supposed to live? That was insane. It's like you're here by accident. Actually, yeah, I've been told that. You've been here because we kindly gave you this apartment for your temporary position, because of course I'm a temporary employer. Employee. Yeah. What's happening here? Somalia. Wow. OK, impressive. Are you still teaching? So that's the thing, right? I cannot stop teaching. I'm a teacher. I became a professor because I love teaching. But it turned out that if you love teaching, it's really hard to be a professor, especially in the US, because here to be at NYU, it's a research university. And you are supposed to be doing research and getting grants and getting funds, which university will take for half of it. And that's how you are becoming an asset for the university. If you are not doing research and getting grants and getting funds, then you are, well, you're teaching students, right? They're actually students paying for tuition. And someone has to teach these students, right? And usually what happens is that research professors are forced to teach. And some of them like it. But then you're not incentivized to teach because you don't make any additional money, regardless. Like you're not independently on how much effort you put in your teaching. It's not going to change the amount of money the university is going to get, right? Students will still come because of the research that other teachers, other professors are doing. And so if you care about teaching, I guess you cannot really be a professor in a research university, which seems pretty obvious. But the thing, if you're in a university like NYU, you have the luxury, let's call it this way, of interacting with very brilliant, very many brilliant people. And you will feel stupid every single day, which is an awesome feeling because you have all of the opportunity to learn something new. And like for example, now I'm teaching this summer course for undergraduates. And the first day, OK, every time the first day with me, it's a bit different because, OK, I'm weird, right? I'm different. But and also I'm a little bit sarcastic. And I'm not patronized. What's the word, the nice version of patronized? I tease, right? I tease a little bit. I tease you because I try to get you to react, to answer, to participate to the things we are doing in class. I've been doing this with this new batch of freshmen basically or whatever. I am weird. But yeah, I'm doing this with these new kids. And the first day, there is almost conflict, right? Because there is this kind of friction of me trying to get them to engage. And they are not used to engage. And they are like, oh, why are you forcing us to talk back? Because you have to actually open up and communicate. Otherwise, I cannot get feedback and I cannot understand what you understand and what you don't. Anyway, what was the point? Hold on. So I got them to interact and they are having a lot of fun. I forgot. I forgot the point. Oh, no, hold on. I remember how we came here. OK. I don't know. What happened? I forgot. I guess the point was that, oh, yeah, maybe the thing that I teach in a different manner, right? I like you to get to watch it. Oh, yes, yes, yes. Yes, I remember. So yeah, I was saying that here at NYU, you always get to interact with people that are smarter than you, which is amazing. And they were complaining. So the first day, they told me, oh, wow, you're like, I was talking about biology or science or something and something, technology. And they're like, oh, and I was saying, oh, yeah, this is something you learn in high school. And they're like, ouch. You're so harsh on us. We don't know these things. And you're telling us that we should have learned them in high school. Now we feel like idiots, right? Well, don't worry. It's OK. As long as once you're aware that you don't know something, you can always go and look it up, right? So that's the major thing that I also saw that also kids, when I was teaching in primary school, they struggled a lot. I mean, everyone struggles because the educational, like how education is done in a classic way is maybe not the optimal way, let's say this way. And it's also another thing, right? So whenever we are in class, I abolish rising hands for talking. Why are we rising hands? Just wait for your turn, then talk, right? And then they are not used to this, right? So they are this class that I have this summer, 10 kids, and they are all rising hands. Why are you rising hands? What's going on? Just talk, right? I mean, oh, I don't want to interrupt. Well, why not? Who cares? Like, you can have a conversation if people are supposed to raise hands to talk. Like, have you ever had a conversation with a friend? Are you raising hands? What the fuck? Like, no, it doesn't make any sense. So because also, OK, at least in Italy, there is like a distinction between when you talk to your peer and when you talk to the professor, you have to change the grammatical form. In English, you don't have that, right? You don't have a respectful you and normal you. But in French or Italian, Spanish or Chinese or anything, you have also the formal way of the accursed word in class. Yes, I do. Yeah, so you have to use the respectful form when you talk to, when you address the professor, right? The professor. You don't call them by name. You cannot call a professor by name in Italy. That would be hilarious. But when I teach in Italy, there is no way people will call me a Professor Cantiani. What the fuck is Professor Cantiani? Alf, right? I guess they can actually pronounce my name. But most of the people outside Italy struggle. Yeah, I know, also in India. So India, the Indian educational system, is very similar to the Italian one. And that's why, also, I find myself very, how I say, like whenever I exchange experiences with my Indian colleagues and friends, we see things in a very similar way. Anyway, the point was that, yeah, so you shouldn't be raising hands, no, when you have a conversation with your friend. Therefore, you shouldn't have to raise hands whenever you ask questions in my class. We are a conversation, not just who cares. But anyway, so they still keep raising hands. They gradually start learning not to. And then what happened is that there was one of these kids that was not used to, well, none of them were used, but one particular kid was struggling in asking any question. And so now there are two options, right? There are two cases. One case is when you don't ask questions because you already know everything, and so I'm boring you. And therefore, I cannot force you to ask questions because I'm like, if you know more than I'm teaching right now, then you should just, you can do whatever you want. I shouldn't, I cannot demand your attention if I'm telling you something you already know. But so then I don't want to force you. But then on the other case, if you don't understand what's going on, you're not asking question because you feel that you're like somehow ashamed of not knowing that that's a major issue with the educational system. And again, I have seen the same issue in primary school when I was teaching in January. These kids are like, so I ask question, a question, and then they ask answer to me, not to their peers, I asked for example, oh, can you explain how something works? And they're like, oh, blah, blah, blah, blah, blah to me. Like, why are you explaining this to me? I know how it works. You should explain it to your classmates, right? And they don't realize that, right? They don't realize they can communicate to each other and they can actually move forward together. So yeah, that was it. And yeah, there was this one student that really struggled in getting to participate and ended up in tears basically because of the pressure that I somehow brought upon the student, right? And then I'm like, okay, maybe I'm too strong, like maybe I'm too forceful. Like then, I may doubt myself, but I don't think so. Let me see a few of these questions, right? Because I've been talking because there were no questions, but there are questions about the panic attacks. That was horrible. That was two nights ago. That basically was the before, wait, when was it? Three nights ago. I didn't sign the contract for the new apartment. Oh yeah, because how do I afford? That's the funny part. How do I afford a apartment similar to this one without a proper paycheck, right? So this apartment is amazing. It's facing a huge internal yard. It's like 10 minutes from Washington Square Park, actually less seven minutes from Washington Square Park. It's provided by the university and I believe this apartment, it may cost like $5,000 a month as a rent. But the university cover most of it, right? So then I can afford it because it's just a small fraction. But then if I'm leaving this apartment and I would like to get something similar, where on earth am I finding $5,000 to pay a rent like every month, that's insane, right? That's why that's what caused all these kind of troubles because the lady from the apartment, she's like, oh, please leave. And I'm like, leave where? And she's like, not my problem, okay? Whose problem is that, right? How am I supposed to afford a $5,000 apartment in New York City without having a decent salary, right? I mean, you're supposed to earn $40,000, 40 times as much the salary, right? So if you have a $5,000, right? If you have like, hold on, you should be earning $200,000, right? Is it correct? Am I hallucinating things? It should be, right? I cannot do calculations on the, yeah, I think so. I mean, the zeros are, there should be right, right? Anyway, so yeah, in New York City, you're supposed to make 40 times the, yeah, yeah. You should, you're supposed to make 40 times the monthly rent. So if the rent is $1,000, you're supposed to make at least $40,000. If it's a $2,000, you're supposed to make $80,000, right? If it's a $5,000 apartment, you're supposed to make $200,000. I'm not making $200,000. So there is no way I can afford a $5,000 apartment or they don't let you, right? So you need actually guarantee and so on. So who's paying for the apartment? We don't know. Okay, now we know, Jan is helping. But yeah, before, yeah, before knowing that, I didn't know where to go, right? Because where am I going, right? Going to live in a, yeah, I don't know, okay? What happened here? I see, I saw a heart. I don't know, are you saying a heart? Okay, something happened here. Yeah. Arizona teaching, how about coming to Arizona? Yeah, so, okay, you're sending some emojis. I have no idea how you're doing that, but okay. So no, the point is that in NYU, right? I have very smart colleagues and also there is Jan from which I've been learning a lot in these past six years. And that's why mostly I'm still here because although I might have finished my, let's say, apprenticeship with him and I kind of know what he thinks and how he works, I'm writing the book, right? So I'm trying to give back to him, right? All his mentorship. I'm trying to put it in a written form and he's supposed to be reading this thing but now he's traveling again. So it's been actually hard recently to get a hold of him, but that's kind of the, he used to be always the case, only during COVID he kind of calmed down. But then now he's back again traveling everywhere. But yeah, so then it's gonna be vacation time, so not sure when he's gonna read the damn thing, but hopefully soon. Yeah, yeah, so I need to, that's actually one of the reason why I didn't want to quit, right? Maybe that's the only reason that the fact that I need to finish this book, right? I cannot, I mean, it's unreliable. No, I'm realistic. It's unrealistic that thinking that I can finish to write a book if I don't have a position where I have free time, like time to write a book, right? If I go work for a company, then I, oh no, you should do what we tell you to do not writing books, right? So that's the major issue, right? Like how did you get to write a book? You need free time. How do you get free time? You need to get a place where you get free time. Let me see a few of these questions, right? I didn't read anything because I was doing this monologue. Hi, hi, hey, Prof. Yes, oh damn, what's up with the mic video? Kinda laggy. No, what do you mean it's laggy? Is it laggy? No. Hi, everyone. Your mic is fine, thank you. Are you still teaching always? Somalia, okay. Wondering what happened since you said you almost got jobless and thought you will stop teaching. Yes, that was the thing, right? So I almost quit my solutions. So how am I going to be affording something similar? I couldn't, right? Then they have to fix this. Otherwise, well, I had to fix it, right? How do I fix it? I find a job that is pain enough for me to have a decent quality of life. Much love for France. Employee, thank you. I know that I couldn't remember the word, yes. I followed your classes lesson two years ago on YouTube. That's awesome. Stay strong, thank you. I learned a lot from you. Thank you for that. You're not weird. No, I am weird. I know I am. We love you for, yeah, yeah. Sure, but I am weird. You're talking about having a hard time. I know you because you're, oh yes. No, no, I'm talking, right. So the hard time at NYU is that it's okay. Everything comes from the fact that I got here as a postdoc, right? I finished my PhD in 2017 and then Jan hired me basically as a postdoc. Then from two years of, two and a half years of postdoc, I graduated to a faculty fellow it's called, which is like some sort of junior faculty if you want to call it this way. And it's still kind of considered a postdoc. But then your faculty at the time, right? I mean, in the title. And they reduced the paycheck. Then they're like, okay, we give you, well, I forced them to give me an apartment, right? And now I've been having the apartment since I got promoted with the less money salary position. And but I was happy, right? Because I'm here and 15 minutes walk, walking from the office. It's a nice area. But then now that I got further promoted to visiting professor, they're like, oh, okay, no more apartment. Okay, where do I go live? I don't know. Do you curse words? Yes, are you answered? Yes. Oh, even in India, yes. America is very informal, yeah. Rising hand and in grain through. So, wait. Rising hand is ingrain though, yeah. Consider polite in a group situation, yes. But we cannot have a conversation if we all have the right hands. How do you end up with panic attack? I told you before, I just tried to slip through. Thank you. So said, you're not too strong. Some people are just very shy. Oh, yeah, yeah, I see that, yeah. Well, I know I am different, right? I'm kind of more hands-on type of teacher. At least like if I'm in dancing class, you know something funny here in this country? It's like, if you're, okay, when I did take my dance lessons, well, during my PhD, the teachers were from Europe. And so, based on who they were teaching, they were, like if they were teaching to a European kid, they used to put their hands on the kid, like to show them the moves, right? Because when you dance, you have to move your body, right? If someone helps you move your body, you understand better. But then in here, in US, you are kind of not supposed to touch other people, otherwise they freak out. So it was so funny. But we are more direct, I think, in Europe. I mean, we don't kind of care about this much space that they have here. Although I see when I go back home, I'm like, wow, this is way too close. I kind of got used to this kind of additional space. Um, but hold on, what happened? Many things. Let me get it up. You're not too strong. Oh, sometimes, right? Hello there. I cannot read Russian. Pak, oh, Pak. I know Pak, right? I think I know this. You translated a lot of stuff, right? I know Pak. I cannot read the first name. Okay. My Cyrillic reading skills are very bad. New York is way too expensive to live in. Yeah, I already answered. A video is laggy. No, why is it laggy? Sorry. Yeah, I don't know how to fix that. Video keeps freezing. Ouch. Yeah, I don't know. Maybe the internet connection sucks. I just thought about consulting for different industries on how to implement specific AI technology. Technology to keep them on what's possible. Tons of companies in e-commerce are looking for to adopt AI. So I guess that would work. But I would be having, I need to create then a company, right? Because I'm on an expiring J1 visa. Like if I have one more month on J1 and then the university will be filing for H1D. And I think if I'd like to be self-employed, I either should have my own company or I should have a green card. Currently, I don't have either. So it becomes a bit hard to do this kind of consulting. That's why also there are no ads on my video, right? YouTube videos, ads will be legal for me to put on. You know, welcome to US immigration laws. Okay. Hey, Alf, I'm a new to deep learning. I have found so many playlists regarding deep learning on your channel, but I am not sure from which playlist I should start. Would you have? Yes, of course. So I would say the spring 2023, no fall 2022. Hold on, what is it? Fall 2022 is the latest most updated version, but it's gonna be just a partial update, okay? So if you see the website, let me see the website link. If you go on my, oh, maybe I didn't update the, let me see. If you go on my website, which is at call.github.io, then you go to the dactics. Yeah, there is the first link, which is going to be NYU Deep Learning Fall 2022. Let me post the link here. Copy, paste. Right, so this is the latest link, like the latest course I taught with the most updated material. As you can see from the website, it's just a partial update from the spring 21. So spring 21 is the most complete version of the course, which has the highest, the all like special effects I learned. So again, if you go on my website, you can find more information, right? Copy, paste. So here you can find information. Also, if you click on the YouTube section, YouTube archive, you can read there a few things spring 22 through. Okay, I actually didn't update at that part. So I have to update the website with the YouTube archive, where I tell you exactly which are the most, okay, I'll update that. Thank you for the reminder, okay? Yeah, so again, what is my suggestion to, in order to, you know, to eat my, no, eat, how do you say? Feed from my videos and so on. We'll be checking out the spring 2022. Spring, right? Spring 20, no, fall, fall, sorry, my bad. Fall 2022 edition, right? Which is the one I linked here on the chat. And that is just a partial update. So you have basically the diff, only the new videos are there, because it doesn't make sense to upload all the videos. It takes way too much time, like each video, if it's one hour video, it takes me maybe four hours, if it's just normal editing, or if I have to put in all special animations and so on, it's gonna be 12 hours per hour. Like if I make all the math animations and things and the axis and the things flying around, then that's taking a lot of time. And if I already have some, if I already have a video where I spend plenty of time to make animations and things, then it doesn't make sense to me to upload a newer version. We doesn't have really new content, but has less animations, right? So I would consider, yeah, I would suggest you to use the Fall 2022 edition for the newer version of the videos and then rely on spring 21. Let me scroll back. Okay, whatever lies in the future, I just wanted to thank you for being such an awesome teacher and researcher. Can't wait to read your book, thank you. That's nice, thanks. Yeah, the book should be very nice. Thank you, Ricardo. Arta, do you work with any Turkish student? Of course I do. Serkan Karakulak, he was one of my master students, like he was in my class and then he actually, we worked together for a few months and now then he was working for Pepsi, then he moved to Meta. He lives in Kashin, made sure too much information, but he's an amazing friend and I often cook for him and a girlfriend and he has an amazing apartment and you spend quite some time with him and with them. Yeah, he's a very good friend. And now I have a Turkish, well, there is a Turkish girl on the floor, but she's not in my group. Very thanks, yeah. Thanks too. I decided to start deep learning now and I work in this field, that's great. I'm from though, I'm based in Chelsea, oh, in New York City. And if there is anything I can do, I'm happy to try to help whatever, whatever way I can, even if you just need to chat, okay, that's so nice, Akash, thank you. Time to set up a company for Alf. Yeah, I know, maybe. Hello, hi, be right back, okay, Spring 21. What about maybe working with Lightning? Sebastian is doing, yes, that was my actually, that is exactly what I was going to do in case they were not going to give me an alternative here in terms of finding a place where to live. What can you say about the prospects of graph neural networks? It's interesting to know your opinion, thanks. Okay, sure, I think you may want to hear instead, Petar Belichkovich's opinion, or Michael Bronstein, or, I guess, Taco, Taco, what's his surname? Taco, I forgot his surname, Cohen, yeah, I think so. I think they have a much more refined opinion about the subject, okay? I only teach, well, I teach everything, but in a kind of, I wouldn't say shallow way because it's not shallow, but in the fundamental part, right? I try to cover all topics in my class, and given that I try to cover everything, I'm therefore no expert in anything, right? So that's always the trade-off. Either you go for breadth or you go for depth. You can't really do both unless you're young, LeCun, but okay, sorry for the joke. Again, if you want to know a better actually opinion, like a more sounding opinion, ask one of these three guys, okay? And Petar is amazing as well, like speaker, so I really recommend listening to them, okay? As a master student in Turkey, planning to apply for a PhD program in the United States, I would like to know about the possibilities of remaining in the USA after completing my PhD. It's not a big issue. Like, again, my master student got a very well-paid position now, before at Pepsi now at META, and he's very happy, he has a very nice house, I told you before, he's doing a very nice life, I think. Yeah, so yeah. Yeah, the major issue is that if you're like, in my case, I don't have a permanent job. I have this kind of one-year contract job, which is creating some sort of issues. So yeah, well, yeah, it is scary here, but if you find a job, right? But again, if you are getting a PhD or if you get a master in this field, it shouldn't be too hard. Yeah, so I was thinking to, well, the corporate jobs would offer you twice as much that amount, right? Like Bloomberg or so would be like, yeah, twice as much, what you said. And that would be totally like, I could do that, but I am working on this book first, right? So I need to finish up this thing. It doesn't make sense to me to move on while I'm not done, right? So I can move on. I don't have any issue with moving on. But yeah, I cannot leave things half done. It doesn't make any sense. And I liked first to finish what I've started. It would be like one of some down with the book would be like some sort of graduation, like thesis kind of, right? At least in my opinion, such that, okay, I've done this, it's packaged, I move forward. Let me see what are the other questions here. Will there be a book released soon? Well, I guess when Jan reads that thing, Jan hasn't start reading. So the first draft is ready. So the first draft are basically first four chapters. Now I'm writing the fifth, well, the fifth chapter is actually the second chapter, which is about that propagation. And something funny is that I found lecture notes from Jan from 20 years ago, which are still relevant, most of it, and actually very well written. That's very surprising since Jan doesn't really, okay, Jan doesn't spend any much, Jan doesn't spend much more time, additional time in teaching, right? He's already spent a lot of time getting where he got. And he definitely, he used to spend time on these things. Now he spends time differently, right? Then anyway, I found this lecture notes from him and I'm, you know, updating them to the 2020 something, right? Like with 20 years in the future, sorry, from when he wrote them. Right, so whenever he read the book and fixes the obvious mistakes I made, then there will be a release of the draft, not before, right? Because it doesn't make sense for me to publish, well, publish, release something that he taught me while he's not approving what I've done, right? Let me see. I really like your classification energy perspective lecture. Are there papers that you can recommend on the topic? No, there are no papers. Okay, there is maybe one paper. I don't know if it's good or not. I mean, Jan doesn't like it, but because he says obvious, but your classifier classifier is secretly a energy-based model. It's been cited 400 times by Will Grathful. Let me post the link. Again, I thought it was nice, but again, so that's the major issue, right? With the fact that people are not aware about these energy-based models and then anything that is somehow telling you something, like people are like, oh, wow, this is so good. Well, yes, we knew that, but it's somehow people are not listening, that maybe that's it. Let me go back to the questions. Again, I don't know how to fix the lagging part for the video. I don't know whether I can do anything. Maybe I should have used the OBS, right now I'm using Chrome as the live camera sharing. Maybe OBS could have created some compression or something better next time, sorry. The book with its pretty representation. Can like an undergraduate student skip straight to PhD? Kind of long shot, yes. Many do actually a direct PhD and skip automatically directly from undergraduate to PhD. In US it's actually possible. It's gonna be a slightly longer PhD, but you can do that easily. One of my coworkers, during my PhD time, she completely and she started doing the PhD directly after undergraduate. I thought it was unfair, right? I have two master degrees and I'm like, wait, why did I have to spend so much more time than you did? Anyway, sure. I know we are all a bit afraid, but it's okay. Why you are not looking for 200? Yes, because I want to finish first what I started. It doesn't make sense to me to... I mean, I see it in the other way around, right? I would like to finish first what I started and I can move on. And of course I keep teaching, right? So my idea, like my plan was to finish this book and then find a job and be adjunct. So basically teaching part-time. Can you find time to write a book? Plenty of authors have full-time jobs and families with kids and are still managed to write books. Writing, I guess. Yeah, this book is running code, right? So the book trains neural networks and then draw figures by itself. It's a major example of literate programming. I don't know if you are familiar with the term. It's something that Donald Knuth actually proposed. Literate programming, you know? Let me show you. Literate programming here. So it's like a, it's a major reproducible thing. And everything is coded in tech. Can I get a sneak peek? Yeah, the sneak peek is in the fall 2022 edition of the course. That's why I suggested before you should start from there because you actually have part of the book. I teach with the book as well. Yeah, it's addictive, I know, the book writing. But yes, yes, I will take a rest. Also, I will be going to Hawaii in three weeks for ICML. So I hope to see you there if you are coming. Jan often mentioned that unsupervised pre-training of computer vision models is always better than using Siamese networks scheme instead of in painting. Which paper discusses this topic? Wait, what? Is always better using Siamese networks instead of in painting? Okay, I see that. So there is the, okay, fine. So I don't know, okay, I don't know the paper discussion but the rationale behind that is that if you want to do in painting, you have to come up with details and details are something that you don't necessarily remember. But what you remember of something is usually like the big picture, what, how are things, you know, who were at a specific party? You don't exactly remember what is the shirt that a specific person was wearing, right? So that becomes irrelevant. And so if you would like to improve your understanding of the scene, you shouldn't be focusing on details. Therefore, in painting, which is, you know, exactly determining the specific detail, it's not the way we actually remember things. Whereas if you're using Siamese networks, you have like basically, regardless of the dress someone is wearing, you just try to remember who came to the party and what they've done, right? Regardless of the details. And so the Siamese networks are basically invariant to small details perturbation, which make them, you know, focus on the actual major structure of the scene. I guess that the major take away there and what paper discusses this topic. I guess every paper, depending on what they're pushing forward, they will say one thing or the other, but I believe that if you want to read something, would be the latest paper from Mehta on Jepa, right? Jepa, paper, Mehta. I, Jepa, right? Here, copy, paste. I haven't read it. So, but I assume he's gonna tell you what I just told you. Because that's what Jan keeps telling. You haven't read any paper? No, I don't write papers. I should maybe, but I don't have really the time. Well, I'm writing a book, I guess. Many, okay. All the things I write in the book to me are obvious. So I don't see how a paper, like papers are, you know, research paper are like things that people don't know. I think I write in the book to me are very basic. So I don't know whether any of these things can make it to a paper. Actually, one thing there is, I develop a technique for like, dividing the sinks and sources of a stationary field, which are otherwise not distinguishable that may get into a paper. There is a kid jiggling outside my front door. That's second time, well, not second, multiple times he comes by, so it makes me smile right now. Thanks, I did it in paper. There's no lagging anymore. That's good. You and Jan looks so similar. Okay, I take it that is a compliment. Hold on, what happened? I scrolled too much. Oh, did you answer my question? Could you repeat the undergraduate to PhD one? So I think there is going to be the recording, right? This video will be up on the channel afterwards. You can check out the thing. Time to read this part, okay? You left me hanging on Twitter, by the way. What? Last week's quiz, oh yeah, I know. Yeah, no, yes, I know, because my life kind of got interrupted in the past time, right? I was trying to push work even if I was going crazy, right? So that's how we cope with this kind of help. You just keep doing what you're doing and try to go through, but then yeah, I had to really put some effort to fix all the things. So that picture actually is going to be coming up in Jan's book, right? He's writing finally the English version of his general public book, and I'm the one behind the illustrations of his book. So the figure you're talking about, let me show you whether on some Twitter, like my God, Twitter is broken these days. It's horrible. It's Elon Musk, it's really bad. Like, oh, the irony of hitting your limits due to a complaint. Yeah, I don't know. I think it's not very bright. Like, why would you limit people's ability to view tweets? Like, we don't even get paid to write the tweets and I'm like, are we customers? I'm the one providing the tweets, right? How the fuck are you limiting my ability to work for free for you and for the stupid? So we are talking about this, this curve, right? I don't know if you saw it before. Do we know what it is? Yes, because you read the solutions underneath, right? Do we see the link? Does it work? Oh, there is the lag, right, for the messages. Do you think, hey, when will the book be released? That's a good question. You can check more, if I find more information on my website. Which is going to be here. Oh, we can't see anything you're seeing. You can't see anything you are seeing. What do you mean? You cannot see the links I put. What do you mean you can't see anything you're seeing? You don't see the chat? There's so much lag in this chat. It's not like Zoom gets immediate back and forth. The link wasn't showing up. Oh, it did, finally, I see. Okay, so I guess, yeah, there is some lag in the scene. Right, right, so what are these curves, right? It shows you three different curves, right? Can't see other screens you're seeing. I put the links on the chat, right? I'm talking about the links I put. The second last, the one on Twitter. If you click, you should be able to see the other screen. So are those these send curves? No, these are interpolants, right? Those are approximators. It's like doing, well, no, linear regression with a neural network, right? It's a real neural network with 330 and 3000 hidden units. If you were on your PhD, what would you work on? It's a good question. I think I like control, like model-based reinforcement learning, or I wouldn't work on, back in the days when we had this kind of boom in guns, everyone was working on guns, so it becomes overcrowded. So I don't really like when it's too many people trying to work on something, like it's way too noisy, it becomes too crazy. Yes, autonomous agents, I would say yes. Let me think. Someone asked me this question the other day. I forgot, what was my answer? Hold on, I should remember what I said. Okay, why are you waiting for your book? Do you have any suggestions about statistics, math books to deepen the understanding of the energy-based perspective of the planning? For instance, some physics book. So I don't think so, because we are just slightly getting inspired, so I mean, some basic book on statistical physics could be very interesting to read. There is this actually, okay, good question. Let me show you. What is the name of the book? Books, physics, no, nonsense classical mechanics. No, nonsense classical mechanics. How do I show you the link here? So I like this book, which is somehow going directly to the point. Are you also solving industrial problems if so, tell us more some, please. Just for inspiration, thanks. So the industrial problems. Well, I think the major problem here is the fact that data is expensive. And so again, I guess like the major thing that every industry is going for is going to be this such supervised part. Why is that, right? Because all these networks take forever to train, takes a lot of resources. The amount of data they need to ingest is huge and everything gets much easier if you have some pre-trained models. Those pre-trained models are biased towards the type of data you're using. And if you're using supervised learning, then your model is going to be very kind of specifically tuned to do a specific task. Whereas if you just use A arbitrary, like some technique or self-supervised technique that is well made, then you're supposed to be able to train this large network, which then can be just fine tuned. So I guess that's the best industrial problem that we are trying to solve here. Same for self-driving cars, right? So how can you use these networks? How can you learn all possible things? You don't have enough data, you will never have enough data, right? So if you're on a road, something will happen that has never happened before. And then if you're just trying to rely on supervised learning where your model is just figuring out whether you're in one of the black cases, then there will be always one case that has never been occurred before. That's why you have to try to invest in this kind of generic learner that are able to perceive reality and then deal with whatever you have. So we have six minutes left on this cast. So let's wrap up. That was pretty nice. I didn't take you for a tour for the house because I think I don't have a cable long enough to show you around. This camera is connected with a HDMI cable. Yeah, I don't know. Maybe I should show you without the lights, right? Okay, the last two minutes I turn off the lights. I show you a little bit around maybe. Let me see. Hi, sorry, I'm late. Any news about Wendell Boogie? I'll show you the link. I wonder if it's possible to apply the approach with the energy on grass. There must be something interesting. We need to think about this topic. Thanks, yes, somehow. New ideas come up somehow. Thank you. So this energy thing is not anything different. Like it's just something that's why I think that what is the major confusion? It's a alternative way of seeing, looking at things, it's not nothing new. It's everything you already work with and dealt with can be seen from a different perspective. It's like you have a coin. You can always look at two different sides. But then one side is actually superior because it allows you, it has much less restrictions which is the energy perspective, right? Which is very more flexible, let's call it this way. You can have many more degrees of freedom and therefore allows you for a possible larger exploration like a larger pool of options, okay? Whereas if you go for other techniques, it's like, oh, you only have to do this, this, this and it's somehow, you know, you are not let, it doesn't let you explore that amount of possibilities, right? Can you briefly describe the energy perspective, how it differentiates probabilistic and models? Yeah, that's the book about, right? So everything will be there, don't worry. Or your YouTube channel, I don't know. To me, the energy-based model makes more sense. Yeah, I guess it's more, yeah, it doesn't require additional, the additional initial framework, a probabilistic framework. So at least for me, it's easier to explain because it doesn't assume additional things. It's just, it has less assumption. It's more bear, like it's, no, no bear, like this bear, like bear, like B-A-R-E. B-A-R-E is the same word, no, bear, but no. I just spell bear. I think it's spelled differently, right? Bear, B-A-R-M-B-A, yeah, R-R-E. Yeah, yeah, yeah, yeah, yeah. Yeah, there are no priors. It's just, there is less, there is less to it, right? So it's just like less complicated, okay? All right, we have two minutes left. Let me show you the apartment with a bit of light without these all fancy things. And I will say goodbye, okay? Thank you for coming here. You've been 40 people at least, that was nice for this kind of improvised, impromptu goodbye to my apartment. I watched all your videos. Will there be new material in the book compared to your videos? Yes. So the thing is that I make, so one lesson takes me one week to make, right? And now I'm writing the back propagation chapter, which I just started teaching last year, right? And every time I go through something, I change it every time. So that's why every semester's things are slightly different. And now I'm actually writing things I didn't know, like that kind of tweet with that half circles. I think that was super cute and something I didn't know, right? Let me show you. This is kind of that bug me for so long. Like why on earth parameter sharing gets the accumulation of the gradients when you do back prop, right? And that was something I didn't understand. I knew kind of where I knew. I've been told, so I was repeating it, but it was not making sense. Like why is the case, right? And then I had to dig through and then it's like, oh yeah, obviously. Everything is obvious once you know it, right? But then at the beginning it was like, I don't know. It's blurry, it's foggy. And then I'm writing down. Yeah, the fact that when you write the book you have to be actually more not consistent but not only consistent, but also you have to be like, what's the word? Thorough, I guess, right? If I talk, I may forget something. But if I write down, I will write down everything, right? Like sometimes talking, you may forget a few small things, right? We've got your back whenever you need us. Thank you. I never imagined you would be having difficulty in career but yeah, major difficulty, right? I'm a temporary employee because I'm not a professor, standard professor, right? I'm not a researching professor so I don't get to go, I didn't go through the normal pathways. Are you familiar with the maximum code rate reduction from IMA? No, I don't know. I have to look it up, I have no idea. Maximum code rate reduction, no, I don't know. You're welcome. That's bear, okay? A friend just wanted to say thanks for all your teachings. They helped me immensely, that's sweet. Oh, you have actually understand? Yeah, yeah. Just want to reiterate if you need anyone to just chat, yeah, sure, it's okay, thanks. All right, let me show you the department. Like, let me show you a bit of lights and then let me say goodbye. Hold on, there's battery here. Yeah, boxes here. I'm gonna be playing, I guess, soon. Kitchen, it's pretty nice. But again, I cannot go there because the cable, there's a cable connected to my, for you to see what's happening. Vincenzo and the things, the lighting, that's one of the lights they can see, right? Over here, the other light is on the table. Where is it? Can you see the thing over there, the blue one? Yeah, then there is this light over here, which is showing my face with some light. Otherwise, you don't see anything, right? So without this one, you wouldn't see my face. And then outside is pretty nice, but you cannot see anything. There's a balcony and there's a very nice yard. And that's it. So now I can finally pack. See, now without the, now you can see me because there is the light on. Actually, everything is burnt, right? You can see there is all white here because the camera is configured to work with the other and blurry tool. The camera is supposed to work with this light, right? You can see also there is light and dark on my face. If I turn on this one, then it's even more bright. Hold on, I have to close this one, right? So there's a lot of setup you have to get right. Now it is correct, right? There is no burnt version, okay? Although the background is burnt, right? Because the, I have manual exposure on the camera. So if you don't have manual exposure, then you're gonna get like different colors based on how bright your face is, right? So if it is automatic aperture and exposure, if I go closer or back, like closer or further away from the light here, you're gonna be having the auto gain of the camera trying to fix things. Then, yeah, you need a microphone, proper microphone, proper lightning, proper, you know, decoration, right? Such that you can still be fooling around. That's it. Let me see what you've written here. Let me see. Okay, thank you. All right, tata, hi, friend, I just don't know, okay? Oh, you're actually not? Okay, all right, all right, sweet. All right, so take care. I'll see you in the next apartment, I guess, okay? Bye-bye. And stream.