raw-transcript-cristian.md

Okay. So, Welcome, Christian. And could you start by saying a little bit about? And who you are and wait from. My name is Christian Botner, I'm a senior researcher at Microsoft and I'm originally from Romania, and also graduated from Manchester in 2018 18.

So five, probably five years ago. Yeah. Okay. Good. Thank you. So, what I wanted to do today is is capture a little bit of A your story from. Starting studying peer signs even longer ago, eight, eight years ago now, and then your journey to where you are now in a short space of time.

So can we start with Right. Beginning, what was it that made you want to come and spend three years at university studying computer science. What was it was there with their particular thing that made that Made you want to study computing, a university? Yeah, I mean, I still remember when actually John Shapiro interviewed me for, for Manchester and then I feel remembered, And he asked me like, okay why computer science then?

I remember answering him that, You have the opportunity to kind of be multiple persons in some sense. So I was being a journalist and I thought you know, okay you could do many things with a computer like, you could be a biologist. If you really like, you know, you can do computational biology You can mathematics like all kinds of mathematics.

I don't know where the computer, right? You simulate things. You have all this. Um all these things like simulations that are kind of not analytically trackable. So pretty much anyone works. But with a computer, you know, in any field and And at the end of the day, you know, you could be a problem.

So you can do algorithms and all these kind of things. So so I really was kind of for me, the easiest way to, to do a bit of everything, really, but have the chance that do something, you know, a bit of everything later on once. Once I finished the degree, I guess you'd study didn't high school so you'd had the reasonable amount of exposure to it in high school in terms of, You know, not just theory, but a bit of practise stuff as well, but their engineering as well.

Yeah, yeah. I was quite deep into kind of algorithms and we are competitive programming, right? They many people coming from Romania, a little relate here. So like Olympiads, that sort of thing. Yeah, both mathematics and also kind of yeah informatics and I think lots of algorithms in general and I really like that and solving problems with using computers and so on Good.

Okay, So you're right here in Manchester eight years ago and You've got to the end of your first year. I think we, you looking for you, do a summer internship in your first year, or is it something happened a bit later? Yeah, I sign in my first year, I after my first thing Internet Amazon in Scotland, Um, this was quite funny because i, i knew I wanted to do an internship.

I like all the applications were kind of, we don't accept first year students. Yeah. And including the Amazon one, the explicitly said that and I was like okay I'm gonna apply anyway and see what happens. So I did and then you know they interviewed me and that went pretty well and then they made me an offer And yeah later that summer I found out that you know they got some applications from first years and it is okay maybe let's part of this.

See how it goes, right? So you know you never you never know where to expect You know, something opportunity to our eyes. I think, you know, I think something I learn from there was just, you know, try your luck anyway. Even if you know, kind of break the rules and because they're kind of arbitrary after That's good advice actually.

So I think I say is as a problem, I see a lot of first years, that question I get as most is like, I can't find an internship because it all says like, A penultimate year only but actually, you know, sometimes there's no harm in applying anyway. And you know, you might in your case, you've got lucky I guess with Amazon.

So that was Amazon in Edinburgh was? Yeah. And what was what was that doing in Edinboro you? Yeah. There were I think the team I was in there or having like kind of recommendation engine for Amazon data kind of recommending you Products you might want to buy and things that which is, you know, how they make a lot of money and Yeah, I was in this time, it was crying to think it also the machine learning which, you know, what's kind of I guess less height than now, it's going to a bit more daily days of the learning and so and so let's quite cool to see, you know, how they were actually applying it that's scale.

And then all these things Um, and i think i also learn a lot, you know, just in terms of software engineering, I realised later on when I was doing the software engineering course, in the second year, I think Holy little idea. Like i had no idea what an integration.

Why you need this was and the fingernails could just kind of asking around, I think. Yeah. I was, I was maybe a bit scared last questions. That summer maybe also because of how little I need you. But, you know, I think since then I learnt, it's actually severity important to the United.

Just kind of be relentless with asking questions, even even if you're annoying, Yeah, she got, I've got trying to find a picture here. There's a picture of you And say if I can find it. In. I think it it might have been in your first year when you did the that was a BBC Barkley.

I had the university challenges, it was annual first year was, yeah, I think I was the first year for agreement. So you were part of a team as I'm trying to find pictures here. And your part of a team that That there is so that would have been about 20 15 I think.

Yeah I've been in your first year was winter spring maybe and a new did a you did a challenge at the BBC where you had to propose I don't think you liked it wasn't much engineering involve was it it's more of a proposition that you yeah. Yeah. I think kind of making yeah some sort of these design for an app.

I think for Westbury to summer Olympics. So yeah, some things some something probably related to iPlayer as well. Also something like that. Yeah so that was 2015 with With the Those people sound on the bench there And so that was in first year, you did your, you did your Internship.

But Amazon came back about your studying software engineering. I think you did a--. I think to an issue on industry experience. Weren't you and did you transfer? Yeah to, yeah. Doing you figured right? Like if I can get good experience in there? Up in. In the summer, there's who don't need to do a whole.

Yeah, exactly right here. Which is, which is pretty good. Good way to get. So did another internship? At the end of second year, as well as all right. Yeah. So next time I interviewed with improbable, Which is this well at that time, it was like relatively small startup, or maybe medium sized.

Let's say. In Londoning large skill simulations. Um, yeah, there was quite interesting. Also, I think the kind of interview Stage for that summer. I think that was quite difficult. I think after the summer and Amazon maybe I got a bit arrogant. I was like, okay, it's gonna, he's gonna be easy, you know, kind of, you know, you kind of rely on your past success and your assume is going to happen again.

And I interview like, you know, the big tech including Amazon again and even go well, Um and yeah i was kind of yeah getting quite annoyed how things went and you know then I can realise, maybe I should have preferred better and all these things. I think that's kind of a bit over confident after the first summer.

And was it the sort of technically interviews or behavioural interviews that you are struggling with or Yeah, I think in I think maybe a bit of both like I think is just you know I didn't treat it seriously enough, the preparation because I just assume the same as last year, will just happen.

You know? I mean I did prepare also the year before but I think not as well. The next the next year and I think, you know, you don't need to at least solve a bit of competitive. Programming problems to get back in the shape and all that stuff. And you know kind of being that nice enough hard to import this libraries quickly is this dictionary is a lot of you need to work with in you know, and so on.

Um so you need to get these things, you know, kind of quickly implemented. And I think I was a bit of a rusting in terms of that. Um, yeah, so yeah I think I was I think I must have interviewed like more than 30 companies or something else. I said, I have an internship.

And then, you know, I started kind of, you know, Getting more serious about it and you know, preparing better and and then I had these interviews within probable which I think went really well, was also some of the hardest. I had like, well, their engineers but like an ex-goodlers, it was very much like a good girl or culture you could see from from just visiting the offices and I think of some of the toughest interviews.

I had like almost an entire day of interviews. I just deadline and I finished But they went extremely extremely well and actually they just called me on my way back from the interview that they want to make me an offer. And I kind of felt it's going to happen now.

And yeah, I think it was really, really cool. Up, you know, also the kind of experience the startup life. I think it was very different from Amazon, of course. And i i would say like it more just because you know I mean just they're not just the textile but also kind of all this Leadership business things and you know it wasn't interesting to witness the kind of decisions they have to make that, you know, you don't have visibility into in the company of the scale of Amazon.

And then next deal with them again, so I got the returning offer. Um, and yeah, let's also shocking to see how much the company go as well in the span of one year. It was much, much bigger company. The next year from the year before and I think, I don't know.

A number of employees must have doubled or something early if they felt like so yeah, it's kind of going to that was after graduate that You just end of? Yeah, graduated but Before that you did. You know, project with John, I mean, mentioned John. So you did Third year project, I think did you propose your own project or is it something you had an idea of something you wanted to do?

A new. A machine learning and you propose the project to John, or was it one that John would proposed? Yeah, I proposed I propose my own project. I can't, I think I had some original proposed, I can't remember exactly what it was. I think. I think probably I've done that.

Before the 30th I might like before the summer. Yeah, because I think that's when you have to do it. But also I wasn't, you know, I guess I was busy without a courses so I not she was the best proposal. But anyways, John, John was happy with it so he took me Um and then i think over the summer, you know, I started reading a bit more just to figure out what exactly I'm going to the next year, in terms of the project.

And then I think I grew more interested in other kinds of things. And yeah, I talked to John and I think I switch my project to like this text to image synthesis. Right project and using. Against generative. Adversar and networks. These were still kind of the the early days of deep learning so to speak.

I mean maybe not quite the early days, depends because I guess some people started three years earlier. But like it felt like I think definitely not as high as it was now, you know, and Um, and actually use my, my money from the probable internship to buy like an expensive GPU, because I knew you're not going to get any from the university.

So, I just had this this computer just for that project and And yeah. I mean it was just one GPU, right? Yeah. Talking. Like lots of GP. Yeah. I just want I was a pretty powerful. I think that's kind of one of the best you could buy at the time video.

Yeah. Yeah. I think it's like a 1080 Ti actually. I'm trying to sell it right. It's been kind of staying in my office in Cambridge, the Unused for a while. Yeah, so you actually got back. I think got paper out, I didn't you get some things. He had something in archive or something.

Yeah, but it up on archive. Um, yeah, i think yeah, that's pretty much state of the art I think of at the time. I think unfortunately like yeah, I think I could have You know that a bit more and maybe publishing a workshop. But yeah, even a conference. But I was so unexperienced that I had no idea about these options.

Um, so that's maybe something out of regrets slightly because it's just an icon and you know, I didn't do any effort to publish it but you're not to know at that stage. I mean one of the questions you the talk you would just doing earlier. Students were asking this question about publishing and it's it's quite difficult at the beginning because yeah you don't really know what your sposed to published and how you publish and where you publish, who you can publish where they're not this kind of stuff and so it's not it's not You know, in hindsight you can see executive done a little bit more work and got something a bit more at that but I was impressively, got anything out to be, right?

Because a lot, you know, getting something out and undergraduate projects, he's good. Yeah, so you did your your graduated, this is 28teen, you've done this internship. Going back in probable. And you, i think by that time, you decided you want to do, PhD, So what made you think? I mean, was it, was it kind of Doing research stuff that made you think actually really enjoy the research.

I'd thinks I'd like to go and pond to a PhD. Yeah, I mean I think this project I did really got me into I mean, you know, I know I loved machine learning and Also, as part of the projects can just reading lots of books on my own and paper and I think also for the same reasons I really like computer science, you could do A lot of everything in machine learning also more advanced mathematics as well.

Also, going to the Mathematics. So it felt like kind of a nice way to combine all these things. I like even more Um yeah also again i got these undergraded words price for that. For that 30 year project so that kind of incares me as well as it wasn't a good track.

You know it wasn't good work. Yeah, so then i, i went to Cambridge for for an Enfield. Um, in advanced computer science. So that point a dude, so you this is kind of like a So that I understand that where the M Phil program works is that you can kind of if you decide after yet actually.

I don't want to do research. You can leave at that point and go and work somewhere. You've got an M4, you've got a qualification or if you are, you know, if the PHDs or the research is going, well, you can carry on and turn. Actually do that. A turn into a PhD or carry on in the same, very hard to do a PhD.

So you'd actually did a You have you did a proposal for your m fill project. Is that, right? So part the application process is Picture your idea of what you'd like to do your research on. So was that that was machine learning related. Yeah. Appointment. Thanks. And you picked a supervisor.

So, rather than just writing a generic. Yeah. Pose or you said, right? I'm interested in The person who, Became your PhD supervisor, I think was it always at somebody? Yeah, so I think actually I haven't there was about that was recommending. The students are like that was the mistake.

I did actually didn't do that. I can't remember back to what my proposal was but I'm pretty sure I was pretty general. Like I didn't do any research. As far as I remembering to like potential advisors, I just kind of something that came to me and I don't know what it was.

Um, and of course, i wasn't necessarily paired with the right advisor because of that. But the kind of during the year, Once i was accepted, I I switched to catch earlier who's been my amphi advisor NPHD advisor. Later on. And yeah, I think we were getting along very well.

He also gave me a lot of freedom so so yeah. I think, yeah, I kind of changed my my direction later. And I think you always have that freedom with this proposal. So, you know, you're no one kind of constraints you to the exactly what you said you're gonna do.

But it's a pretty good filter for you're gonna be matched to. So, you know, at least in terms of broad interests, you would be useful. You know, if you actually match with the right advisor Yeah. So So, the MPL you do you is it's one near them for and then it's another two years for a PhD, or is it another three years, three years.

So you do one year of Enfield and then three years. Yeah, PhD after the right. Okay? So then you start, you start your PhD after that. So what was, what was your PhD in? What was the sale a little bit about? What your PhD was in? Yeah. So again, I think, yeah, looking backwards.

Can, I'm amazed how, you know, these organised? Like I said, last year. Did you say it's 2022? Is a day on your thesis is that? Yeah. Yeah, well I submitted it in, December 2022, and the five hours in February this year and official. I got you asking questions. Thank you.

Yeah, and official I got it like a month ago, right? The actual declare, right? Um, but i like looking backwards I think even with the PhD proposal. I think I just Yeah, I think I just proposed the some sort of extension of my master project just because it was easier to write something on you know about.

But it doesn't like that. But I mean, it was published, there was my master project was published that Triple AI, and it happened in New York. That was kind of the first conference. Like, and the last year was right? I think COVID hit and there was no conferences. Well, was I think I was like, right in favourite before COVID or something.

So I was great. So you got, I mean, you got first taste of going to a conference. Exactly. You're only good conference to go to. Yeah, yeah, huge conferences as well. Yeah. Yeah, so I think I propose an extension of what I did in the masters but again during that summer I think I was I was at google accent California's and the PhD and you know ours working on like machine learning for robotics I think was one of the best summers of my life.

Like you just deploy, your machine learning model on an arm form of robots around your knowing that they're running your softer. It was kind of amazing to you know to witness and I just watching them all day. How they pick up things trying to spot differences compared to previous models, you know, is doing that picking this subject then another object So that was pretty amazing.

But i think he also kind of made me realise. Maybe I should you know, do something else for the PhD and what I propose the regionally, I mean I'm not sure if I took my own purpose, a very seriously anyways because I think I knew I'm going to be accepting the in the program.

Because I was quite highly reacting the Enfield. Um, yeah, so and I think also, that's somewhere in my spirit time, I was watching this Course on geometry and topology. From. This mathematician Tadashi Tokyara. He was a professor in Cambridge for a while but I think he left before I arrived there and I think he's in US somewhere.

I think maybe I think might be a Princeton or something like that. And it has like this amazing youtube courses on geometry Anthropology and he's not just kind of, you know, like A good teacher. I also going to get you excited about the subject. Not precisely what Not necessary just about what precisely he's teaching and the way to think about mathematics.

And so and so that got to be very passionate and you know, then Pietro's group was just doing lots of graphing on networks. Um and this was kind of coming back track a little basic, Can you tell us what, what a graph neural network is? I mean, I guess most people listening this hope will know what neural network is.

But how is a graph neural network different from Other kinds of neural networks of people might be familiar with. Yeah, I think there's days people. See, that's part of this bigger field of geometric did learning. And is going to wear this connection with geometry and topology came in. Um, so in some sense you can have data living on all sorts of geometric objects like surfaces, like, if you're entry graphics or something that you have lots of data kind of living on, I don't know.

Also animated things or things like that meshes But also on grass, in particular, for instance could be, could be one type of geometric object where you can, we can have data on, and it's quite common, like they have a social network, Facebook is then all sorts of machine learning on grass because all day or data is based on graphs.

And, and so on, so, Yeah, basically it's a model of a processing data that leaves on graphs and all sorts of statistical tasks. Involving graphs. Um, yeah, and there's also kind of you can also see As some form of discrete geometry in some sense, like you, if you see the graph as a discretized, the geometric space, if you want.

And then this gives you many connections with geometry and topology, and I think that was kind of my unique perspective into this. And I think at the time, maybe not many people were sharing this, this perspective of you bringing some of the sort of more theoretical aspects of topology in geometry into what is quite an applied area of.

Yeah, computer science, isn't it? Yeah. Bringing those two parts together. Yeah, exactly. Yeah. Kind of, I was trying to connect the dots between the two, but, you know, ultimately also to kind of lead to some practical, advancements for that, kind of problems. People are defined just in and, yeah.

And I think, you know, to someagram manage that, that like, you know, some of the work I published in the beginning, they were, for instance was, Fly to molecules, for instance, for predicting properties of molecules and offer instance, some pharmaceutical companies were looking at using this models. Um, yeah.

So Yeah, I think that's kind of mildly mildly successful and also kind of getting some of these stuffing to actually apply things that you know Latimately work. But at the same time I was also an opportunity to do quite a bit of theory. Get more deep into the mats and so, yeah, so good or not, a nice parts of both.

There's nothing at the you mentioned, Twitter court, was it Twitter cortex, right? So that that come after Google X was that before, Google. So I need to figure. So, yeah, I think my first summer, doing the PhD, I was, I went to Google brain. So, right, when I was a good building, Ex I work a lot with brain people.

There are kind of very close cooperation between the teams. Um, and that got me under their radar, so to speak and I knew people already and I wanted to kind of you know come back and work with them the next summer. So that's what I did. Unfortunately was also the first summer under COVID.

And yeah, i was kind of interesting. Sorry to be honest like I was supposed to go to California, you know. Yeah, and a tinier apartment and Oh my yeah. All my team was in obviously, in California. So there's like huge time zone difference. I had to work kind of more in the evening, right?

So yeah, that was kind of depressing because that's kind of like a very short time spend where we would actually overlap in terms of times and we could communicate And then at some point also had like an kind of, a third advice. I have to advisors from US from Google and the third advisor from Europe.

So it's kind of also harder to coordinated so many advisors. So, you know, you things got worse quite quickly. But yeah, we still publish the paper at the end of. I mean, I was not my best favourite, I would save it, you know. I think I turned out I think to something And I guess in the way, some kind of good to It's something you were saying earlier that you did these internships during our PhD and they you were lucky in that since you had Didn't have to be something with directly related to your PhD a bit like.

Undergraduate internships. You can just go off and do whatever you want. And but that was quite good because you were able to do different things throughout your PhD just to come up. I guess kind of right now I'm doing research but was related to So those those came about you were just applying for them while you were while you were doing your PhD basically.

Yeah, yeah. Okay. Um, so Um, we're now so you Graduated. Well, you graduated from the PhD. Last year, I guess you have you had the big graduation ceremony yet. Have you done that with a gown? And yeah, so that was a math ago. Yeah. So you didn't know you've done all that and then you You, you know, work at Microsoft research which is is based in Cambridge.

So that how long have you been at Microsoft? Yeah, as I've been there since Late might see. And you did continuing doing on. Obviously can't appreciate this things. You can't talk about what you've been doing, but continuing the stuff that you've been doing in your PhD, In in a, in a commercial environment.

Yeah, I think it is dangerously related. I mean, it's not what I was doing specifically. I think I was be quite hard because there's many obscure things. Yeah, the same extent I was doing so. Buddy. I like there's it's an yeah, I'm working on weather prediction, and simulation using machine learning and yeah, there's yeah, some of the best models out there in this space.

You know, the use graphing and networks describe, it of geometric, deep learning. Obviously, you're honest here. Um because you know, you could approximate idea as a sphere and then you know, you predict things on the series. So so yeah, there's also sorts of related aspects with what I do what I was doing before.

There's also these kind of differential equations that describe the evolution of weather. Um and yeah i was also doing some work on this neural differential equations how you can. There might be HD on how you can kind of approximate solutions to this equations using your networks. Um, so yeah, don't relax.

Quite a bit with some of these things. I've been exploring. So I guess it's one interesting thing here is I think, I mean, you mentioned because you mentioned the GPU earlier in terms of like access to hardware, right? So one one barrier that people face in trying to get into AI machine learning is that?

And you know, there's the theory part but then there's the if you actually training models, sometimes you need, depending on what they're doing. You need it, but you need access to quite A expensive compute. So, And as a student I guess you had access to good computational resources but for undergraduate so interested in getting in this kind of stuff, what would you recommend in terms of How they can get access to the hardware and And you know GPUs or TPUs or whatever it is that they need in order to train and work on these models.

Yeah. So yeah I think scares resources in the sense. What's going to also problem? And when I was doing my my third year project and I knew I'm going to be quite, you know, I'm going to work quite hard on this and nine year. You know it's going to be yeah something want to dedicate a lot of time to to the third year project.

So I decided to use my internship money from in probable, basically to buy a kind of a state of the art and video GPU that I could have and I could just use for whatever. I think the reason I went Might chose to that. I think I was doing some calculation.

Like, how much would I paid WS for kind of the same amount of computer? And I think every some crazy conclusion, I would just spend Those money like I'm maybe in a few months so maybe even less than I've been weeks. So in my total sense, you know if you're like I think was I think 1.5 k or something.

On the machine, I bought. Yeah, it might more sense to just buy a machine and you can use it entire year and I think, you know, even today like I know there's lots of research into life skill, machine learning and so on but I think you can do pretty well with one big good TPU and especially if you kind of choose your problem set and you know what you want to work on quite carefully.

Um, i think that, you know, you can do a lot of progress and Um, yeah, you just have to be a bit clever about, you know what? Of course you can't. You know, you can't produce GPT free or something, you know. But there's lots of value but research that can be done with just 1 GPU.

And you can publish papers at nearest one GPU. And I mean, even during my PhD, I did. That's sometimes. You know, I had just had A lot of access to just ones if you and, you know, just use it in smart ways and right, you know, so there the economics of that still stack up.

Now, it's saying that that's still stands. Now, it's sometimes it's the most economical way of doing it. So, the slightly scary thing about using these cloud services. Whether it's, you know, Googles or A Microsoft's year or Any others or Amazon's is that? The kind of the bill can get quite big right quickly.

Yeah, exactly. So at least, you know, Running it on a local machine that you've got, you know, you're not going to run into any huge bills. You invest in a bit hardware to style And now the GPUs I kind of also even better like price and you know, is dropping and you can get kind of The same power you could get like a few years ago for a lower price.

So now also kind of works through your advantage and because, you know, if you want to do like smaller scale, kind of Experiments. I don't think you need more compute now than you needed a few years ago, so yeah. But yeah, GPU has got cheaper, so I think probably that should work.

The ones advantage. Good. Okay. So Um, so Paula the reason you're here, manages you've done this talk on graph neural networks, so you've just given a nice introduction to grab your networks and some of the stuff that you doing at the moment. I guess you're saying, And you know, there's there's working progress there that you're in.

The point of, A publishing at the moment. But if we could take a step back and go back to With my questions to edit this bit out, but One second, right.

Right.

Yeah, so we talked about a little bit about your journey from Big undergraduate. Travelling passing through your and graduate degree doing PhD in now, starting with starting your career with Microsoft. And, I want you to imagine that. You know, you've studied in two universitys. It's so the university, Manchester, and the University Cambridge I'm now going to make you vice chancellor of either or both of those universities, as you wish.

And as somebody who spent a lot of time as a student, as an undergraduate student, and a PhD student and a master student as well. And what would you do if you were to be made by chance or in terms of making And university more. I don't know better from a student experience point of view.

Yeah I think something I found out all these years is that the best way to learn is to actually, The things and discover things on your own. So think of very nice way, would be to have some program that is going to teaching. But research based in the centre.

They have an opportunity to do some sort of research project. And I think there was like, even in Manchester I think I was reading. I think when I was another good was this. I don't know. Learning through research program or something like that. But I don't know how widespread it was.

Or, you know, at least I didn't, you know, I didn't affect me directly and I think, you know, you're my always have kind of opportunities like local contextualize opportunities to do. A little research on your own by working with someone, but it's not kind of institutionalised. Like, it's not, you know, someone that everyone can have access to.

Um so i think that could be quite useful to you know, actually got to the boundaries of what's known in in some particular sub area. Right? And maybe discover is actually not that hard right? And A year. I think that's going to the best way to to learn after all and just like kind of like being an intern in a research lab at the University.

As a, you know, like I mean you did internships in in commercial companies but the same thing. A perhaps earlier on in the degree because you don't really get a Chase for research until your final year. Yeah. Right. But you're saying it'd be nice if you could do that earlier on.

Exactly first. Yeah, second. Yeah. I mean, I'm I did that as an undergraduate and it's something you remember years later, but So more opportunities to do that kind of thing in turning research labs. Exactly in the university. Yeah. Before you graduate Exactly and not necessarily as part of an assessed piece of work.

Yeah. Yeah I guess that's kind of optional. Yeah, I think the main thing is the You know, actually experience that yourself And I guess in your case. I mean, a good experience was And because you wrote your proposal for a third year project and that probably helped a little bit then when you applying for Masters programs.

Yeah. They're asking you to write proposal is that you've already had some experience of that, you know, blank piece of paper problem of what am I going to do Now just coming up with a good idea. Yeah, yeah. And then you need to say, And then you can also show that you kind of have a track record.

Yeah. And experience of You know doing this you know what it means to do research, you can be quite challenging it at times, you know, especially when things are not working and Should have no idea why and things that right? So, Uh, yeah, i think i think can help a lot especially if you want to have a research career afterwards, but not necessarily like it is a very different kind of Experience from, you know, standard teaching where there's always going to the correct answer, you know, do anything.

You know what's the right answer to that is like everything's kind of right or wrong. Whereas in research, it's been, you know, You don't know, right? Yeah, no, no dark yet. So it's kind of very different way of thinking about things. So it's not it's not just a technical challenges, more of a sort of cultural challenge.

Exactly. Your being an undergraduate or being a student is, like doing what you told basically exactly following instructions and ticking boxes. And then all of a sudden, It's completely different doing research because you'll have to come up with the ideas yourself and come up with good ideas and ways to test them.

So, yeah. So when you're making that transition actually from being a student to being, you know, as a student who just studies to a student who does research what was sort of main challenges you faced in And in. Making that transition and how did you overcome them? But I think one is that.

You're search spaces here. I think that's that's one challenge. Up your one, human and your time is limited and you could do anything right. Like And that's one of the big differences when you have like a pretty good teaching curriculum. Someone puts this road ahead of you and you just walk.

Yeah, whereas here, now you You know, you have 100 avenues plus a few others that you're not seeing so it's not clear in what direction you should go. Um so i think that's that's you know, just kind of this huge conveniatorial search space. You want to explore and with limited information about what each Avenue will actually provide.

You I think that's the most challenging thing and I don't have a selection credit if that's your full love person. Oh yeah I think I like what you're saying earlier it's about you. So someone was asking you off the talk about exactly that's problem of like how do I robe proposal and that I thought your advice about Start from the research papers, right?

You know, you can do it bottom up and start with the textbooks and try and understand theory and building that way, but they'll take you forever. Yeah, so your advice to them, which I thought what was quite nice? Which was like, well, just read the latest research papers and you even if you don't understand it all, yeah, at least you'll be at the leading edge of wherever.

Yeah, whatever it is that you're interested in is at. So then you're you can, you can Right to better proposal come up with a better idea, then if you're just reading textbooks and stuff you've been told to read. Yeah, I think it could give you a good glimpse of the destination that road can kind of lead to write and Yeah you wouldn't know what to expect if you get there, you should, you know, produce research of this kind?

Does it actually excited here? And I think probably that's one of the best kind of filters. You know what, actually gets you excited from all these but potential things if you know, if you read about kind of latest research in an area and you know, and then probably not the right one for you, you know, but like I think that's what kind of got me to deep learning goes.

Also, these are powerly papers that came out at the time and also into kind of this text image synthesis project. I think I really like you would be surprised by the thing you created like you. It's all these kind of jarrative AI there was, you know, I was doing at a time like you don't know what your model will generate, right?

You'll be surprised by my own work and that was kind of what something that you know made me excited like right? Yes. I think that's kind of the team ingredient and probably the best filter to search through this place. No. Okay, good advice. Why? So find something you're interested in.

Yeah, rather than something I mean is a lot of stuff where when you see papers published it's like somebody else's interested in this? Yeah. And trying to make you interested in it and convince you that that you should care about it as much as they do. And sometimes you do by the times, you just think, like you're saying, Oh yeah, i don't know where he said not.

So interesting that Okay, good. So There's two more things to finish up then, so, The first one is one tune, one podcast, one book, one film. So I ask all guests to come on the show to recommend. A tune. A piece of music that you. That more added to our code, is playlist.

So is there a piece of music you'd like to recommend Up people to listen to his podcast, go and listen to, and say, a little bit about why that piece of music is important to you. Um yeah, maybe i'll go with Do I want and are from Arctic Monkeys because I'm trying to play it on kids high.

And is the only song I can have started to, you know, trying to learn recently. And that's kind of one of my first things I like, because I really like the song. So say I'm gonna go with that one. Good. Okay. And And do I want to know by the acting like satellite?

There's a video to that with the sound way we've seen. Yeah, at the same way, going down, that's quite nice video. And then one podcast I don't know if you listened to do listen to podcasts much or yeah the radio programs if not or maybe. Yeah. Actually way too cheap.

This is it only get my Spotify? There's a few podcasts, but I don't know the exact names of Let me see. Oh yeah, there's the Joe Walker podcast here. And I was that one episode was about, Ah, I think this guy from, I think it's from Australia, but you kind of Yeah, and I quite like invites very kind of diverse groups of people like historians and economists.

You know, machine learning people. So it's, yeah, it's like kind of a nice way to learn about all these kind of different areas. Yeah, Joe walk poker. Okay, good. And then one book, any book, could be a novel, could be something. Else. Well, i fiction. You know, I'm reading now, 0 to 1 by Peter Thiel.

It's about building startups. It's okay. According to interesting. Um, so yeah, I'll just go without one. So, Peter Tails, founded. And what balancery? I always find people before that as well. Yeah, I was a people before we go. Right? Okay, good. And and then that's when he's one film, Any any film that you recommend?

Favourite film. That's a harder one. No, there's tons of them but I'm gonna go with gravity right. Okay. Recently there's this 10 year anniversary in London at the British film Institute, right? Came for a QRA I was like an in the first row so I really enjoyed re-watching it and you get chance to ask him any questions or I did at the beast.

I was decide to understand anything anyway. But yeah, it's just going to enjoy the seeing him answering us this questions. But yeah, I was great movie so I'm gonna go with Ellen. Good. Okay. And then the last one is time traveller. So my colleague in the physics, departments invented a time travel machine which they assure me works.

I have lent us You can travel back in time. To meet yourself in the first year, arriving here, from Romania, eight years ago. And what advice would you offer? A younger Christian about. Making the most of the short time at university, because three years goes by very good, even eight years goes by a very quickly as well.

What advice would you want? Yeah, I think maybe. I think my timeside, you know, i realised, As we also kind of discuss develop a lot of things, I didn't know at various stages in on my journey and you would have been nice to know On and realistically, I don't think there would have been any other way other than just talking to more people.

Oh, maybe even outside kind of your immediate circle. Um, or maybe just by expanding your immediate circular even, you know, Because by definition, you can't think about things you don't think about, right? Like, The only other way to think of this things you're missing about the kind of unknown unknowns.

If it someone comes to you, right? And tells you about something, you don't know. Um and i think that probably something I should have done more. I mean I didn't involve a lot in you know like also organising hackathons and so and kind of met a lot of people but I think I could have done more of that maybe also in another areas just to kind of learn from from, you know, other people what are doing am I doing something wrong?

They, you know, it's something doing some I'm not doing and I should be doing and so on. And, and that kind of thing, you know, maybe finding men potential mentors, you know, maybe I'm talking to PhD students more something like that. For instance, I don't think I talked to many PSD students when I was here and, you know, I think I would have definitely learnt something from that.

I had qualo exposed to PA teaching for that realising because they would have been TA. Yeah, exactly exactly. Yeah. I guess it's hard to connect sometimes a people. Yeah, that's true. That's true. Yeah. And so, yeah, i think kind of all sorts of things that, you know, would kind of diminish the unknown and us.

Right. You know. Okay, good. Let's get advice. All right, so that's Of the last time my questions. So thank you. And end up pocket. Right. Good. I don't know how long, how long was that to be doing a bit short of them? I thought about half now that and I was a bit fluster because I didn't have I didn't have my script.

I'm just gonna messing around trying to do the script and in the middle. A good. Thank you. Yeah, there's always fun. Um, I was going to mention this picture as well because I was looking at this from Sammy's profile, but this is That would have been the graduated. I think, wasn't it?

Yeah, something still. You remember Bill? Yeah, So Bill Came back for. What is now? Great uni hack this year. It's he basically set up student here. Yeah. Maybe you might have been at the first student, hack. No. Possibly. He came back this year for 10 years afterwards to talk just students about 19, but no that must have been 2018, i guess probably yeah.

Something that I think so. Oh maybe no. I think it might be 2007. 17. Yeah, I think we were won the MLH. Traffic order. I can watch it probably says on the trophy doesn't mean. So yeah together, I can find out was just under date the picture. Right. Okay.

Smarter feminist faces, I'm still staying in touch with the Robert and he was in electrically. I don't know if you remember. He was not in CS but I just doing housing. Yeah, he's in Germany. Has the startup which is going quite well for trading printing in space. Was always interestingly printing and they want to build this 3D printed that can 3D print in space and it's very cool.

They want to launch it into orbit and things so it's kind of amazing. Yeah. And then me high, I think he's in this mess. Yeah, I think he's in Austria. No, I think for working for Qualcomm. Actually ego still here. Yeah. Okay. Sammy's Sammy. You know is that deeply?

Now he's a deep mind. It's he gonna submit his PhD. Well there's a question, I've run away team is like he's not sure. Right? You know, I guess he doesn't need it now, right? Yeah. God job at beat mines. Yeah. But yeah, although I'm always a most high managed to get in before kind of finalising.

Everything. Yeah, Sammy way. Yeah, he knows. He's dark magic for this. I think and raise it. Google is me. Yeah, he's still on Google. Yeah. I'm I ran i bonded to him in London, the next accidentally a few weeks ago. Janina. Because she was on the team. Yeah, other people from, she's actually think it was my like that.

Yeah, she's in Lana. She's working a booking.com. Oh, she switched recently, thank you. All right, and then she switched jobs recently. Yeah. Nice, good. Okay. Well cool, thank you for doing that. Yeah thanks for having me. Is there anything you want me to end out? Apart from my clumsy?

Nothing halfway through. I think it's good. Yeah. And, I just get to have right, it's good to have. A stories of failed applications. So talking about like 30 fail applications in. So when you already had experience It's useful, because A. And having stories like that is useful in teaching is to have examples where I can say rather than just saying, oh yeah, you should just keep trying.

You can say well, This Christian christian did this and this happened and having sinners like that. It's quite useful for teaching. I think I had like such a long list and I've come, I think I like, I made this Google lock, I think I'm 100 something. Companies in there, just kind of taking no one by one that's talking to my brother because the, you know, kind of give him the same advice, apply?

Anyways, even if they say, you second years or something. Well, again, that's good but I hadn't hadn't hadn't. No one's told me that before. I didn't know, I just assumed you did a few things by the book, but yeah. They can own the worst they can happen is that they reject you right?

Yeah, exactly. And fine. I mean I think it's such an arbitrary all anyways because like if you're a really good there is absolutely no reason. You know, they were sail, you're a second you're hook here is, you know, and there are companies that don't care. So there are there's a company called necraft that do so security stuff and they're like, we don't care what year you're in.

Yeah. All we care about is that you, you're interested in your capable. So that's what matters. Not what year study US? Yeah, I mean, to be honest, I think also Microsoft, I think not the people hiring are actually going to putting these rules on the job application. Because like, it's HR probably.

Yeah, exactly. Because we just have some. Okay, we're looking for this skills and it's not like I thought that I need a second year of 30 minutes. I just put it there and then when you read the job description, you see these extra thing. They try and then Um, and i i've done really care about them So, and at the end of the, I'm, you know, the person highly.

So yeah. So now I'm really kind of tears. If Yeah. Well, all right like this thing. Which, oh, this cube thing is. Oh yeah. So this is Hi, it's very nice. I got invited to a conference of a Googleplex in In. That main mountain view and they gave like a gift because it was a jointful thing between a Riley.

Yeah. Google and nature. And they pick the Buckyball one. They had several ones but I like the book people once but geometries called and I should go one of these on right here, Have yeah. Are you much stuff with chemistry? Because you're making you mentioned, you have proteins and you're talking a little bit about that is that kind of one of potential applications of yeah, you doing.

And so we have quite a few projects in our for science and Microsoft kind of looking in these areas. Oh yeah. And i think it's kind of one of the also for drug discovery, and it's just what they do. All they Um, you know, original, I was supposed to kind of spend half of my time in one of these projects on drug discovering the Microsoft, but then Plus the rear that I wish you welcome one thing.

Um, so then I kind of dedicate for time to these PDs and simulations space. Um, yeah, that's also quite interesting I think always Yeah. Also there's these isomorphic labs company now in London, I think it's kind of a deep mind it's been out, right? That Dennis is leading and the drug disco and it kind of apply all these alpha fold stuff now to Drug discovery.

And yeah, they do a lot of this proteins and molecules and so interesting stuff. Yeah. It's a very hot space and this plan of startups in I just stuck into some startup in Berlin doing proteins and things. Yeah. I was reading that, you know, that That there's that famous paper in there, is it attention?

You attention is all you need. Every author on that paper, Sometimes I startup or started the startup. They were all that. I guess a lot of them were at Google or yeah, it corporations. Yeah, people since God made their own starter she joins us to do stuff. Yeah. I mean it makes sense if you are like, kind of a hit paper, like back as there is not like they're gonna promote you rights right away in Google.

So, you know, just go out and make a lot of money. Yeah, I guess so. Yeah. It's right. Thanks a lot for having me. Yeah, you see coming And Up. Yeah, well I'll let you know when I publish it but it's quite might take me a It a month or two?

Well, what normally happens is. I published the audio first and then right here, Sharing it around, okay, all right. Okay. Thank you. Thanks for helping the password. By the way, I got my password recently, So it was right. Where would they have one? Yeah, so you have a whole ceremony and yeah, yeah, that was fun.

Yeah. Probably do you have to do some crazy quiz that you about about British values or something in that, is it? Oh yeah. There was like a citizenship test and like, yeah, I mean a history part was easy because I like this girl and every dollar and it's kind of more like the personalities and sports and okay, another line in my room but we, it's like that little bit, you know, for I was swimming and leather Olympic sports.

I had no idea. Play the think of it as moreization that I always quite easy one. Yeah. You probably know more about Britain than the most quickly. Yeah, I mean, I forgetting gradually not. So I'm gonna go back to the beginning in a few months. Right. Right. It's

Oh,

Cuts short for some reason?