How Data Science is Evolving Technology...

data.bythebay.io: Ryan Orban, How Data Science is Evolving Technology Education

Recording: data.bythebay.io: Ryan Orban, How Data Science is Evolving Technology Education

um it'll be uh first part really just telling you about galvaniz in the program and kind of what we're doing and then um some thoughts around how data science uh is going to continue to transform education and uh when I was thinking about this this morning and putting my slides together you I said okay how is data science evolving technology education and I said um you know actually I should have wrote how data science will evolve uh technology education because we're not there um and we're not there yet um but we are in a very very exciting time um but before we get to Galvanize you know people always ask me uh why do you do what you do um and and what I do is on the CTO galvaniz I run our curriculum and content teams help kind of shape the future direction of what we teach and how we teach and before Galvin I was I was the co-founder and CEO of a little company called zipan Academy which is the first uh data science immersive boot camp um in the country and we joined galvaniz about a year and a half ago and brought the whole team over um and people always ask me why why do you why do you teach right um isn't there this whole old adage that those who can't do teach right did you not want to do data science and I said no no no I love data science my personal passion is how they two intersect but in in order to have a context of of why education is so important especially going into the next 10 to 20 years I want to have a little bit um of a history Liston and this popped up in my newsfeed today so um NPR has done some great work um they had this article that came out in 2015 in February and it was tracking the basically most common job in each state between 1978 and 2014 and I spared you guys the animation and just went directly to 2014 but uh what based on this image is the most popular job in America right now truck driver Y and then this morning if you guys were watching Hacker News you saw that this company came out of stealth that was called Auto it was a bunch of uh X Google and Facebook and Tesla engineers and they are building a $30,000 apparatus that's the thing that you see right there in the front of the semi TR that essentially will allow it to do um National autonomous driving right putting all of those people out of work and they uh they went out of South you you also see that Volvo and Mercedes also launched their own self-driving cars and self-driving semis and so we're in this kind of perilous future right where our most popular jobs in America will be automated and uh I found this as well also done by NPR uh this is also done in 2015 and um the title of the article is will your job be done by a machine in the future um um I suggest you guys check this out um statisticians it was the closest thing I could find to a data scientist um it's kind of our proxy here are in good shape it's about 21.8% of being automated but um those pesky software Engineers not so much 50% chance um so I think everyone in this room is probably relatively safe if youve self- selected into this conference but um there is going to be a need for people to retool and res skill and a lot of that is going to be working with artificial intelligence working with data being able to program the machines rather than working for them and so I think making sure that we are providing opportunities and honor amps for people to retool either at the beginning of their life midlife or later in life so that they can take advantage of this new economy is going to be key and based on my experience um traditional higher education institutions just can't keep up and can't move at the pace of Industry so that's what brings us to galvaniz that's it's where you're here uh we're not just a pretty event space uh we're primarily an education company and our motto is it's where learning goes to work and um this is really born out of the need for education to be more responsive to move fast at the pace of industry and in order to do that we blend the line between learning and working and so when you're on this campus today um take note you'll see that there's over 200 almost 250 students on Campus Learning web development and data science there's over 700 members who coming from big companies like IBM and Google to small one and two person startups and they're all working hand in hand in the same environment and they're learning from each other they're taking programs and mentoring and completing projects with our Mentor companies and our companies are getting access to the high level of technical education that we provide and I think this is really the model for what education can be and what it needs to be going forward is that people are not hiring based on degrees anymore they hire based on competency uh they based on what you can do not necessarily what piece of paper you have and there's no better way to learn skills uh than on the job itself and so we always say that we're solving this Chicken and the Egg problem between you need experience to get a job um but you need a job to get experience and so for someone um coming into let's say data science or web development for the first time um we we plan on being the bridge we have a couple campuses Nationwide we're primarily us-based uh so you're in our San Francisco campus which is our largest um but we also have campuses in Colorado with the company was initially founded um Seattle we're opening up Phoenix soon we just opened up Austin so on want a plane quite often but they're all beautiful and they're all large and they all follow the same pattern of bringing together industry um and education and we do this primarily through three programs so we have a full stack immersive uh a data science immersive so full stack is three six months our data science is three months um and we also have invested pretty heavily in doing data engineering training so this is part-time workshops in corporate and we've graduated over 500 students last year and we're on track to probably triple that this year so lots of growth and it's by large worked um 97% of our students from our web development program have found jobs within six months uh 94 from our data science program and more often than not they are doubling their salary um in a a 3 to six month program um these aren't easy programs you know when we say it's an immersive boot camp we mean it um people are saying goodbye to their friends and families for a period of three or six months and this is all they do and I think that personto person commitment in that inperson component is um something that's really important it allows a true acceleration of learning um because you're working with pairs you're working with a group you're not just uh working on a problem set are going to lecture in our data science program uh which is probably of most interest to you um we cover pretty much all the greata setes primarily in Python um and it's a 12we full-time program but we cover both the machine learning um the big data and the underlying Theory the bar is relatively High we look for people who have uh a background in software engineering um not necessarily as a professional but they know how to code and the right math and stats background so if you've taking linear algebra and statistics in college you should be good to go and they work at companies like Tesla Uber uh Facebook Udacity Twitter Assa um we've had a tremendous luck placing data scientists and web developers and some amazing companies and then they come back and continue to support the program um both through mentoring and get speaking uh we recently just launched a data engineering program as well um and that was built in collaboration with industry which is you know part of our big ethos and that really covers the kind of back end of data science so how do you process and shuttle around large you know petabyte scale amounts of data and we see that our data scientists and data Engineers typically work in concert to help break down that wall between engineering and data science so galvanized sales pitch is over over let's let's talk about um some more fun stuff you know one of the reasons I think that we're successful is we think about um our curriculum as a product um and one of the ways that we do that is having a very quick iteration Loop in our development process um I was in a conference last week with a bunch of really Innovative people from higher education and we were talking about how do immersive schools and higher education work together and one of the things that fell out of that is their current ly isn't really a bridge between how fast we can move and how fast they through regulatory requirements can um the way that we look at our curriculum is we're trying to meld the best parts of what higher ed has done which are things like adaptive learning um standards and objectives mapping things like Mastery tracking and thinking about learning experiences with the speed of a Silicon Valley startup and you know so we actually start um and we have a Consortium of Industry partners that we work on this standards and objectives right we help map out what are the skills that are required um for a particular job say it's web developer or data science um and then how do we actually validate those in an automated way so we're you know you can think of it as test driven development and that we're writing tests um for things that can be Quantified and then actually writing the curriculum learning learning experience is just similar if you were doing test driven development you would write the test and you make them pass and then we get feedback from students in industry and also from um the outcomes of our students and and our curriculum is changing daily if not weekly you know we're graduating a data science cord every seven weeks and we're continuously updating and getting feedback both from how we teach but also how the you know industry continues to evolve because um data science as it stands today will not be the same a year from now a lot of machine learning will still stay the same but the new technologies the new ways of thinking new algorithms so we have to stay at the pace of industry and part of that is not um expecting that we're going to have all the answers we need you guys to tell us um to make sure that we are producing the best content that's available and we think about this as a lifelong journey right um it's not just a oneandone come in here and do uh a three-month program and then you're kind of off and you're a full- fetch data scientist uh we think of ourselves as an accelerator program and that you are going to be learning throughout your career and we hope that you come back someday and so we think about you know someone discovers us on the website maybe they take a course at Udacity or code academy we go in and have an assessment and actually figure out their skills and what they can do based on that they come in and do some pre-work and learning get placed into an amazing career and then start the cycle over again from within industry discovering what they need to know relearning on the job getting that industry immersion and then coming back maybe five or 10 years down the line and being able to work with Galvan I to reschool rescale and retool um as the industry continues to evolve so let's talk about data um we measure a lot of things in our programs um from our admissions to our Student Success which is really owning the student life cycle within the program and then Career Services after they're placed and there are some opportunities for automation here right now our admissions are all done by hand and we create what we saw we say a quality score that's made up of both technical interviews s skill interviews and Tome assessments um one school hbert and school you probably probably have not heard of it they just launched they've actually completely automated this process and it was really inspiring when I was talking to them last week um but we don't want to take the human element out of it I think that's what a lot of online education gets wrong and that they try to put everything at the mercy of the algorithm uh one of the reasons we're successful is because we have that human element um and that we're actually building that relationship with the people um based on that they go into Student Success so we collect a lot of surveys we try to do things like measure their problem solving how fast they complete exercises whether is their engagement um automated assessments which I talked about earlier and then getting things like exit tickets on a weekly and monthly basis so that we have a good pulse over how things are changing overall and largely we're not just looking for competencies we're actually looking for Deltas um someone can come in and if they are learning a ton we can give them the same assessment week to week and if we see a big jump there then they're doing well in the program and then we obviously we um measure things on our placement rate how fast it takes to get people played they're starting salary and their net promoter score but the dream is not just these metrics I think when we talk about how data science can really be effective in education is how do we actually build this feedback loop you know one of the most interesting data sets is a longit to in the record where I can take both their education and work history how they did on an admissions test how they did throughout the entire program from automated assessments to uh what data they fed us and then what their ultimate outcome was right where did they get placed how happy are they with their job um and then from that we can actually build a very interesting model um where we can then feed that back into our admissions in Student Success process process um and be able to find Diamonds in the Rough and you know one of the the hard Parts about admissions is that it's primarily a filter and that you have to have the opportunity and the time uh and the money in a lot of cases to actually go through a process like this and by creating a model and being able to sus out things in a much more quick manner um and have a higher degree of accuracy we can actually open this program up to a lot more people and we can do it at a much cheaper cost uh which is our ultimate goal and then from there we can do things like Predictive Analytics uh on Student Success and that's why I kind of mentioned earlier that we're not there yet we're still collecting the data and one of the hardest parts about what we do is just the sheer amount of scale um you can think about it it's a very wide data set with not a ton of uh students we've only been about 2,000 already so we're starting this process now um but as we continue to scale and get better data uh and and we're launching an online component as well that will help us collect even more data um I think we can actually create some really interesting insights and and one of the ways I've been playing around with doing this has anyone heard of item response Theory we got one all right so this will be new I was hoping someone would alert something today um so item response theory is this really interesting branch of of mathematics um and it's used uh anyone a GMAT or an act any of those so if you have you've actually used that in response Theory um it is the successory to what's called classical test Theory where you would look at a student in their grades as the you know kind of accumulation of points within a particular test and what that doesn't do is it doesn't allow for variability in how hard the question is um and so what item response Theory does is essentially it applies a logistic sigmoid and we're trying to actually estimate the probability or the underlying lat in feuture which we call Theta here of how successful or um how much Mastery does an underlying student have um and we can model this in two directions one this allows us to figure out which questions are actually difficult and which ones are not and we think about that as um how much information is contained in that question if it's easy and everyone gets it um that's not a good qualifier but if it's way too hard and no one gets it and someone who has a very low comp just happens to have guessed the right answer um those should be treated differently right you want to look at the the student as a whole and being able to model um and really throw the assumption that all questions are the same or they all have the same value um allows you to have a much more personalized experience and so on the x-axis here we have Theta which is their um basically perceived Mastery neg3 being really bad three being great and then you there's essentially a couple parameters that you tune with this model one has to do with the person and the other has to do with the individual question so you get a matrix um and then from there you're trying to uncover the underlying Theta for that person which is um how much they've mastered and then through that what's really interesting is what we're starting to see is you can actually do Predictive Analytics here so based on what we've seen so far um we can actually recommend which questions um or recommend which interventions we need for that student based on this underlying model so this is something that's very exciting it also needs a lot of data um so one of the you know kind of programs that's successfully rolled this out as Con Academy um they very finally named it guacamole which is why the guacamole is there and so if you guys want to check this out or actually run some simulations with some data um on your laptop just go to GitHub con guacamole and um they have a a pretty much TurnKey solution there for you so they were we were good enough to open source that and then you can actually take this one step further and do something called uh multiple item response Theory uh and this is when you're actually creating a surface this is is over I think 30 different um latent features and so we're not only looking at Mastery but we could think about general intelligence we can think about grit and gumption um all of these components that fall out of a large test Suite we can actually model across the board and create better assessments and create better experiences for our students so with that um are we in a world of the robot tutor I don't know if you guys remember this from the the Jetson you know are we going to um see a world where you know we have you know essentially AI coming in and um you know teaching us everything we need to know and is that really the future and can we get rid of humans after all and um in some ways we're already doing this um I don't know if if anyone saw the story that came out a few days ago it was called Jill Watson um a professor essentially put IBM Watson um as a virtual TA in his artificial intelligence class and he didn't tell anybody um he named it Joe Watson through IBN Watson in there and they trained it using real life Tas uh until it had 97% accuracy um or confidence in that it was actually answering the questions you see should we be aiming for a TH or 2,000 words I know it's variable but that's a big difference Jill steps in there is no word limit but we'll be able to grade both on death and succinctness blah blah blah blah blah and then um they ask a more elaborate question and then lth an actual human ta jumps in because Watson was not comfortable answering that question and um you know some people start to wonder if Jill is a computer if there's anything this has taught me I should always question if someone I've met online is Nai or not and before they actually announced that it was nii Jill was nominated for the best TA of that year because of how fast and complete and succinct her answers were because she would get back to people at all hours of the day and nights um and provide uh amazing resources and it was only until she could not answer something that a human have to step in so are we there no but are we getting there yes and this is really exciting to see um but I think what this underlies is the need that AI is not um a solution unto itself what we need to be thinking about and you see this thrown around a lot is human assisted AI um and how do we leverage these amazing Technologies and algorithms that we're doing to create um better lenses in which to view the world so that humans can actually get involved um and take some of the labor out of the equation take taking a bit of a right turn I think the other aspect to this um is is not just the in-classroom experience but it's also understanding the market and and where the trends are going and I think um bright.com did an amazing job of this they required by LinkedIn I think last year and um their goal was to understand and be able to quantify someone's ability to um actually do a job and they did this primarily through um a collection of outside data and data science um looking at their resident resume and doing some uh natural language processing and ultimately came out with a bright score and they could say for each individual um what is the probability or what is their score that they're actually going to U be able to get you know placed in this job and be successful and when we look at the the future of kind of what education might be I think LinkedIn is definitely on to something um LinkedIn has acquired both bright and Linda and if you take a moment to think about it it's actually a pretty brilliant ma master stroke if they can pull it off um LinkedIn kind of owns the world's economic graph right they understand where people move and uh what their skills and competencies you've given them all of that data Brite can understand and and kind of quantify at a particular level and match people within a particular job and then Linda is an online training um institution that has a lot of technical aspects to it as well and so you can think if if linkedin's job is to create Economic Opportunity um wouldn't it be great to be able to take anyone on the the planet who's in linkedin's network be able to analyze their skills and then uh provide them with a Linda course and say these are the five courses that you need to take to get to X job um I think it's really interesting if they can pull it off people I talk to at LinkedIn are being a little closed mouth about it but um I think under the under the surface that's secretly what they're doing and we're trying to do uh something similar uh we're obviously not going to go head-to-head with LinkedIn um but we are building our own skills matching service it's much more tailored to what we do which is data science and web development and we're calling it flight it's in beta right now if you guys go to students. galvaniz you can sign up but um right now it's a manual process right where companies can come in they can provide the skills and the jobs that they're looking for um we can get a curated list of galviz graduates and then our outcome staff actually matches them manually but really this is just building a training set for us um over time you know this is easily automatable um through you know the algorithms we don't know but we just haven't had enough data to to to go through um and so I think you know in each aspect we're trying to create you know like I said uh human- centered or or human assisted AI where we can leverage technology to create better scalability in the business um but ultimately we're enabling humans to do their job much more effectively and ultimately change more lives and then when we think about um the kind of the next steps after item response Theory and having good assessments it's really about ad learning um this has been used to a pretty good degree in some K through 12 schools um with companies like Newton being able to actually do um sorry a fair amount of adaptive learning but we haven't really seen it in adult education we haven't really seen it in technical education where we're taking content um we're actually collecting data we're building predictive models that enable both students to understand uh where they kind of where they are in the course and for the instructors to really see at an individual level um how they can actually come in and do do uh an intervention and and talk to that person potentially get them more courseware or um you know some sort of medial training and then have that go into in a full adaptive engine um where you know essentially just it follows the same theme of human assisted AI how can we use and collect data in order to improve the student experience you even get to things like um you know out of this lecture model and into a fully baked personalized plan where students can you know grab an iPad or grab a piece of paper and they know exactly where they need to go all day but it's based on the needs uh for them and what competencies they needs to master rather than a factory model which is um kind of where we are now where everyone s to the same lecture everyone s to the same exercise it works but it can't always be better so I'm almost out of time and I'm only leave some for Q&A um so like I said galvaniz provides access to Talent uh expertise and an ecosystem so um you know we're always looking for um potential hiring Partners so if you're hiring data scientist or developers I would love to talk to you um we do a lot of corporate training as well we're obviously hiring because we've got a lot of work to do for both data scientist and uh software Engineers um we've got amazing event space as you guys already well know um and we're home to many many startups and established companies so if you want to get involved uh in kind of our vision of the future of Education um if you want a mentor or work on a Capstone with a students uh we're always looking for data but like I said um we're industry aligned and we can't do this with you guys so you're the stuff that makes it work thank you thank you very much Ryan thank [Applause] you