Fireside Chat with Dr. Fei-Fei Li & Anthony Goldboom | Kaggle

okay we're gonna get started so welcome everyone to career Kahn's fireside chat with dr. faith daily and today's chat is moderated by kaga CEO Anthony Goldblum who's in the room with me dr. Bailey is currently the chief scientist at Google cloud AI she was also a professor of computer science at Stanford University and the director of the Stanford artificial intelligence lab she's authored nearly 200 scientific articles among her best-known work is the imagenet project which has revolutionized the field of large-scale visual recognition and deep learning so without further ado I'm going to hand it over to dr. Bailey and Anthony Kalume all right thanks very much Anna and thanks everyone for joining so it's a great honor to be interviewing Feifei today so a lot of a lot of you who are tuning in a relatively new to the field and and so you know may not be aware of the background here but Feifei is one of a very small number of people who powered the new techniques that are behind the current AI revolution and safeiy is most famous for creating imagenet which is a huge data set made available to researchers and really allowed researchers to start trying deepen your networks and also gave a standard benchmark that that research because I could use to measure their performance against other researchers and so imagenet directly relate to this covery of deep learning which is caused a big step change in machine learning and what machine learning is capable of so it's very exciting to be speaking to Fei Fei today because without faith phase work many of us wouldn't be working in this exciting new field now the conference this conference is of course dedicated to learning more about how to get started in their data science and machine learning career and Fei Fei has an unbelievable career story she was an immigrant from China didn't speak a word of English when she came to the US as 16 as I understand it a phase first job in the US was actually cleaning houses and now she works right up at the cutting edge of machine learning and AI so without further ado I want to get into learning a bit more about a phase career getting some advice from Feifei on this exciting new field and then we'll take some questions from the community safe a thanks very much for speaking to me today now this is a pleasure Anthony first of all I'm a huge fan of kegels work so when you Anna and Kat go team you invited me I thought this is gonna be fun and just always a pleasure to talk to you Anthony yeah likewise so let's start at the very beginning you were born in Chengdu in China can you tell us a little bit about your childhood and your early school experience I was actually worried amazing and and then I grew up in chendo China it's just a small city of more than 10 million people it was a you know I grew up in the 80s in China when China was opening the door to the to the west and I was lucky to have a pair of parents who loved learning my my dad loved nature and my mom is very much into literature so I had a good good childhood in terms of you know growing up with books and and I moved to the state's when I was 16 which life did changed dramatically since then I can imagine I'm in your early schooling what was your favorite subject that was an interesting question so the standard answer wouldn't be math I don't recall it was math so I think most memorable subject was physics but that was around you know middle school probably before that was Chinese actually I loved language and and but physics was really what caught my imagination and it was it was just these deep questions of where we come from or what is the universe where we're going that that was fun for me and and and so you discovered physics at school in China then your family as you said moved to the US and when you were 16 what was the trigger that made the made family move that is a great question that we have to interview my mom for that I don't know I was too young to fully comprehend what's going on it was you know I moved with my parents and you when you arrived do it you had pretty limited English and in two years you managed to go from you know learning a new culture learning a new language to join almost a full scholarship at Princeton and can you talk about that like what was the process of adjusting to life in the u.s. and and high school in the US so I landed in pursue pretty New Jersey I don't know if there's anyone out there dialing from perséphone New Jersey I heard it's cold over there today and it was a small town actually it was a ton of some already during the the mid-90s a town of immigrants and I went to participate in high school I was very lucky that the teacher it's a modest high school public high school but the teachers are phenomenally kind I had I remember my math teacher Bob Sabella he we kind of ran out of course for me kind of early and he used this lunch break to do advanced AP calculus for me and you know our teachers science teachers they were just kind and and I you know it was a time that I was focused on learning in a way I feel I was lucky because I didn't have much in distraction there was no internet for me so I was I was able to focus on learning learning the language learning some of the new subjects and also surviving as an immigrant enhance various sorts of jobs and and I think I was also lucky Princeton gaming an offer I don't recall having a phenomena SAT score score really I remember my verbal score was pretty bad but luckily Princeton still took me and I suspect they don't regret it us we'll see so you studied undergraduate physics at Princeton why you mentioned that early in life you the first object you fell in love with was physics what was it about physics that grabbed you yeah that's a good question um it's the cosmic wonder that grabs me I think early on as a child there's part of me that was precocious that I loved the philosophical ponderings of really the the the universe meaning of life were we going or where did we come from and physics to me is the epitome of the signs that touches on these cosmic wonders and Princeton of all places is where Einstein was it's Fineman was there and these you know deep questions it was just fascinating and I was lucky enough to you know to be good at math I guess and and it was a mecca for me and but it was also the interesting that I kept a leg in engineering I got a Princeton doesn't have minors they call it certificates but it's just like minors I did get a minor in engineering physics because that gummy started to be exposed to engineering and computer science so you did some software as part of the engineering yeah um so after you finished at Princeton you you probably liked a lot of Princeton students you had companies like McKenzie and Goldman Sachs very elite employers who were interested in you and and as a new immigrant this would have been a way to establish yourself very quickly I assumed turning those those off as Dan was a hard decision what was it a hard decision and what was behind my decision it wasn't hard actually Anthony I think part of it is I I think that I'm lucky that I didn't I kind of this since early on just pursue passion more than um some more Tikku concerns so on one hand I think it's important to fulfill my responsibilities and some you were mentioning I did run a dry cleaner shop at that time actually I opened the dry cleaner shop you're your CEO of cago I was CEO of a dry cleaner and that was helpful for surviving as a family with my parents so that was part of my responsibility that I I had to you know it was important for me but in the meantime I want to pursue dreams as well so so turning down Wall Street was not hard it was actually harder to switch out of physics and pursue AI and neuroscience because I loved physics but one thing that got me thinking was that as I not only studied that the mechanics electrical you know all the electrodynamics and all those quantum physics I also read essays by physicists like Schrodinger like Einstein like Roger Penrose and it was curious that for these giants in physics towards the end of their lives they ponder about life more than the universe they talk about life more than the universe so that got me fascinated by another deeply cosmic question is the question of life and and then you know being influenced by philosophers including people like de Carr you know I think therefore I am to me the the essence of life has a lot to do with intelligence and so I become very fascinated by neuroscience and I also went to Berkeley during a sophomore year intern to do a neuro science experiment and replicated Hubel and Wiesel and those neuroscience experiment and that really got me a path to to want to know more brain and making artificial intelligence systems so I feel like we've we've got our first life lesson here which is that sometimes in a long term you can be better off following your passion and following this short-term most lucrative it's a nice nugget so it sounds like you had to have some exposure to neuroscience was Berkeley your first exposure to neuroscience oh I heard you had you've done some of it at Princeton where did you you know when you went to Caltech you did AI in computational neuroscience as he said but where did the interest first come from yes so I'm actually as a college student so so if there's any college student out there use her summers wisely I my my freshman summer was a molecular biology internship in brothers University in New Jersey and I got exposed to you know genomics and those it was fascinating it's part of life science as well but um you know just like I am very bad at cooking today till this day I'm very bad of bad at hands-on the experiment so I learned that wasn't my strength so so the the second summer I explored more system neuroscience which is less molecular experiments and I was fascinated by that but I also learned you know it there is these stages of introspection that I I want to be doing things that the cross-section of computer science mathematical science and intelligence not just purely on the biology side so so that got me to decide to pursue artificial intelligence and cognitive neuroscience at Caltech got it well very good and I think there's another good piece of advice in there which is use your summer wisely right and how often do you get a chance to experiment with different career paths and try new things out you can do it for free during summer vacations so when did you start working on imagenet and where did that idea come from that's image that came post PhD time so so my PhD time was in in today's world prehistoric time of you know AI in computer vision even though it was just I started 2000 so it was 18 years ago and already we were asking some of the key questions in AI and computer vision in the in the subfield which is you know how do we recognize objects and describe objects today you open the Google photo app it tags objects for you but at that time that technology not only wasn't there that problem wasn't even formulated well and it was also a lucky time that machine learning was starting to you know to take home form as a subfield of a AI and the marriage between machinery and computer vision became very powerful so I was among the first generation of PhD students who was able to use the modern tool of machine learning and explored questions that are hard to answer like getting a computer to see objects so that was all good and I became an assistant professor and first job was University of Illinois urbana-champaign but quickly my alma mater Princeton called me back not to the physics department but a few buildings down to the cs department and it was funny because that was the time I was deeply thinking about machine learning modeling about object recognition and I was doing a campus visit I didn't even start my Princeton job yet I was doing a campus visit and I talked to a linguist a cognitive scientist Kristin crostini album and she showed a piece of work called bird net by another linguist called George Miller and that was a knowledge ontology that organized the entire English language and it was a cognitive science work and I had some training cognitive science to really appreciate and admire that piece of work and then she was saying you know it'd be nice to show a picture next to each of these words in English and that moment really was a bit of a epiphany moment for me I was thinking sure one picture doesn't solve the problem the world is big the world is continuously there's a lot of data babies get too exposed to a lot of data and a continuous seeing of the world and machine learning algorithm at that time was struggling with you know as always generalization issues because we don't have the right amount of data or our parameters are being over fitted and and everybody's trying to think about clever tricks to to get over this overfitting problem to try to generalize and suddenly I thought well if we can just rethink about computer vision rethink about object recognition from the point of view of giving huge amount of data to machine learning algorithm that would really open a new way of thinking about how to model our visual world how to represent our visual world so we started because of wordnet we call our project image net and that was end of 2006 beginning of 2007 we began that crazy project wanting to wanting to collect a huge data set of images for each concept in word that and there are tens of thousands of concepts in word net and people did think it was crazy right and my understanding is that the computer science discipline was quite hostile to the idea that someone could spend their research time collecting images and classifying images is that correct well I think that people who really thought it was crazy was my graduate students because they're the ones with me suffering day to day on that we had several failures and the first time we wanted to do it we wanted to do it about hiring undergrads at Princeton and they clearly had better things to do then to do Dino idling yes and it was just not scalable because we were thinking about we downloaded billions and billions of images and then we thought of using current machine learning algorithm and we spend a month after month developing very fancy machine learning algorithm some of the some of the milestone even became research papers themselves but at the end of the day it was philosophically the wrong thing to do because if we use machines to collect data for us then the data best-case scenario was the what the machines could do at that time but what we wanted to do was to improve machine learning and to give a benchmark gold standards for permission learning algorithm so we and one day we heard literally in the hallway again from a master student who actually just graduated from Stanford came to Princeton and he said have you heard of Amazon Mechanical Turk and I was like no what is it and then he showed me I still remember it was a project of naming colors a very very early crowdsourcing project it wasn't nothing to do with the image net and I saw that interface user interface of Amazon Mechanical Turk online market and that day we knew we found the path to image that and you were the biggest customer of Mechanical Turk for a period right I know I want to um so I'm gonna skip ahead a little bit and because we're running out of time and I want to get to the advice section but from Princeton you went to Stanford and now you split your time between Stanford and Google cloud and how do you split your time what's the you've got one foot in academia and one foot in industry what does that split look like well I'm on sabbatical from Stanford and I spend four days a week at Google cloud working with amazing people like you Anthony and I I still spend a day at Stanford especially focusing on my students and their research so that segues nicely into a lot of people in our audience will be thinking a lot about academia versus industry and you have experience at the top of academia and at the top of Industry and do you have advice for people on who are trying to decide between the two is there a certain character is there a certain personality type that might be more suited to being you're more suited towards industry I absolutely don't think there is a certain personality type we have a whole wide range of people and wonderful creative people on both sides I think that first of all I think it's worth exploring for especially for young people out there it's hard to imagine your first job your first you know degree your first anything is gonna determine the rest of your life right so give yourself that freedom to explore so why would you want to get a PhD in academia there's several reasons there are people who want to be professors and I probably belong to that camp I saw myself as a professor so that was easy for me but another part of me and for many of my students is that you want to dive deep because PhD is about devoting four to five to six years going extremely deep and expand human knowledge through a place that no one else has has take has done that and you have to do something original challenging and give them give back to the world something new in terms of knowledge discovery or a new algorithm or something but it's also for those who I spent that time to go so deep in in for five years Industrial another hand I think it's equally awesome here at Google cloud just coming to work every day thinking about we are democratizing AI we are taking the biggest computing platform humanity has ever invented which is cloud and delivers such a powerful technology AI onto this platform so that I can reach to our hundreds of thousands of customers and they in turn reach to billions of users that is also very empowering and being part of that is also empowering so there there are all kinds of different choices for academia versus versus industry one and one of the very popular questions we asked our community what questions they wanted to ask you the very popular questions was do you feel like a pH if you're if your ultimate goal is to get a job at an elite employer in a Google cloud AI or Google brain or deep mine do you think a PhD is necessary or an important prerequisite or are there other paths into those elite employers there's absolutely other paths I think I mean Anthony by being a successful entrepreneur are you good that's also another path and meaning of the many of the amazing people I meet that Google doesn't have a PhD I think AI especially in AI R&D a PhD is a very good path because AI is a very deep technology and there's a lot of new things to do and Stanford and other universities has phenomenal programs so I think in AI R&D and especially research part it is a very I think it's a very fun you know path for for young people and I agree my experience of being at Google you were an edit the curve he focused on computer vision at a time where as he said in your New York Times up here we could barely detect shopping so somebody starting their career out today what is what are the things that are ahead of the curve today what should be people be working on if they want to be the next face a way well it's very hard to predict the future I when I wanted to do computer vision I was thinking to be ahead of the curve I was thinking about two things one is I'm fascinated by this topic of intelligence and making machines intelligent and and I think also I believed in some of the potential impact of intelligent machines to the world so in a way I think for young people choosing their career thinking ahead about impact probably is a good important important ingredient I'm sure when you start a candle you are thinking about the impact a successful company like ago would have to to you know to the data scientist community and democratizing AI so so that was part of the consideration yeah the the advice to follow you follow your passion really resonated with me and last question for me and this is a bit of a different take on the careers question but in your recent New York Times op-ed I think it was published last week and you talked about human centered AI and you referenced that the the looming threat of job job displacement caused by AI do you have any guidance for people entering the field you know how do we make sure that we harness the good bits of AI without causing large-scale social upheaval yes so that's a very important topic for all of us as I said and I'm quoting quoting contemporary philosopher Shannon Baylor on this there is no independent machine values machine values are human values humanity has never created a technology that resemble ourselves as much as a I and as we develop a I know it'll be more and more so so I think it's is our opportunity to use AI in the best way possible it's also our responsibility to consider the social and humanistic impact of AI job displacement is one of them fairness safety privacy and many other issues are critical to AI so so from this point of view I actually think AI has entered an age that we need a much wider set of talents not just software engineers who can do AI algorithms we need AI policy thinkers we need social scientists we need humanists we need law legal scholars when the ethicist s-- to work on AI from all different aspects okay thanks vici and very last question this comes from our community I'm willing to do anything to be faithful to PhD student what does Fife I look for in a candidate to have to be her PhD student so first of all there's no favourites PhD student there's ten first computer science departments PhD student and your application has to go through a very rigorous committee multi stage multi professor review and I don't know if I can speak on behalf of stem for computer science department but we definitely look for students who are authentically passionate about the computer science area you want to study who are good students in AI good math skills matter good coding skill matter research experience demonstrated research experience matter you know being a cowgirl winner is definitely a plus so so some of the criterias are spelled out on the Stanford career science website Thank You Phi Phi this is dr. Anthony

Channel: Kaggle

Views: 3,608

Rating: 4.9534883 out of 5

Keywords: Kaggle, Kaggel, coffee chat, live-coding, live, learn, api, cli, python, data, data science, interview, questions, transfer learning, coding, networks, programming, technology, tech, machine learning, AI, artificial intelligence, coders, programmers, help, tutorial, projects, 101, rstats, stats, statistics, what is kaggle, how to, github, developer, kernels, datasets, data visualization, deep learning, sql, challenge, competition, whitehat, code, lesson, CS

Id: ElDWanLOc4c

Length: 28min 40sec (1720 seconds)

Published: Tue Mar 20 2018