>> [Announcer] Live from New York, it's theCube, 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 the Waldorf Astoria, everybody. This is theCube, the worldwide leader in live tech coverage. We're covering the IBM Machine Learning announcement. IBM bringing machine learning to its zMainframe, its private cloud. Dinesh Nirmel is here. He's the Vice President of Analytics at IBM and a Cube alum. Dinesh, good to see you again. >> Good to see you, Dave. >> So let's talk about ML. So we went through the big data, the data lake, the data swamp, all this stuff with the dupe. And now we're talking about machine learning and deep learning and AI and cognitive. Is it same wine, new bottle? Or is it an evolution of data and analytics? >> Good. So, Dave, let's talk about machine learning. Right. When I look at machine learning, there's three pillars. The first one is the product. I mean, you got to have a product, right. And you got to have a different shared set of functions and features available for customers to build models. For example, Canvas. I mean, those are table stakes. You got to have a set of algorithms available. So that's the product piece. >> [Dave] Uh huh. >> But then there's the process, the process of taking that model that you built in a notebook and being able to operationalize it. Meaning able to deploy it. That is, you know, I was talking to one of the customers today, and he was saying, "Machine learning is 20% fun and 80% elbow grease." Because that operationalizing of that model is not easy. Although they make it sound very simple, it's not. So if you take a banking, enterprise banking example, right? You build a model in the notebook. Some data sense build it. Now you have to take that and put it into your infrastructure or production environment, which has been there for decades. So you could have a third party software that you cannot change. You could have a set of rigid rules that already is there. You could have applications that was written in the 70's and 80's that nobody want to touch. How do you all of a sudden take the model and infuse in there? It's not easy. And so that is a tremendous amount of work. >> [Dave] Okay. >> The third pillar is the people or the expertise or the experience, the skills that needs to come through, right. So the product is one. The process of operationalizing and getting it into your production environment is another piece. And then the people is the third one. So when I look at machine learning, right. Those are three key pillars that you need to have to have a successful, you know, experience of machine learning. >> Okay, let's unpack that a little bit. Let's start with the differentiation. You mentioned Canvas, but talk about IBM specifically. >> [Dinesh] Right. What's so great about IBM? What's the differentiation? >> Right, exactly. Really good point. So we have been in the productive side for a very long time, right. I mean, it's not like we are coming into ML or AI or cognitive yesterday. We have been in that space for a very long time. We have SPSS predictive analytics available. So even if you look from all three pillars, what we are doing is we are, from a product perspective, we are bringing in the product where we are giving a choice or a flexibility to use the language you want. So there are customers who only want to use R. They are religious R users. They don't want to hear about anything else. There are customers who want to use Python, you know. They don't want to use anything else. So how do we give that choice of languages to our customers to say use any language you want. Or execution engines, right? Some folks want to use Park as execution engine. Some folks want to use R or Python, so we give that choice. Then you talked about Canvas. There are folks who want to use the GUI portion of the Canvas or a modeler to build models, or there are, you know, tekkie guys that we'll approach who want to use notebook. So how do you give that choice? So it becomes kind of like a freedom or a flexibility or a choice that we provide, so that's the product piece, right? We do that. Then the other piece is productivity. So one of the customers, the CTO of (mumbles) TV's going to come on stage with me during the main session, talk about how collaboration helped from an IBM machine learning perspective because their data scientists are sitting in New York City, our data scientists who are working with them are sitting in San Jose, California. And they were real time collaborating using notebooks in our ML projects where they can see the real time. What changes their data scientists are making. They can slack messages between each other. And that collaborative piece is what really helped us. So collaboration is one. Right from a productivity piece. We introduced something called Feedback Loop, whereby which your model can get trained. So today, you deploy a model. It could lose the score, and it could get degraded over time. Then you have to take it off-line and re-train, right? What we have done is like we introduced the Feedback Loops, so when you deploy your model, we give you two endpoints. The first endpoint is, basically, a URI, for you to plug-in your application when you, you know, run your application able call the scoring API. The second endpoint is this feedback endpoint, where you can choose to re-train the model. If you want three hours, if you want it to be six hours, you can do that. So we bring that flexibility, we bring that productivity into it. Then, the management of the models, right? How do we make sure that once you develop the model, you deploy the model. There's a life cycle involved there. How do you make sure that we enable, give you the tools to manage the model? So when you talk about differentiation, right? We are bringing differentiation on all three pillars. From a product perspective, with all the things I mentioned. From a deployment perspective. How do we make sure we have different choices of deployment, whether it's streaming, whether it's realtime, whether it's batch. You can do deployment, right? The Feedback Loop is another one. Once you deployed, how do we keep re-training it. And the last piece I talked about is the expertise or the people, right? So we are today announcing IBM Machine Learning Hub, which will become one place where our customers can go, ask questions, get education sessions, get training, right? Work together to build models. I'll give you an example, that although we are announcing hub, the IBM Machine Learning Hub today, we have been working with America First Credit Union for the last month or so. They approached us and said, you know, their underwriting takes a long time. All the knowledge is embedded in 15 to 20 human beings. And they want to make sure a machine should be able to absorb that knowledge and make that decision in minutes. So it takes hours or days. >> [Dave] So, Stu, before you jump in, so I got, put the portfolio. You know, you mentioned SPSS, expertise, choice. The collaboration, which I think you really stressed at the announcement last fall. The management of the models, so you can continuously improve it. >> Right. >> And then this knowledge base, what you're calling the hub. And I could argue, I guess, that if I take any one of those individual pieces, there, some of your competitors have them. Your argument would be it's all there. >> It all comes together, right? And you have to make sure that all three pillars come together. And customers see great value when you have that. >> Dinesh, customers today are used to kind of the deployment model on the public cloud, which is, "I want to activate a new service," you know. I just activate it, and it's there. When I think about private cloud environments, private clouds are operationally faster, but it's usually not miniature hours. It's usually more like months to deploy projects, which is still better than, you know, kind of, I think, before big data, it was, you know, oh, okay, 18 months to see if it works, and let's bring that down to, you know, a couple of months. Can you walk us through what does, you know, a customer today and says, "Great, I love this approach. "How long does it take?" You know, what's kind of the project life cycle of this? And how long will it take them to play around and pull some of these levers before they're, you know, getting productivity out of it? >> Right. So, really good questions, Stu. So let me back one step. So, in private cloud, we are going, we have new initiative called Download and Go, where our goal is to have our desktop products be able to install on your personal desktop in less than five clicks, in less than fifteen minutes. That's the goal. So the other day, you know, the team told me it's ready. That the first product is ready where you can go less than five clicks, fifteen minutes. I said the real test is I'm going to bring my son, who's five years old. Can he install it, and if he can install it, you know, we are good. And he did it. And I have a video to prove it, you know. So after the show, I will show you because and that's, when you talk about, you know, in the private cloud side, or the on-premise side, it has been a long project cycle. What we want is like you should be able to take our product, install it, and get the experience in minutes. That's the goal. And when you talk about private cloud and public cloud, another differentiating factor is that now you get the strength of IBM public cloud combined with the private cloud, so you could, you know, train your model in public cloud, and score on private cloud. You have the same experience. Not many folks, not many competitors can offer that, right? So that's another . .. >> [Stu] So if I get that right. If I as a customer have played around with the machine learning in Bluemix, I'm going to have a similar look, feel, API. >> Exactly the same, so what you have in Bluemix, right? I mean, so you have the Watson in Bluemix, which, you know, has deep learning, machine learning--all those capabilities. What we have done is we have done, is like, we have extracted the core capabilities of Watson on private cloud, and it's IBM Machine Learning. But the experience is the same. >> I want to talk about this notion of operationalizing analytics. And it ties, to me anyway, it ties into transformation. You mentioned going from Notebook to actually being able to embed analytics in workflow of the business. Can you double click on that a little bit, and maybe give some examples of how that has helped companies transform? >> Right. So when I talk about operationalizing, when you look at machine learning, right? You have all the way from data, which is the most critical piece, to building or deploying the model. A lot of times, data itself is not clean. I'll give you an example, right. So >> OSYX. >> Yeah. And when we are working with an insurance company, for example, the data that comes in. For example, if you just take gender, a lot of times the values are null. So we have to build another model to figure out if it's male or female, right? So in this case, for example, we have to say somebody has done a prostate exam. Obviously, he's a male. You know, we figured that. Or has a gynocology exam. It's a female. So we have to, you know, there's a lot of work just to get that data cleansed. So that's where I mentioned it's, you know, machine learning is 20% fun, 80% elbow grease because it's a lot of grease there that you need to make sure that you cleanse the data. Get that right. That's the shaping piece of it. Then, comes the building the model, right. And then, once you build the model on that data comes the operationalization of that model, which in itself is huge because how do you make sure that you infuse that model into your current infrastructure, which is where a lot of skill set, a lot of experience, and a lot of knowledge that comes in because you want to make sure, unless you are a start-up, right? You already have applications and programs and third-party vendors applications worth running for years, or decades, for that matter. So, yeah, so that's operationalization's a huge piece. Cleansing of the data is a huge piece. Getting the model right is another piece. >> And simplifying the whole process. I think about, I got to ingest the data. I've now got to, you know, play with it, explore. I've got to process it. And I've got to serve it to some, you know, some business need or application. And typically, those are separate processes, separate tools, maybe different personas that are doing that. Am I correct that your announcement in the Fall addressed that workflow. How is it being, you know, deployed and adopted in the field? How is it, again back to transformation, are you seeing that people are actually transforming their analytics processes and ultimately creating outcomes that they expect? >> Huge. So good point. We announced data science experience in the Fall. And the customers that who are going to speak with us today on stage, are the customers who have been using that. So, for example, if you take AFCU, America First Credit Union, they worked with us. In two weeks, you know, talk about transformation, we were able to absorb the knowledge of their underwriters. You know, what (mumbles) is in. Build that, get that features. And was able to build a model in two weeks. And the model is predicting 90%, with 90% accuracy. That's what early tests are showing. >> [Dave] And you say that was in a couple of weeks. You were, you developed that model. >> Yeah, yeah, right. So when we talk about transformation, right? We couldn't have done that a few years ago. We have transformed where the different personas can collaborate with each other, and that's a collaboration piece I talked about. Real time. Be able to build a model, and put it in the test to see what kind of benefits they're getting. >> And you've obviously got edge cases where people get really sophisticated, but, you know, we were sort of talking off camera, and you know like the 80/20 rule, or maybe it's the 90/10. You say most use cases can be, you know, solved with regression and classification. Can you talk about that a little more? >> So, so when we talk about machine learning, right? To me, I would say 90% of it is regression or classification. I mean there are edge case of our clustering and all those things. But linear regression or a classification can solve most of the, most of our customers problems, right? So whether it's fraud detection. Or whether it's underwriting the loan. Or whether you're trying to determine the sentiment analysis. I mean, you can kind of classify or do regression on it. So I would say that 90% of the cases can be covered, but like I said, most of the work is not about picking the right algorithm, but it's also about cleansing the data. Picking the algorithm, then comes building the model. Then comes deployment or operationalizing the model. So there's a step process that's involved, and each step involves some amount of work. So if I could make one more point on the technology and the transformation we have done. So even with picking the right algorithm, we automated, so you as a data scientist don't need to, you know, come in and figure out if I have 50 classifiers and each classifier has four parameters. That's 200 different combinations. Even if you take one hour on each combination, that's 200 hours or nine days that takes you to pick the right combination. What we have done is like in IBM Machine Learning we have something called cognitive assistance for data science, which will help you pick the right combination in minutes instead of days. >> So I can see how regression scales, and in the example you gave of classification, I can see how that scales. If you've got a, you know, fixed classification or maybe 200 parameters, or whatever it is, that scales, what happens, how are people dealing with, sort of automating that classification as things change, as they, some kind of new disease or pattern pops up. How do they address that at scale? >> Good point. So as the data changes, the model needs to change, right? Because everything that model knows is based on the training data. Now, if the data has changed, the symptoms of cancer or any disease has changed, obviously, you have to retrain that model. And that's where I talk about the, where the feedback loop comes in, where we will automatically retrain the model based on the new data that's coming in. So you, as an end user, for example, don't need to worry about it because we will take care of that piece also. We will automate that, also. >> Okay, good. And you've got a session this afternoon with you said two clients, right? AFCU and Kaden dot TV, and you're on, let's see, at 2:55. >> Right. >> So you folks watching the live stream, check that out. I'll give you the last word, you know, what shall we expect to hear there. Show a little leg on your discussion this afternoon. >> Right. So, obviously, I'm going to talk about the different shading factors, what we are delivering IBM Machine Learning, right? And I covered some of it. There's going to be much more. We are going to focus on how we are making freedom or flexibility available. How are we going to do productivity, right? Gains for our data scientists and developers. We are going to talk about trust, you know, the trust of data that we are bringing in. Then I'm going to bring the customers in and talk about their experience, right? We are delivering a product, but we already have customers using it, so I want them to come on stage and share the experiences of, you know, it's one thing you hear about that from us, but it's another thing that customers come and talk about it. So, and the last but not least is we are going to announce our first release of IBM Machine Learning on Z because if you look at 90% of the transactional data, today, it runs through Z, so they don't have to off-load the data to do analytics on it. We will make machine learning available, so you can do training and scoring right there on Z for your real time analytics, so. >> Right. Extending that theme that we talked about earlier, Stu, bringing analytics and transactions together, which is a big theme of the Z 13 announcement two years ago. Now you're seeing, you know, machine learning coming on Z. The live stream starts at 2 o'clock. Silicon Angle dot com had an article up on the site this morning from Maria Doucher on the IBM announcement, so check that out. Dinesh, thanks very much for coming back on theCube. Really appreciate it, and good luck today. >> Thank you. >> All right. Keep it right there, buddy. We'll be back with our next guest. This is theCube. We're live from the Waldorf Astoria for the IBM Machine Learning Event announcement. Right back.