>> Announcer: Live from New York, it's the CUBE. Covering the IBM Machine Learning Launch Event. Brought to you by IBM. Now here are your hosts Dave Vellante and Stu Miniman. >> Welcome back to New York City everybody the is The CUBE the leader in live tech coverage. We're here at the IBM Machine Learning Launch Event, bringing machine learning to the Z platform. Steve Astorino is here, he's the VP for Development for the IBM Private Cloud Analytics Platform. Steve, good to see you, thanks for coming on. >> Hi how are you? >> Good thanks, how you doing? >> Good, good. >> Down from Toronto. So this is your baby. >> It is >> This product right? >> It is. So you developed this thing in the labs and now you point it at platforms. So talk about, sort of, what's new here today specifically. >> So today we're launching and announcing our machine learning, our IBM machine learning product. It's really a new solution that allows, obviously, machine learning to be automated and for data scientists and line of business, business analysts to work together and create models to be able to apply machine learning, do predictions and build new business models in the end. To provide better services for their customers. >> So how is it different than what we knew as Watson machine learning? Is it the same product pointed at Z or is it different? >> It's a great question. So Watson is our cloud solution, it's our cloud brand, so we're building something on private cloud for the private cloud customers and enterprises. Same product built for private cloud as opposed to public cloud. Think of it more as a branding and Watson is sort of a bigger solution set in the cloud. >> So it's your product, your baby, what's so great about it? How does it compare with what else is in the marketplace? Why should we get excited about this product? >> Actually, a bunch of things. It's great for many angles, what we're trying to do, obviously it's based on open source, it's an open platform just like what we've been talking about with the other products that we've been launching over the last six months to a year. It's based on Spark, you know we're bringing in all the open source technology, to your fingertips. As well as we're integrating with IBM's top-notch research and capabilities that we're driving in-house, integrating them together and being able to provide one experience to be able to do machine learning. That's at a very high level, also if you think about it there's three things that we're calling out, there's freedom, basically being able to choose what tools you want to use, what environments you want to use, what language you want to use, whether it's Python, Scala, R, right there's productivity. So we really enable and make it simple to be productive and build these machine learning models and then an application developer can leverage and use within their application. The other one is trust. IBM is very well known for its enterprise level capabilities, whether it's governance, whether its trust of the data, how to manage the data, but also more importantly, we're creating something called The Feedback Loop which allows the models to stay current and the data scientists, the administrators, know when these models, for example, is degrading. To make sure it's giving you the right outcome. >> OK, so you mention it's built on Spark. When I think about the efforts to build a data pipeline I think I've got to ingest the data, I've got to explore, I've got to process it and clean it up and then I've got to ultimately serve whomever, the business. >> Right, Right. >> What pieces of that does Spark unify and simplify? >> So we leveraged Spark to able to, obviously for the analytics. When you're building a model you one, have your choice of tooling that you want to use, whether it's programmatic or not. That's one of the value propositions we're bringing forward. But then we create these models, we train them, we evaluate them, we leverage Spark for that. Then obviously, we're trying to bring the models where the data is. So one of the key value proposition is we operationalize these models very simply and quickly. Just at a click of a button you can say hey deploy this model now and we deploy it right on where the data is in this case we're launching it on mainframe first. So Spark on the mainframe, we're deploying the model there and you can score the model directly in Spark on the mainframe. That's a huge value add, get better performance. >> Right, okay, just in terms of differentiates from the competition, you're the only company I think, providing machine learning on Z, so. >> Definitely, definitely. >> That's pretty easy, but in terms of the capabilities that you have, how are you different from the competition? When you talk to clients and they say well what about this vendor or that vendor, how do you respond? >> So let me talk about one of the research technologies that we're launching as part of this called CADS, Cognitive Assistant for Data Scientists. This is a feature where essentially, it takes the complexity out of building a model where you tell it, or you give it the algorithms you want to work with and the CADS assistant basically returns which one is the best which one performs the best. Now, all of a sudden you have the best model to use without having to go and spend, potentially weeks, on figuring out which one that is. So that's a huge value proposition. >> So automating the choice of the algorithm, an algorithm to choose the algorithm. what have you found in terms of it's level of accuracy in terms of the best fit? >> Actually it works really well. And in fact we have a live demo that we'll be doing today, where it shows CADS coming back with a 90% accurate model in terms of the data that we're feeding it and outcome it will give you in terms of what model to use. It works really well. >> Choosing an algorithm is not like choosing a programming language right, this bias if I like Scala or R or whatever, Java, Python okay fine, I've got skill sets associated with that. Algorithm choice is one that's more scientific, I guess? >> It is more scientific, it's based on the algorithm, the statistical algorithm and the selection of the algorithm or the model itself is a huge deal because that's where you're going to drive your business. If you're offering a new service that's where you're providing that solution from, so it has to be the right algorithm the right model so that you can build that more efficiently. >> What are you seeing as the big barriers to customer adopting machine learning? >> I think everybody, I mean it's the hottest thing around right now, everybody wants machine learning it's great, it's a huge buzz. The hardest thing is they know they want it, but don't really know how to apply it into their own environment, or they think they don't have the right skills. So, that actually one of the things that we're going after, to be able to enable them to do that. We're for example working on building different industry-based examples to showcase here's how you would use it in your environment. So last year when we did the Watson data platform we did a retail example, now today we're doing a finance example, a churn example with customers potentially churning and leaving a bank. So we're looking at all those different scenarios, and then also we're creating hubs, locations we're launching today also, announcing today, actually Dinesh will be doing that. There is a hub in Silicon Valley where it would allow customers to come in and work with us and we help them figure out how they can leverage machine learning. It is a great way to interact with our customers and be able to do that. >> So Steve nirvana is, and you gave that example, the retail example in September, when you launched Watson Data Platform, the nirvana in this world is you can use data, and maybe put in an offer, or save a patients life or effect an outcome in real time. So the retail example was just that. If I recall, you were making an offer real-time it was very fast, live demo it wasn't just a fakey. The example on churn, is the outcome is to effect that customer's decisions so that they don't leave? Is that? >> Yes, pretty much, Essentially what we are looking at is , we're using live data, we're using social media data bringing in Twitter sentiment about a particular individual for example, and try to predict if this customer, if this user is happy with the service that they are getting or not. So for example, people will go and socialize, oh I went to this bank and I hated this experience, or they really got me upset or whatever. Bringing that data from Twitter, so open data and merging it with the bank's data, banks have a lot of data they can leverage and monetize. And then making an assessment using machine learning to predict is this customer going to leave me or not? What probability do they have that they are going to leave me or not based on the machine learning model. The example or scenario we are using now, if we think they are going to leave us, we're going to make special offers to them. It's a way to enhance your service for those customers. So that they don't leave you. >> So operationalizing that would be a call center has some kind on dashboard that says red, green, yellow, boom heres an offer that you should make, and that's done in near real time. In fact, real time is before you lose the customer. That's as good a definition as anything else. >> But it's actually real-time, and when we call it the scoring of the data, so as the data transaction is coming in, you can actually make that assessment in real time, it's called in-transaction scoring where you can make that right on the fly and be able to determine is this customer at risk or not. And then be able to make smarter decisions to that service you are providing on whether you want to offer something better. >> So is the primary use case for this those streams those areas I'm getting you know, whether it be, you mentioned Twitter data, maybe IoT, you're getting can we point machine learning at just archives of data and things written historically or is it mostly the streams? >> It's both of course and machine learning is based on historical data right and that's hot the models are built. The more accurate or more data you have on historical data, the more accurate that you picked the right model and you'll get the better predictition of what's going to happen next time. So it's exactly, it's both. >> How are you helping customers with that initial fit? My understanding is how big of a data set do you need, Do I have enough to really model where I have, how do you help customers work through that? >> So my opinion is obvious to a certain extent, the more data you have as your sample set, the more accurate your model is going to be. So if we have one that's too small, your prediction is going to be inaccurate. It really depends on the scenario, it depends on how many features or the fields you have you're looking at within your dataset. It depends on many things, and it's variable depending on the scenario, but in general you want to have a good chunk of historical data that you can build expertise on right. >> So you've worked on both the Watson Services in the public cloud and now this private cloud, is there any differentiation or do you see significant use case different between those two or is it just kind of where the data lives and we're going to do similar activities there. >> So it is similar. At the end of the day, we're trying to provide similar products on both public cloud and private cloud. But for this specific case, we're launching it on mainframe that's a different angle at this. But we know that's where the biggest banks, the insurance companies, the biggest retailers in the world are, and that's where the biggest transactions are running and we really want to help them leverage machine learning and get their services to the next level. I think it's going to be a huge differentiator for them. >> Steve, you gave an example before of Twitter sentiment data. How would that fit in to this announcement. So I've got this ML on Z and I what API into the twitter data? How does that sort of all get adjusted and consolidated? >> So we allow hooks to be able to access data from different sources, bring in data. That is part of the ingest process. Then once you have that data there into data frames into the machine learning product, now you're feeding into a statistical algorithm to figure out what the best prediction is going to be, and the best model's going to be. >> I have a slide that you guys are sharing on the data scientist workflow. It starts with ingestion, selection, preparation, generation, transform, model. It's a complex set of tasks, and typically historically, at least in the last fIve or six years, different tools to de each of those. And not just different tools, multiples of different tools. That you had to cobble together. If I understand it correctly the Watson Data Platform was designed to really consolidate that and simplify that, provide collaboration tools for different personas, so my question is this. Because you were involved in that product as well. And I was excited about it when I saw it, I talked to people about it, sometimes I hear the criticism of well IBM just took a bunch of legacy products threw them together, threw and abstraction layer on top and is now going to wrap a bunch of services around it. Is that true? >> Absolutely not. Actually, you may have heard a while back IBM had made a big shift into design first design methodology. So we started with the Watson Data Platform, the Data Science Experience, they started with design first approach. We looked at this, we said what do we want the experience to be, for which persona do we want to target. Then we understood what we wanted the experience to be and then we leverage IBM analytics portfolio to be able to feed in and provide and integrate those services together to fit into that experience. So, its not a dumping ground for, I'll take this product, it's part of Watson Data Platform, not at all the case. It was the design first, and then integrate for that experience. >> OK, but there are some so-called legacy products in there, but you're saying you picked the ones that were relevant and then was there additional design done? >> There was a lot of work involved to take them from a traditional product, to be able to componentize, create a micro service architecture, I mean the whole works to be able to redesign it and fit into this new experience. >> So microservices architecture, runs on cloud, I think it only runs on cloud today right? >> Correct, correct. >> OK, maybe roadmap without getting too specific. What should we be paying attention to in the future? >> Right now we're doing our first release. Definitely we want to target any platform behind the firewall. So we don't have specific dates, but now we started with machine learning on a mainframe and we want to be able to target the other platforms behind the firewall and the private cloud environment. Definitely we should be looking at that. Our goal is to make, I talked about the feedback loop a little bit, so that is essentially once you deploy the model we actually look at that model you could schedule in a valuation, automatically, within the machine learning product. To be able to say, this model is still good enough. And if it's not we automatically flag it, and we look at the retraining process and redeployment process to make sure you always have the most up to date model. So this is truly machine learning where it requires very little to no intervention from a human. We're going to continue down that path and continue that automation in providing those capabilities so there's a bigger roadmap, there's a lot of things we're looking at. >> We've sort of looked at our big data analyst George Gilbert has talked about you had batch and you had interactive, not the sort of emergent workload is this continuous, streaming data. How do you see the adoption. First of all, is it a valid assertion? That there is a new class of workload, and then how do you see that adoption occurring? Is it going to be a dominant force over the next 10 years? >> Yeah, I think so. Like I said there is a huge buzz around machine learning in general and artificial intelligence, deep learning, all of these terms you hear about. I think as users and customers get more comfortable with understanding how they're going to leverage this in their enterprise. This real-time streaming of data and being able to do analytics on the fly and machine learning on the fly. It's a big deal and it will really helps them be more competitive in their own space with the services we're providing. >> OK Steve, thanks very much for coming on The CUBE. We'll give you the last word. The event, very intimate event a lot of customers coming in very shortly here in just a couple of hours. Give us the bumper sticker. >> All of that's very exciting, we're very excited, this is a big deal for us, that's why whenever IBM does a signature moment it's a big deal for us and we got something cool to talk about, we're very excited about that. Lot's of clients coming so there's an entire session this afternoon, which will be live streamed as well. So it's great, I think we have a differentiating product and we're already getting that feedback from our customers. >> Well congratulations, I love the cadence that you're on. We saw some announcements in September, we're here in February, I expect we're going to see more innovation coming out of your labs in Toronto, and cross IBM so thank you very much for coming on The CUBE. >> Thank you. >> You're welcome OK keep it right there everybody, we'll be back with our next guest right after this short break. This is The CUBE we're live from New York City. (energetic music)