H2O Open Tour: NYC - Opening Keynote From CEO Sri Ambati

thank you so much for having it's joy in New York so many New Yorkers bright faces and so many people from different parts de Ville and I met a few from from outside the country how many of you are here outside the US and there where are you from UK welcome welcome one of the things I learned from the whole brexit situation is that the Atlantic Ocean is much closer than this shorter than the English Channel right because we actually speak English on this side of Atlantic Ocean I heard another Israel thank you for coming from we again have it h2o users how many of you have actually used such - OH before Wow I want to take a picture thank you so much how many of you are hoping to learn a lot about h2o this time it's awesome so hitched or the product hitched or the company have both evolved quite a bit quite a bit and we'll talk about the plans we have I think the biggest welcome for us was last evening how many of you watch the thunderstorm and the rain was dramatic rain right cut off the humidity we could enjoy the evening was fun so I think there's a lot of fun schedules rest of the afternoon a lot of great events we also are doing the unconference as as we know was mentioning and there's a little bit of good hands-on hack sessions data hack sessions so hacking data is kind of why we are all here and data has a singular way of transforming how we think over the world and I think that's kind of what we are here to talk about today how many of you are here from other parts the City office of the state's I heard I heard caught over the LexisNexis folks early this morning from Orlando Florida any other flop Florence Tampa Florida awesome awesome we saw we did an analysis of the RSVPs there's a big chunk of part of the population from the area here we've done annual conferences at the Computer History Museum for the last two years in a row and they were we're getting flooded with the hundreds of people there and we decided to decentralize and have truly go where the users are so I think our mission is to get this to such a good machine that we can do a dozen of this in 2017 not just three or four of these so our goal is to do more and there is enough content and and that's something one of the topic you'll hear as the keynote you'll realize that we are actually trying to build multiple products and they're a product company and that's not changing so you're trying to innovate the core platform in so many ways it'll be truly fun to see a lot of the collaboration and the community that brought us this far I want to call out a couple of folks a few folks have really done it tremendous amount of work putting this together especially the organizing team behind and you have two fillets of real will push your limits to the point where you can it have to produce some great work likewise the marketing team at h2o so without further ado when I talk about data when data is actually a datum right bottom is Latin for to give and so there's a lot of giving to be done but in terms of energy in terms of direction to the space and AI is actually one of those things that people will ask you will ask what is your company's AI strategy alright that's the fundamental piece that we are seeing a lot of different customers of ours the question about why right I mean a lot of my talks about setting this stage or why and the rest of my team which is behind me the still our team is going to demonstrate the how and the what right so my goal is to set the tone for Y the y is to put transformation we want to bring change and every entrepreneurs vision is to bring change and that's kind of what we are we are using data to bring change data science is the ultimate search for truth and in that search for truth we have found some amazing tools but some amazing opportunity for data as well the whole idea of open source and data coming together is such a phenomenal time in our existence as a software people that we can really look for truth in every walk of life and when I say every walk of life from a business standpoint a software is eating all parts of life you'll see that so we have a jokes that software is eating everything except food right so don't forget the lunch that people are sending up that's the only thing that will left for us right but all the verticals if you think about the verticals every vertical has to transform because traditional ways of doing business the base product that you're selling is going to become modified software follows faster than Moore's law of depreciation right Moore's law is the depreciation so if software is actually faster than that two people in a garage can build any the code that that you you hold private in so precious to you so the code is a commodity and as a result what you're doing today how you're making today whatever curve it is on it's going to go very flat very soon so all you have is your data and your data in the defense of your community vertical is the new horizontal right and and that's kind of your the crux of where we feel the world is headed towards is if you can actually truly bring your thinking to to bear in other words the data you have if you can use your data to solve your customers problems your community's problems and not just your own community go beyond it if your community is connected to other verticals so if you're a airline company you could be working on a real estate problem because of how you know flights are flying between different cities so think of your data as not just siloed in your vertical think of it as a truly horizontal product data like water will flow across different planes and so if you can actually connect data products from your domain to other domains that's where you see value so data products is kind of where the real rubber meets the road so h2o as a company is has learned how to build a little bit of community thanks to your love but I think what we are looking to do now is give you platforms to build data products and I think and data ecosystems and that's kind of where the next band in our evolution has is that's where the band end of it in the river for our customer sees as well is how to build data ecosystems and trusting alliances in data I think that's going about what you'll see is a big change in theme data is the only vertical we're all in and there is two four-letter words that I'm really pop popular saying one is of course I'm happy to trade data and cash in cash and data which is very important right but I think love is a very powerful part of it and love is is what you want to get from your customers you want to have almost almost a spontaneous reaction to your work and one of the things h2o experience which is quite transforming to us as a company is most of stuff we've done almost 80% or even 90% of stuff we have done had immediate reaction from the audience from the customers from the users and I think that's something that we've been fortunate to have and we continue to treat that as as a as a responsibility to keep that kind of trust from the community but building beautiful data products is a new transformation to h2o and we would not use Beauty in sense and data in one line four years ago but today we have built a team we can credibly say that we are going to bring true visual intelligence to the space and I think beautiful transformation beautiful AI is it actually assumes that there's a human in the AI I'll drop a little bit of that in my talk but really you'll find that there's rest of the day couple of talks will hit on that tawny shoe and esteem for three Lee Wilkinson we've had some really real power houses of of visual interpretation Jeff for Mika Stubbs will talk about that as well some of my demos will talk to that but building beautiful data products getting your user to interact with your data product so you create new data and that's the crux of where the cycle starts got a shift focus from the internal to the external and the value has to shift from just your product to the co system and that's true to our own transformation its Joy's going through building vertical ecosystems and that's kind of one of the things that we'll talk about it there we've always built horizontal ecosystems sparkling water is a great example of that where we embraced other people's work other open source work and combined to get Best of Breed open source to our customers and our community I think what you're going to see is how we now build vertical ecosystems so some of the folks in the audience so I met earlier are here to find partnerships with us we are looking for them so we want to build them so let's nurture that whole ecosystem now we talked about software eating the world well AI is eating software right so so that the true end of code as we know it as people say is right around the corner and and it's when I say right around the corner 20 22 it's not really far off before AI is going to dominate most all of software even today most rule-based applications are being rebuilt with AI engines with pattern recognition based applications so what was truly a data science movement has flipped over to be pop science real software engineering movement right so what we are seeing as AI today is true software engineering engineers being part of the whole revolution I think that will transform much faster because software there's this there's a saying I used to is to say or here sometimes it's to way to change the world and to try to be Deepak Chopra or try and be a software engineer right and software does touch as many lives right so that's the fastest way to touch a lot of lives other than of course Pokemon go which is probably also software right so the vertical eco system for h2o again a crayon diagram is we built six thousand plus companies have used such to now sixty thousand data scientists use it on a weekly basis of which eleven hundred of them have used it before eleven o'clock today right which is a fun set of six for us to hear and a few of those open-source users do use us do pairs there are customers as well as our community but as we start building them some of those customers started doing larger interactions with our team sometimes in form of little bit of services but trying to truly extract the product into a software engineering engine plugging it into their software and that's led us to now discover some really interesting verticals we're looking at building some verticals in CFO for CFO dai for sheer force with help of another large auditing house they're looking to build a ml based anti money laundering verticals with the help of real banks their partnerships the data whoever has data and analysis but one thing that's close to your heart is cancer we have now started building open data ecosystems for cancer looking at imaging pattern recognition in images an ultrasound that's where some of our other innovations that we're looking at from deep water from the deep learning side are going to be applied and this is a hundred percent open source vertical is the first true community oriented vertical in cancer and that opens up a whole new ecosystem on the top for us especially if you look at ultrasounds are different than the way look at ultrasounds for breast cancer is different further look at it for pancreatic cancer and so that opens up a whole slew of how do you monetize that in the end as IOT so there's a whole audacious plan of one at a time opening up into the verticals and really partnering with the particular ecosystem I think the beauty about what where we are is just getting started there is a data alliances that need to be built truly work with the new stars of the world experience of the world they could access to the world and build real data articles that we can then plug in back into the hto community into the steam products and so you can actually add data to your existing data so I think that's kind of the true vision that we have of enabling not just algorithms but data right into your own data so connecting different data sets is kind of going to be the the real true road ahead for us all of this delivered through cloud and eventually through IOT that's the real vision that we have for the next ten years and we're working at it one quarter at a time right as we as we are startups but the audacious plan is truly slowly emerging in itself with the help of our customers the help of our own community and with help of our own set of backers how do we incentivize steam sport none of this was is possible without the amazing team behind me right and that team includes my customers my community some of my earliest customers having the audience Eric from in davao reckon of her first pricing engine customer some I've actually met I've seen we call as well we call Kirk I was also one of my first customers in the same team but now some of them have joined us and our building product within our company but I think the customers and the whole ecosystem of people around us have truly incentivized us the score really takes care of itself for data is really a team sport right and I'm gonna set the tone for why we're doing the next three products that we working on and you cannot build any of these products we thought actually connecting the dots between the domain the data the apps right and then using algorithms and designed to truly bring your data as a sport because we thought applications you cannot connect the feedback loops and connecting those feedback loops are so critical for the applications and the first platform for data products that we are looking at is steam and the team's really busy trying to build several interesting pieces into steam first of which will be just model management how do I store models some of our customers and community have done that themselves how many of you have tried to persist h2o models and then come up with your own ways to store them pickle them reuse them I see some nods will have Alison and the team from SVA speak about it tomorrow but that's actually the crux of the matter database for AI all these models how they're stored how they're governed how there's code how you plug it into applications so you can get a rest endpoint that speaks to the true-true data that's flowing through the system that's the heart of the matter first steam of course comparison selection I mean hitch to is already good at finding model selection how do you plug that into production that's where the question for steam arises and data products need new tools you have a lot more dependencies data now you plug data into your product as a result into your logic of the product code dependencies are much more easier to debug than data dependencies because as the world changes your entire models pipeline has to be rebuilt so if you don't have good model pipelines you and the ones that you cannot see those are the ones that really cost most most issues for you this is the early days of the internet where you had to Cisco switches you plug two wires now we have hundreds of switches that's what's going to happen very soon and your data center is going to be clogged with wires that's the model ml pipeline jungle that you're gonna create for yourself and debugging and unraveling are some of the problems the earliest hitch to our users have already hit and we're trying to create toolchain to debug those visually as well as through an IDE right and I think that's the heart of the road ahead for building stuff how do you trace through these models how do you even keep models in check for if models are intelligent beings you only can keep them in check with other models right so then testing your models becomes really creating an ecosystem of models around them both the upper bound accuracy lower bound accuracy so all of those pieces are the stuff that our our team is now trying to focus they'll give you a tool chain so you can debug them the second product we're looking at is deepwater and folks will talk about that today later today Arno and Fabrizio with a phenomenal team we want to get best-in-class open source deep learning and that deep learning is obviously a framework it's not one algorithm and so all of the all the bests in all of the top frameworks that are out there including tensorflow Cafe MX net and and the ease of use from h2o all of them combined in a way that you can actually use it very easily in a way you can deploy the best-in-class but then morph it into something that's simple where you can delete the stuff from other frameworks that you you're not using the RN and lsdm from some of these frameworks is useful for you as an NLP user while for others CNN is all they need for the Imaging so I think the having best-in-class deep learning is truly what a year from now look at look back and say this is something we've really preserved with open source because I think the wars for algorithms are all headed into the deep learning into deep water in some sense right so that's that's the face you thought our team will talk about it as well it's another very big future for us I think it's Fabrizio in the audience somewhere he joined us recently or he joins us today I suppose and then we have on on the other side so we'll have a MX net I don't know if KK is here yet but we have most of the folks on these different projects already collaborating inside hitch to us so we'll we'll have a lot of fun doing that Nvidia and Intel are both committed to the resources to help us build them on on GPUs visual intelligence is kind of talking about this earlier that storytelling is such a human endeavor that if you cannot understand how the models working and not and trade-off between accuracy as well as interpretability I mean sometimes interpret abilities dominates over accuracy and I think that's kind of the hard problem that Tony and his team are really putting our struggling and we'll come to an answer on that one which would be a fun one because deeply I mean deep learning will obviously use it it's a black box model how do you convert a black box model to a gray box more well where I can understand parts of it how do I take the best out of that and probably bring back GLM on top of it so I can really understand what's happening but there is lots of real hard problems behind this and interpretation becomes far more valuable than accuracy at times and I think that's the part we we are have started UCR to d3 the making of our to d3 at some point during this conference on how the thought process behind how to truly tell a story and I think every business needs this today's business needs this and and the future also needs this and here's why there's several distinctive futures for AI one of them has the humans in it and others do not right so then the bots are the bots don't need visual interpretation it's the humans that need visual interpretations and for that we do need truly immersive intelligence and I think that's one of the reasons why we're really investing in this is that without visual interpretation we are no longer as as important to debugging the AI paths go ahead and I think that's the key piece it's really something that's going to make or break for the for the for the environment here and I think AI is truly powerful some of the jokes we have in our you know water cooler are interesting AI has been with us for hundreds of years and people have been predicting the future of AI for so many years even in the last 20 years I have seen AI take off couple of times false starts and and and the joke we have is that AI will be with us in hundred years and maybe the only thing left in 100 years if you do not actually bring true human understanding alongside AI and AI on one hand does make us be more human because it takes away most of the menial work but I think if we cannot debug it just as well as a toy as a child t bux the toys we would still be able we're still be at loss in terms of intelligence AI has a capacity to increase the world's intelligence by a hundred x without having a huge bio footprint it's a platform for data many of you have already used it sparkling water how many of you have used sparkling water there's a few hands here your hands they're one 1/4 of our downloads today are sparkling water which is pretty drunk pretty pretty drastic pretty large actually and and people use us both in the cloud alongside data breaks cloud alongside IBM's as well as in in Amazon on Prem their two distinctive architectures one that streaming and one that is bulk so both cases you have seen customers users in different forms and SPARC provides what we haven't done yet if you will in terms of ETL we don't do a and T right we we do definitely to the Intelligence part the modeling part there's clearly design patterns for smarter applications again we are seeing lots of different patterns including and some have not even mentioned here like the closure on some of our customers have stormed this deaf no js' react yes so lots of different interactions but I think the real key piece is to make sure we have a loop the data product needs to be exposed to the end-user to get user interaction and data products are really fun right they actually allow you to take your rebuild your applications in the cloud one one more time so connecting the agenda of your company companies cloud agenda with AI is something that is happening it's happening both from the push from from Google and Microsoft but also from general common sense where if you rebuild a monolithic app today you're going to put it in the cloud for most part and I think this micro service mindset is kind of like really bringing AI into a lot of different applications data is the new clay and and towards that we are putting together a phenomenal data prep engine in form of data doc table so you'll have a talk later today from Matt Dahl I think he's somewhere in the audience and where he will demonstrate 10 billion drawer joins like one of the world's fastest largest distributed joints using radix sort and again open source and I think one of the things you'll see is that we're slowly beginning to help the data prep side as well but it's not something that's right there today it's headed towards production used words end of the year but that's one part of it the second part of it is building real alliances and data products data alliances working with real data products so you can then plug them through steam into your environment so api's simplest example is a weather API we could pass through a weather API into your application through Steam and then now as a retail user you can start adding weather to your existing data set there's lots of little data products that can be coming that can be pickled upfront in deep learning and be available in the steam default so you can actually start using those data products out of the box and start connecting them into your applications I think the reality is to truly it exemplified domain experiences and make them productize them I've heard this saying several times startups are all team sports I'm here because of a phenomenal team behind me and some of them or most of them are listed on the site as well I'll probably pull them up one of the times they're in the audience here but we've been growing rapidly they're a team now of roughly 70 people still dominated more than what more than 60 65 % of them are still engineering and culture is truly your co-founder and in my team I find the partnerships that I have been building every one of the team bonding is not the one moment when you sign that in paper every moment is a founding woman for us every folks every one of them who joined us has this huge finger footprint and fingerprints on the company the my goal for the company is to build a lot of really true good leaders but let me bring up one interesting piece about the company we do have three and F in my small career ever ever I've countable number of times number of folks have worked with African Americans in technology and Silicon Valley has a lot of people chrissy around it and there's something we're going to call straight on we do have about half roughly heading towards half but half of the company is going to be women in civil real leadership positions as well and it's not just numbers we actually have equitable pay as well I think some of the changes you want to see you want to make happen in your own journey and that's something we have really taken as a responsibility for our own company why make one CEO when you make 10 is a question we ask in the company and the way to do that is actually exposed most of the data within the company to almost everybody within the company and almost to my board is one of the most educated board and they could walk into any big day the company and truly take them to to task and to a large part most of my engineers are like that most of my sales team is like that our marketing team is also has been doing that my GA team might add even my my admin team they've all been part of the company rebuilding I think we're gonna build such an amazing set of leaders within the company that will be very the h2o makers will create a new revolution so one one Sunday I woke up with a much more interesting view of what can be done and that's how can I build 10 great women CEOs in Silicon Valley and I think that's also one of the things that we're really pushing a lot of our leaders are being formed like Amy when she joined me she was the first job here right sort of and she's from New York and I said I'm gonna move you from here and she did move I think about a weekend thanks to and they have been growing rapidly several people several four customers will tell you how much they have touched their lives I think we have another like ADP like one of the questions they had was can I get some more time from a bank right but that's just one of many people who poured their hearts into this company Josephine and Carrie who's of course New York and we just have Beth to join us if several real strong strong leaders being formed as part of her journey I think startups are not just technology companies the technology beings we owe it to ourselves to actually being the emotion emotional energy behind the creation that we're doing and change will come in spades once we have leaders on the top one of the change I used to talk about is that every generation is to fight its own revolution and of course the Emperor of all maladies is as this guy we call cancer or Merovingian from the matrix world the thing I was I think was presenting this slide to one of my colleagues at the time convincing him to join us Mike Bukowski and the slide was presented in a very tilted fashion so you see there's love in revolution love powers most revolutions one of the things the most recent June 6 finding invalidated 40,000 deaf MRIs that were done for the last 15 years because of a small software bug which means about 15 years of research is pretty much up for a question and that could be that happens often times you find bugs you fix them but you don't have the time to run all those experiments again and the way to do that is have open source be the guiding path with enough ice all bugs are shallow and that's the fundamental theme behind open source and open data and I think that's kind of what we really want to bring change if you want to bring change as a third bottom line for the company that would be trying to really enter the wonderland an Alice in Wonderland is a fundamentally new way of thinking of the world where everything is possible with AI a lot of things are possible with open source a lot of things can be bring to bring to solution so we're trying to build an open source data ecosystem for cancer truly in a vertical fashion and we started one separate time is now bringing imaging focused on imaging and ultrasound and we're working with several hospitals Stanford Hospital we started working we should work with our customers as well we started working with NHS we started working with our AR Cancer Institute in India really far and wide the ultrasound is actually very inexpensive to to do it's actually non-invasive and there's lots of good data and where data right so you can actually look at both pancreatic cancer and the simplest normal gastric trouble in this copy so all of these are all data sets and they become deep learning problems right once you start applying deep learning problems and bringing in biomarkers we will bring change to this world steadily one step at a time and we won't do it without your help we need your help and your help is going to be coming in the form of data science in the form of engineers in form of truly building products with us alongside us in the community it's going to be open sourced so we will try and get as much of our radiologists and oncologists and writers and the society involved I think this is a true way of unlocking a lot of a lot of the clutter we see in healthcare today and finally we're going to consumer eyes the data right data collection if you consumer eyes it you can actually get enough big data behind you to rally behind this call I think this really a clarion call for creating an open-source health diagnostic systems where if you think about how we've really focused on trying to make sure our laptops software is tested more than the software that does a cardiogram or a software that's looking at our fMRI a software that's looking at our cancer diagnostics the false positive raised in breast cancer is 58 percent so that it's and and the false positive rate in fMRI was 70% that was what invalidated that paper so these are things that cannot be left to be solved by commercial interest themselves these need to be really attacked with smart people like us where this hardware we have this body we have is going to last for just less than a century why don't and and why should I work on a digital twin for a turbine which can be fixed easily versus working on a digital twin for our own human heart I mean these are the things that really are gonna inspire us to build the next generation of open source and it's not just much like how sass or sibling in 40 years ago started with the base statistical software and verticalized into all these ecosystems we're going to do that in open source and pay a very open approach and bring a lot of different customers community and the rest of the world to bear it's the power law of networks once we start doing one network the other networks will emerge and if you are sitting in a circle as each of you are on a circle here there's possibilities for infinite ideas not just one idea and I think we look forward to your feedback so we can bring change one thing we will never forget is that is there still day zero it's still very little and there's some fundamental advantage of being that little being you can build HECO systems you can partner raise a forest not just a tree right and I think that's what keeps us really awake that's what keeps us coming back to work every day and change the world one step at a time time still is the only non-renewable resource and this one has says the keynote is over time it's about time I have to conclude and the key piece that was going to bring up is that people are such an important part of this transformation I'll go plus data is business but I'll go plus data plus people bring change and that change is coming form of business transformation we put together a couple of units for our customers and the world really needs more of these more of the dreamers because I think technology is actually much more easier to build these days than ever before and and I think the rise of the business technologist is everywhere it's - as a movement has matured it's now bringing AI to business it's not just data science it's the combination of data science software engineering the data engineering but also the storytelling aspect of it and that's not been possible without your help without the community that has grown from just mere tens of people who are playing with our product to thousands of them who are using it now and many of them have paid us some of them have used us and given us feedback feedback is in the end more valuable than even cash we are hiring the right kind of crazy people we're willing to join us please send us your notes this original people who dare to change the status quo but thank you so much for being part of this journey and we really are looking to bring change

Channel: H2O.ai

Views: 2,060

Rating: 5 out of 5

Keywords: H2O Open Tour NYC

Id: hPkyNhMR36E

Length: 38min 17sec (2297 seconds)

Published: Thu Aug 11 2016