Advice From a Top 1% Machine Learning Engineer

I've been getting a lot of questions about how do you become an AI engineer what is it like to work as an AI engineer so today I thought I am going to invite one of the best machine learning Engineers working at a big tech company here Mita and she's going to give us advice and tell us more about how she became an AI engineer hopefully so you guys can also learn and become an AI engineer too so Mita thank you for being here tell us a little bit about yourself hey everyone I'm Mita bararia I'm a senior resarch scientist Netflix prior to Netflix I was working in Etsy a two-sided marketplace where I was Tech leading the recommend assistance team I've have done my PhD with specialization on machine learning that's really impressive just the person that we need to talk to because we get so many questions on YouTube about people who are interested in Ai and what is this new thing and how do we do it you started very early on before AI really picked up so yeah tell us the secret like what sport do your interest in getting I did my undergrad in electrical engineering I love some of the subjects of electrical engineering but then towards the end of my undergrad I realized that I love ma mathematics and Computing and programming more I uh decided to take up a job as a software engineer after mandag grad while I was enjoying pure software engineering I also had this itch of knowing more about a field that's when I chose to come for my masters in the first semester I took a course on machine learning introduction to machine learning in the mathematics Department Department wow really sparked my interest and I wanted to know more about it and that's when I decided to go for my PhD in machine learning I really like that because one of the advice that I always give to students or people who are exploring is take a class you build project see if you even like it maybe you do maybe you don't and that's the best indicator whether or not you should pursue it absolutely you figure out whether you're into it because you can only be really good if you're really into it exactly if you absolutely hate it like don't it doesn't matter how popular it is absolutely you did your PhD program that's another question that we get a lot do you need a PhD in order to get into AI machine learning be a ml engineer or AI engineer if you're transitioning from a bu software engineering I think there are plenty of available resources whether it's courses and Cora or other online platforms whether is reading some of the seminal papers and understanding the fundamentals possible to understand the field without actually going to a formal graduate school right and then F get into a team which gives you an exposure in getting hands-on experience then you can slowly be a full-time ml engineer from a p software engineering so I think it's definitely possible somewhat to be successful uh you need to like some of the uh theoretical Foundation which I feel a PhD or grad school gives you that BS to be able to spend that time with the subject so that you mature your intuition about the subject and it gets hard when we are learning that a job because we also delivering at a much faster base if you want the time and space to be able to really appreciate the the study and the learning it is nice to be able to go to graduate school or PhD program where you can just focus on learning and growth exactly because when you're working full-time you just don't have much time it's a luxury to be able to study right absolutely yes if you're not in a place that where you can go to school or if you're not a school type of person there are still ways to get into it by finding opportunities within where you're already at absolutely yeah one of the advice I do give to folks who are getting into ml revise your fundamentals like mathematics or the basics of linear algebra it really helps to build on top of that ML and eii as you know is extremely fast moving as a field like last year in one of the big conference in machine learning kdd one of the kot speaker said seeing advancement per day the way we used to see in a year so the amount of that need to come out in a year is now being published every day almost so it's really fast moving I got my PhD most of the deep learning methods that we used today did not even exist how you like it up is if your fundamentals are very strong so you just like back and revise those fundamentals build intuition for the basic models like learn logistic regression really well and explain deepal network from it because a lot of fundamentals are very translatable and then you build on it and then when we have large language model generative AI foundational model you can build up on your fundamentals so if you have the time read up some of the seminal papers read up your revise your math and then you're set for the future like Basics are clear then as the field is moving you're able to catch up I love that a lot I also used the example like I was in college we didn't have smartphones I know I didn't know anything about smartphones but I got a job in Mobile and I was able to learn it because once you have the basic fundamental is much easier to pick up new things one of the biggest dilemas for a lot of students and we get this question a lot should I go to grad school or should I get a job how do you think going to a PhD program serves you now as a machine learning or AI engineer working in the industry the decision of going pursuing a PhD degree or even grad school if you're getting your Masters I think it's definitely a very personal choice uh and it really depends on where you are at at that point in your career if you're already a ml engineer with your undergrad degree and you're able to perform really well you don't feel the need to pause the working and go back to school and revise and learn the subject further for me personally I was craving that pause and being a part of Academic Program where I can just purely pursue knowledge that really helped me in my trajectory as a ml scientist now in Industry because I could really spend a lot of time with the subject you spend many many hours with yourself you think about one thesis question and spend many hours days and months and years to to find the right answer I enjoyed that Journey my pH was very collaborative I worked on building machine learning models for disease prediction and R stratification so it taught me a lot of collaboration communication skills while I'm picking up a lot of deep technical expertise it also helped me become much more confident in my ability because of the individualistic of the pursuit another skill set that I think PhD program can really give you is we write a lot of papers as a PhD student it helps really really hone your written communication and verbal communication and that's one skill set that is absolutely invaluable in Industry getting those chances to really horn your technical skills while developing communication while developing confidence within yourself learning to compete at a very Global level that builds up a confidence that builds up your belief in your own ability yesterday or a couple days ago I saw this comment on one of my videos this person was like now chbt can read all the papers for you so you don't need to learn how to read did and I thought well that's not like saying we have calculators so you don't need to learn yeah and if Char is reading and understanding that what are we going to train CH on in the future yeah right we don't want just machine generat text we want real people to be still continuing to able to write journals and write papers and the human J text is still going going to be around and I feel like even if you do use chbt is just a tool to help you if you don't have the basic fundamental knowledge it's not that helpful even if cha PT reads it for you Chach PT is the one learning not you that yes I think in the future we will free up our mental space we have already feed up our mental space from memorizing right we use Google we use other information on internet we don't remember everything we needed we used to with chat GPD with gbd4 with these generative models I think we will free up even more of our mental space with the things that can be solved as a result as humankind I feel we'll be more creative eventually it might take a few years maybe a generation to get there every other pivotal moments uh like Industrial Revolution and uh when internet happened there was unsettlement in humankind But ultimately we Le we ended up in a better place yeah I feel like as humans we are going to get more creative we're going to focus on other kind of energy that these tools and in a way you rightly put so that these are tools that will help us be more productive but at the same time we will as humankind start using our brain power for something else absolutely I I 100% % agree I read this book recently called range I think he said something about how human IQ has been increasing every decade as an overall human species we're getting smarter counter argument I hear often is like how many phone numbers do you do you know I'm like why do we I remember contextual things a lot if you tell me something that happened to you if you teach me a concept I'll remember it forever but if you tell me a name I'll forget it here growing up I would hear uh folks say senior previous generation say that oh this person is super smart because he or she can remember stuff I was like why is that equated to Smart yeah good memory is part it's part of it but contextual memory or understanding concept it's hard to quantify that exactly that's why I think it's easier to just say if someone remembers a lot of stuff it's smart versus it's hard to quantify when you understand a concept or you're very creative now I think I also read about how back in days when the writing first came out people would say oh my gosh no one has to remember anything anymore now we're going to get all done these tools are serving us and allowing us to be focusing our energies on more creative things but maybe in the generation or two they would think oh back in the days people didn't do this 100% totally I think we're very aligned on this and another question that we often get a lot is how important is it to balance your technical expertise with other sof skills like communication and Leadership skills to perform your job well of course the technical skills is the foundation right you need the expertise that you are required to perform uh at any given job so technical skills are non-negotiable but beyond that a lot of the soft skills I think go a really long way communication and collaboration I think at various point we spoke about how important it is let's say two individuals who have equal amount of technical expertise but one person has ability or has horned the ability to to communicate succinctly and and articulate things and in a timely manner can go a longer way compared to the other person so technical communication to be effective collaborator to be effective uh researcher to be succinct and effective Communicator itself is a very very valuable skill to be successful in industry in addition to that other soft skills where just genuinely being a good colleague to work with person who others are delighted to actually work with I think also helps because who wouldn't want to work with someone who is who is a happy person to work with compared to who is a grumpy person to work with so those are kind of almost non-quantifiable skill set that I think are good things to think about actively and have the to mind uh beyond your technical skills communication being a collaborator fast decision making is another soft skill that I feel is important because Innovation velocity in any company in industry is fast especially machine learning AI the field is moving fast so you want to see through the all possible options but quickly enough and then commit to something and move on rather than taking a really long time to get to the decision of what we should even build all this thing kind of come together communication decision making collaboration and just being a nice person to work with and you mentioned several times how things are changing really quickly what are some things that we should look out for and what are you excited about as an ml engineer there's tons of innovation happening at a pace that we have never seen before overall I am very excited about uh the impact these large language model and the foundation models can bring to different applications so what I mean by that is traditionally we would build specific models to solve specific problems now with a really large models foundational models it's almost like we're able to train a model to be fifth grader or undergrad level student and then for a specific task we can fine-tune the model to S to be a PhD in a specific area right so that's how I think about it instead of training a model from scratch now we have this well understood about the basics of the world kind of models which then we can and for the train to make them extremely well in specific task my prediction is that soon we'll start seeing high quality Innovation happening for example Healthcare problems or even recommendation and search where we find tune the model Leverage The Power of these greed models but also solve very specific task and do it really really well proprietary data will become even more powerful if you have a very specific data for a specific task like gene expression data or Healthcare data then you can Leverage The Power of these large models that are trained on very big uh data set and then use your propriety data to train fine-tune the model and get really high performance we are entering into this era where we'll see much more improved applications of ml for very impactful task sounds like you're more excited about expanded ability for us to build more and achieve more as sofware Engineers yeah yeah and leverage AI for many more tasks that we have not been able to leverage that reminds me people also ask is it too late to become a Sofer engine should we not do it anymore more is it useless I we have the answer yeah yeah absolutely no no it's not useless I mean the kind of software engineering you will do is different we are evolving we we will we are requiring more and more let's say prompt engineering the skills that require uh will expand I don't see it's disappearing it's expanding yeah like kind of like what you said earlier about it will enable us to do more creative work and take on more tasks and maybe do things that we haven't been able to do in the past yes well thank you for sharing your advice thank you so much for having me yeah and thanks everyone for watching too thank you thank you byee

Channel: Exaltitude

Views: 116,431

Keywords: software engineer, software engineering, software developer, computer science, AI Engineer, Ai engineer career path, Future of Software engineer, Software engineer, ai jobs for freshers, ai jobs salary, artificial intelligence, machine learning, machine learning engineer, software engineer salary, junior engineer, junior software engineer, new grad, career growth, junior developer, ml engineer, women in tech, women in stem, women in ai, big tech, faang, netflix

Id: v45cTIDbj9E

Length: 14min 4sec (844 seconds)

Published: Thu Apr 04 2024