>> (Narrator) Live from Madrid, Spain. It's The Cube covering HP Discover Madrid 2017, brought to you by Hewlett Packard Enterprise. >> We're back at HPE Discover Madrid 2017. This is The Cube, the leader in live tech coverage. My name is Dave Vellante and I'm with my co-host for the week, Peter Burris. Cat Graves is here, she's a research scientist at Hewlett Packard Enterprises. And she's joined by Natalia Vassilieva. Cube alum, senior research manager at HPE. Both with the labs in Palo Alto. Thanks so much for coming on The Cube. >> Thank you for having us. >> You're welcome. So for decades this industry has marched to the cadence of Moore's Law, bowed down to Moore's Law, been subservient to Moore's Law. But that's changing, isn't it? >> Absolutely. >> What's going on? >> I can tell Moore's Law is changing. So we can't increase the number, of course, on the same chip and have the same space. We can't increase the density of the computer today. And from the software perspective, we need to analyze more and more data. We are now marching calls into the area of artificial intelligence when we need to train larger and larger models, we need more and more compute for that. And the only possible way today to speed up the training of those modules, to actually enable the AI, is to scale out. Because we can't put more cores on the chip. So we try to use more chips together But then communication bottlenecks come in. So we can't efficiently use all of those chips. So for us on the software side, on the part of people who works how to speed up the training, how to speed up the implementation of the algorithms, and the work of those algorithms, that's a problem. And that's where Cat can help us because she's working on a new hardware which will overcome those troubles. >> Yeah, so in our lab what we do is try and think of new ways of doing computation but also doing the computations that really matter. You know, what are the bottlenecks for the applications that Natalia is working on that are really preventing the performance from accelerating? Again exponentially like Moore's Law, right? We'd like to return to Moore's Law where we're in that sort of exponential growth in terms of what compute is really capable of. And so what we're doing in labs is leveraging novel devices so, you've heard of memristor in the past probably. But instead of using memristor for computer memory, non volatile memory for persistent memory driven computer systems, we're using these devices instead for doing computation itself in the analog domain. So one of our first target applications, and target core computations that we're going after is matrix multiplication. And that is a fundamental mathematical building block for a lot of different machine learning, deep learning, signal processing, you kind of name it, it's pretty broad in terms of where it's used today. >> So Dr. Tom Bradicich was talking about the dot product, and it sounds like it's related. Matrix multiplications, suddenly I start breaking out in hives but is that kind of related? >> That's exactly what it is. So, if you remember your linear algebra in college, a dot product is exactly a matrix multiplication. It's the dot in between the vector and the matrix. The two itself, so exactly right. Our hardware prototype is called the dot product engine. It's just cranking out those matrix multiplications. >> And can you explain how that addresses the problem that we're trying to solve with respect to Moore's Law? >> Yeah, let me. You mentioned the problem with Moore's Law. From me as a software person, the end of Moore's Law is a bad thing because I can't increase their compute power anymore on the single chip. But for Cat it's a good thing because it forced her to think what's unconventional. >> (Cat) It's an opportunity. >> It's an opportunity! >> It forced her to think, what are unconventional devices which she can come up with? And we also have to mention they understand that general purpose computing is not always a solution. Sometimes if you want to speed up the thing, you need to come up with a device which is designed specifically for the type of computation which you care about. And for machine learning technification, again as I've mentioned, these matrix-matrix multiplications matrix-vector multiplications, these are the core of it. Today if you want to do those AI type applications, you spend roughly 90% of the time doing exactly that computation. So if we can come up with a more power efficient and a more effective way of doing that, that will really help us, and that's what dot product engine is solving. >> Yes, an example some of our colleagues did in architectural work. Sort of taking the dot product engine as the core, and then saying, okay if I designed a computer architecture specifically for doing convolutional neural networks. So image classification, these kinds of applications. If I built this architecture, how would it perform? And how would it compare to GPUs? And we're seeing 10 to 100 X speed up over GPUs. And even 15 X speed up over if you had a custom-built, state of the art specialized digital Asic. Even comparing to the best that we can do today, we are seeing this potential for a huge amount of speed up and also energy savings as well. >> So follow up on that, if I may. So you're saying these alternative processors like GPUs, FGPAs, custom Asics, can I infer from that that that is a stop-gap architecturally, in your mind? Because you're seeing these alternative processors pop up all over the place. >> (Cat) Yes. >> Is that a fair assertion? >> I think that recent trends are obviously favoring a return to specialized hardware. >> (Dave) Yeah, for sure. Just look at INVIDIA, it's exploding. >> I think it really depends on the application and you have to look at what the requirements are. Especially in terms of where there's a lot of power limitations, right, GPUs have become a little bit tricky. So there's a lot of interest in the automotive industry, space, robotics, for more low power but still very high performance, highly efficient computation. >> Many years ago when I was actually thinking about doing computer science and realized pretty quickly that I didn't have the brain power to get there. But I remember thinking in terms of there's three ways of improving performance. You can do it architecturally, what do you do with an instruction? You can do it organizationally, how do you fit the various elements together? You can do it with technology, which is what's the clock speed, what's the underlying substrate? Moore's Law is focused on the technology. Risk, for example, focused on architecture. FPGAs, arm processors, GPUs focus on architecture. What we're talking about to get back to that doubling the performance every 18 months from a computing standpoint not just a chip standpoint, now we're talking about revealing and liberating, I presume, some of the organization elements. Ways of thinking about how to put these things together. So even if we can't get improvements that we've gotten out of technology, we can start getting more performance out of new architectures. But organizing how everything works together. And make it so that the software doesn't have to know, or the developer, doesn't have to know everything about the organization. Am I kind of getting there with this? >> Yes, I think you are right. And if we are talking about some of the architectural challenges of today's processors, not only we can't increase the power of a single device today, but even if we increase the power of a single device, then the challenge would be how do you bring the data fast enough to that device? So we will have problems with feeding that device. And again, what dot product engine does, it does computations in memory, inside. So you limit the number of data transfers between different chips and you don't face the problem of feeding their computation thing. >> So similar same technology, different architecture, and using a new organization to take advantage of that architecture. The dot product engine being kind of that combination. >> I would say that even technology is different. >> Yeah, my view of it we're actually thinking about it holistically. We have in labs software working with architects. >> I mean it's not just a clock speed issue. >> It's not just a clock speed issue. It's thinking about what computations actually matter, which ones you're actually doing, and how to perform them in different ways. And so one of the great things as well with the dot product engine and these kind of new computation accelerators, is with something like the memory driven computing architecture. We have now an ecosystem that is really favoring accelerators and encouraging the development of these specialized hardware pieces that can kind of slot in in the same architecture that can scale also in size. >> And you invoke that resource in an automated way, presumably. >> Yeah, exactly. >> What's the secret sauce behind that? Is that software that does that or an algorithm that chooses the algorithm? >> A gen z. >> A gen z's underlying protocol is to make the device talk to the data. But at the end of the system software, it's algorithms also which will make a decision at every particular point which compute device I should use to do a particular task. With memory driven computing, if all my data sits in the shared pool of memory and I have different heterogeneous compute devices, being able to see that data and to talk to that data, then it's up to the system management software to allocate the execution of a particular task to the device which does that the best. In a more power efficient way, in the fastest way, and everybody wins. >> So as a software person, you now with memory driven computing have been thinking about developing software in a completely different way. Is that correct? >> (Natalia) Yeah. You're not thinking about going through I/O stack anymore and waiting for a mechanical device and doing other things? >> It's not only the I/O stack. >> As I mentioned today, the only possibility for us to decrease the time of processing for the algorithms is to scale out. That means that I need to take into account the locality of the data. It's not only when you distribute the computation across multiple nodes, even if we have some number based which is we have different sockets in a single system. With local memory and the memory which is remote to that socket but which is local to another socket. Today as a software programmer, as a developer, I need to take into account where my data sits. Because I know in order to accept the data on a local memory it'll take me 100 seconds to accept my data. In the remote socket, it will take me longer. So when I developed the algorithm in order to prevent my computational course to stall and to wait for the data, I need to schedule that very carefully. With memory driven computing, giving an assumption that, again, all memory not only in the single pool, but it's also evenly accessible from every compute device. I don't need to care about that anymore. And you can't even imagine such a relief it is! (laughs) It makes our life so much easier. >> Yeah, because you're spending a lot of time previously trying to optimize your code >> Yes for that factor of the locality of the data. How much of your time was spent doing that menial task? >> Years! In the beginning of Moore's Law and the beginning of the traditional architectures, if you turn to the HPC applications, every HPC application device today needs to take care of data locality. >> And you hear about when a new GPU comes out or even just a slightly new generation. They have to take months to even redesign their algorithm to tune it to that specific hardware, right? And that's the same company, maybe even the same product sort of path lined. But just because that architecture has slightly changed changes exactly what Natalia is talking about. >> I'm interested in switching subjects here. I'd love to spend a minute on women in tech. How you guys got into this role. You're both obviously strong in math, computer backgrounds. But give us a little flavor of your background, Cat, and then, Natalia, you as well. >> Me or you? >> You start. >> Hm, I don't know. I was always interested in a lot of different things. I kind of wanted to study and do everything. And I got to the point in college where physics was something that still fascinated me. I felt like I didn't know nearly enough. I felt like there was still so much to learn and it was constantly challenging me. So I decided to pursue my Ph.D in that, and it's never boring, and you're always learning something new. Yeah, I don't know. >> Okay, and that led to a career in technology development. >> Yeah, and I actually did my Ph.D in kind of something that was pretty different. But towards the end of it, decided I really enjoyed research and was just always inspired by it. But I wanted to do that research on projects that I felt like might have more of an impact. And particularly an impact in my lifetime. My Ph.D work was kind of something that I knew would never actually be implemented in, maybe a couple hundred years or something we might get to that point. So there's not too many places, at least in my field in hardware, where you can be doing what feels like very cutting edge research, but be doing it in a place where you can see your ideas and your work be implemented. That's something that led me to labs. >> And Natalia, what's your passion? How did you arrive here? >> As a kid I always liked different math puzzles. I was into math and pretty soon it became obvious that I like solving those math problems much more than writing about anything. I think in middle school there was the first class on programming, I went right into that. And then the teacher told me that I should probably go to a specialized school and that led me to physics and mathematics lyceum and then mathematical department at the university so it was pretty straightforward for me since then. >> You're both obviously very comfortable in this role, extremely knowledgeable. You seem like great leaders. Why do you feel that more women don't pursue a career in technology. Do you have these discussions amongst yourselves? Is this something that you even think about? >> I think it starts very early. For me, both my parents are scientists, and so always had books around the house. Always was encouraged to think and pursue that path, and be curious. I think its something that happens at a very young age. And various academic institutions have done studies and shown when they do certain things, its surmountable. Carnegie Mellon has a very nice program for this, where they went for the percentage of women in their CS program went from 10% to 40% in five years. And there were a couple of strategies that they implemented. I'm not gonna get all of them, but one was peer to peer mentoring, when the freshmen came in, pairing them with a senior, feeling like you're not the only one doing what you're doing, or interested in what you're doing. It's like anything human, you want to feel like you belong and can relate to your group. So I think, yeah. (laughs) >> Let's have a last word. >> On that topic? >> Yeah sure, or any topic. But yes, I'm very interested in this topic because less than 20% of the tech business is women. Its 50W% of the population. >> I think for me its not the percentage which matters Just don't stay in the way of those who's interested in that. And give equal opportunities to everybody. And yes, the environment from the very childhood should be the proper one. >> Do you feel like the industry gives women equal opportunity? >> For me, my feeling would be yes. You also need to understand >> Because of your experience Because of my experience, but I also originally came from Russia, was born in St. Petersburg, and I do believe that ex-Soviet Union countries has much better history in that. Because the Soviet Union, we don't have man and woman. We have comrades. And after the Second World War, there was women who took all hard jobs. And we used to get moms at work. All moms of all my peers have been working. My mom was an engineer, my dad is an engineer. From that, there is no perception that the woman should stay at home, or the woman is taking care of kids. There is less of that. >> Interesting. So for me, yes. Now I think that industry going that direction. And that's right. >> Instructive, great. Well, listen, thanks very much for coming on the Cube. >> Sure. >> Sharing the stories, and good luck in lab, wherever you may end up. >> Thank you. >> Good to see you. >> Thank you very much. >> Alright, keep it right there everybody. We'll be back with our next guest, Dave Vallante for Peter Buress. We're live from Madrid, 2017, HPE Discover. This is the Cube.

>> Announcer: Live from Las Vegas, it's the CUBE! Covering HPE Discover 2017. Brought to you by Hewlett Packard Enterprise. >> Hey, welcome back, everyone. We are live here in Las Vegas for SiliconANGLE Media's CUBE exclusive coverage of HPE Discover 2017. I'm John Furrier, my co-host, Dave Vellante. Our next guest is Kirk Bresniker, fellow and VP chief architect of Hewlett Packard Labs, and Natalia Vassilieva, senior research manager, Hewlett Packard Labs. Did I get that right? >> Yes! >> John: Okay, welcome to theCUBE, good to see you. >> Thank you. >> Thanks for coming on, really appreciate you guys coming on. One of the things I'm most excited about here at HPE Discover is, always like to geek out on the Hewlett Packard Labs booth, which is right behind us. If you go to the wide shot, you can see the awesome display. But there's some two things in there that I love. The Machine is in there, which I love the new branding, by the way, love that pyramid coming out of the, the phoenix rising out of the ashes. And also Memristor, really game-changing. This is underlying technology, but what's powering the business trends out there that you guys are kind of doing the R&D on is AI, and machine learning, and software's changing. What's your thoughts as you look at the labs, you look out on the landscape, and you do the R&D, what's the vision? >> One of the things what is so fascinating about the transitional period we're in. We look at the kind of technologies that we've had 'til date, and certainly spent a whole part of my career on, and yet all these technologies that we've had so far, they're all kind of getting about as good as they're going to get. You know, the Moore's Law semiconductor process steps, general-purpose operating systems, general-purpose microprocessors, they've had fantastic productivity growth, but they all have a natural life cycle, and they're all maturing. And part of The Machine research program has been, what do we think is coming next? And really, what's informing us as what we have to set as the goals are the kinds of applications that we expect. And those are data-intensive applications, not just petabytes, exabytes, but zettabytes. Tens of zettabytes, hundreds of zettabytes of data out there in all those sensors out there in the world. And when you want to analyze that data, you can't just push it back to the individual human, you need to employ machine learning algorithms to go through that data to call out and find those needles in those increasingly enormous haystacks, so that you can get that key correlation. And when you don't have to reduce and redact and summarize data, when you can operate on the data at that intelligent edge, you're going to find those correlations, and that machine learning algorithm is going to be that unbiased and unblinking eye that's going to find that key relationship that'll really have a transformational effect. >> I think that's interesting. I'd like to ask you just one follow-up question on that, because I think, you know, it reminds me back when I was in my youth, around packets, and you'd get the buffer, and the speeds, and feeds. At some point there was a wire speed capability. Hey, packets are moving, and you can do all this analysis at wire speed. What you're getting at is, data processing at the speed of, as fast as the data's coming in and out. Is that, if I get that right, is that kind of where you're going with this? Because if you have more data coming, potentially an infinite amount of data coming in, the data speed is going to be so high-velocity, how do you know what a needle looks like? >> I think that's a key, and that's why the research Natalia's been doing is so fundamental, is that we need to be able to process that incredible amount of information and be able to afford to do it. And the way that you will not be able to have it scale is if you have to take that data, compress it, reduce it, select it down because of some pre-determined decision you've made, transmit it to a centralized location, do the analysis there, then send back the action commands. Now, we need that cycle of intelligence measurement, analysis and action to be microseconds. And that means it needs to happen at the intelligent edge. I think that's where the understanding of how machine learning algorithms, that you don't program, you train, so that they can work off of this enormous amount of data, they voraciously consume the data, and produce insights. That's where machine learning will be the key. >> Natalia, tell us about your research on this area. Curious. Your thoughts. >> We started to look at existing machine learning algorithms, and whether their limiting factors today in the infrastructure which don't allow to progress the machine learning algorithms fast enough. So, one of the recent advances in AI is appearance, or revival, of those artificial neural networks. Deep learning. That's a very large hype around those types of algorithms. Every speech assistant which you get, Siri in your phone, Cortana, or whatever, Alexa by Amazon, all of them use deep learning to train speech recognition systems. If you go to Facebook and suddenly it starts you to propose to mark the faces of your friends, that the face detection, face recognition, also that was deep learning. So that's a revival of the old artificial neural networks. Today we are capable to train byte-light enough models for those types of tasks, but we want to move forward. We want to be able to process larger volumes of data, to find more complicated patterns, and to do that, we need more compute power. Again, today, the only way how you can add more compute power to that, you scale out. So there is no compute device on Earth today which is capable to do all the computation. You need to have many of them interconnect together, and they all crunch numbers for the same problem. But at some point, the communication between those nodes becomes a bottleneck. So you need to let know laboring node what you achieved, and you can't scale out anymore. Adding yet another node to the cluster won't lead up to the reduction of the training time. With The Machine, when we have added the memory during computing architecture, when all data seeds in the same shared pool of memory, and when all computing nodes have an ability to talk to that memory. We don't have that limitation anymore. So for us, we are looking forward to deploy those algorithms on that type of architecture. We envision significant speedups in the training. And it will allow us to retrain the model on the new data, which is coming. To do not do training offline anymore. >> So how does this all work? When HP split into two companies, Hewlett Packard Labs went to HPE and HP Labs went to HP Ink. So what went where, and then, first question. Then second question is, how do you decide what to work on? >> I think in terms of how we organize ourselves, obviously, things that were around printing and personal systems went to HP Ink. Things that were around analytics, enterprise, hardware and research, went to Hewlett Packard Labs. The one thing that we both found equally interesting was security, 'cause obviously, personal systems, enterprise systems, we all need systems that are increasingly secure because of the advanced, persistent threats that are constantly assaulting everything from our personal systems up through enterprise and public infrastructure. So that's how we've organized ourselves. Now in terms of what we get to work on, you know, we're in an interesting position. I came to Labs three years ago. I used to be the chief technologist for the server global business unit. I was in the world of big D, tiny R. Natalia and the research team at Labs, they were out there looking out five, 10, 15, or 20 years. Huge R, and then we would meet together occasionally. I think one of the things that's happened with our machine advanced development and research program is, I came to Labs not to become a researcher, but to facilitate that communication to bring in the engineering, the supply chain team, that technical and production prowess, our experience from our services teams, who know how things actually get deployed in the real world. And I get to set them at the bench with Natalia, with the researchers, and I get to make everyone unhappy. Hopefully in equal amounts. That the development teams realize we're going to make some progress. We will end up with fantastic progress and products, both conventional systems as well as new systems, but it will be a while. We need to get through, that's why we had to build our prototype. To say, "No, we need a construction proof of these ideas." The same time, with Natalia and the research teams, they were always looking for that next horizon, that next question. Maybe we pulled them a little bit closer, got a little answers out of them rather than the next question. So I think that's part of what we've been doing at the Labs is understanding, how do we organize ourselves? How do we work with the Hewlett Packard Enterprise Pathfinder program, to find those little startups who need that extra piece of something that we can offer as that partnering community? It's really a novel approach for us to understand how do we fill that gap, how do we still have great conventional products, how do we enable breakthrough new category products, and have it in a timeframe that matters? >> So, much tighter connection between the R and the D. And then, okay, so when Natalia wants to initiate a project, or somebody wants Natalia to initiate a project around AI, how does that work? Do you say, "Okay, submit an idea," and then it goes through some kind of peer review? And then, how does it get funded? Take us through that. >> I think I'll give my perspective, I would love to hear what you have from your side. For me, it's always been organic. The ideas that we had on The Machine, for me, my little thread, one of thousands that's been brought in to get us to this point, started about 2003, where we were getting ready for, we're midway through Blade Systems C-class. A category-defining product. A absolute home run in defining what a Blade system was going to be. And we're partway through that, and you realize you got a success on your hands. You think, "Wow, nothing gets better than this!" Then it starts to worry, what if nothing gets better than this? And you start thinking about that next set of things. Now, I had some insights of my own, but when you're a technologist and you have an insight, that's a great feeling for a little while, and then it's a little bit of a lonely feeling. No one else understands this but me, and is it always going to be that way? And then you have to find that business opportunity. So that's where talking with our field teams, talking with our customers, coming to events like Discover, where you see business opportunities, and you realize, my ingenuity and this business opportunity are a match. Now, the third piece of that is someone who can say, a business leader, who can say, "You know what?" "Your ingenuity and that opportunity can meet "in a finite time with finite resources." "Let's do it." And really, that's what Meg and leadership team did with us on The Machine. >> Kirk, I want to shift gears and talk about the Memristor, because I think that's a showcase that everyone's talking about. Actually, The Machine has been talked about for many years now, but Memristor changes the game. It kind of goes back to old-school analog, right? We're talking about, you know, login, end-login kind of performance, that we've never seen before. So it's a completely different take on memory, and this kind of brings up your vision and the team's vision of memory-driven computing. Which, some are saying can scale machine learning. 'Cause now you have data response times in microseconds, as you said, and provisioning containers in microseconds is actually really amazing. So, the question is, what is memory-driven computing? What does that mean? And what are the challenges in deep learning today? >> I'll do the machine learning-- >> I will do deep learning. >> You'll do the machine learning. So, when I think of memory-driven computing, it's the realization that we need a new set of technologies, and it's not just one thing. Can't we just do, dot-dot-dot, we would've done that one thing. This is more taking a holistic approach, looking at all the technologies that we need to pull together. Now, memories are fascinating, and our Memristor is one example of a new class of memory. But they also-- >> John: It's doing it differently, too, it's not like-- >> It's changing the physics. You want to change the economics of information technology? You change the physics you're using. So here, we're changing physics. And whether it's our work on the Memristor with Western Digital and the resistive RAM program, whether it's the phase-change memories, whether it's the spin-torque memories, they're all applying new physics. What they all share, though, is the characteristic that they can continue to scale. They can scale in the layers inside of a die. The die is inside of a package. The package is inside of a module, and then when we add photonics, a transformational information communications technology, now we're scaling from the package, to the enclosure, to the rack, cross the aisle, and then across the data center. All that memory accessible as memory. So that's the first piece. Large, persistent memories. The second piece is the fabric, the way we interconnect them so that we can have great computational, great memory, great communication devices available on industry open standards, that's the Gen-Z Consortium. The last piece is software. New software as well as adapting existing productive programming techniques, and enabling people to be very productive immediately. >> Before Natalia gets into her piece, I just want to ask a question, because this is interesting to me because, sorry to get geeky here, but, this is really cool because you're going analog with signaling. So, going back to the old concepts of signaling theory. You mentioned neural networks. It's almost a hand-in-glove situation with neural networks. Here, you have the next question, which is, connect the dots to machine learning and neural networks. This seems to be an interesting technology game-changer. Is that right? I mean, am I getting this right? What's this mean? >> I'll just add one piece, and then hear Natalia, who's the expert on the machine learning. For me, it's bringing that right ensemble of components together. Memory technologies, communication technologies, and, as you say, novel computational technologies. 'Cause transistors are not going to get smaller for very much longer. We have to think of something more clever to do than just stamp out another copy of a standard architecture. >> Yes, you asked about changes of deep learning. We look at the landscape of deep learning today, and the set of tasks which are solved today by those problems. We see that although there is a variety of tasks solved, most of them are from the same area. So we can analyze images very efficiently, we can analyze video, though it's all visual data, we can also do speech processing. There are few examples in other domains, with other data types, but they're much fewer. It's much less knowledge how to, which models to train for those applications. The thing that one of the challenges for deep learning is to expand the variety of applications which it can be used. And it's known that artificial neural networks are very well applicable to the data where there are many hidden patterns underneath. And there are multi-dimensional data, like data from sensors. But we still need to learn what's the right topology of neural networks to do that. What's the right algorithm to train that. So we need to broaden the scope of applications which can take advantage of deep learning. Another aspect is, which I mentioned before, the computational power of today's devices. If you think about the well-known analogy of artificial neural network in our brain, the size of the model which we train today, the artificial neural networks, they are much, much, much smaller than the analogous thing in our brain. Many orders of magnitude. It was shown that if you increase the size of the model, you can get better accuracy for some tasks. You can process a larger variety of data. But in order to train those large models, you need more data and you need more compute power. Today, we don't have enough compute power. Actually did some computation, though in order to train a model which is comparable in size with our human brain, you will need to train it in a reasonable time. You will need a compute device which is capable to perform 10 to the power of 26 floating-point operations per second. We are far, far-- >> John: Can you repeat that again? >> 10 to the power of 26. We are far, far below that point now. >> All right, so here's the question for you guys. There's all this deep learning source code out there. It's open bar for open source right now. All this goodness is pouring in. Google's donating code, you guys are donating code. It used to be like, you had to build your code from scratch. Borrow here and there, and share in open source. Now it's a tsunami of greatness, so I'm just going to build my own deep learning. How do customers do that? It's too hard. >> You are right on the point to the next challenge of deep learning, which I believe is out there. Because we have so many efforts to speed up the infrastructure, we have so many open source libraries. So now the question is, okay, I have my application at hand. What should I choose? What is the right compute node to the deep learning? Everybody use GPUs, but is it true for all models? How many GPUs do I need? What is the optimal number of nodes in the cluster? And we have a research effort towards to answer those questions as well. >> And a breathalyzer for all the drunk coders out there, open bar. I mean, a lot of young kids are coming in. This is a great opportunity for everyone. And in all seriousness, we need algorithms for the algorithms. >> And I think that's where it's so fascinating. We think of some classes of things, like recognizing written handwriting, recognizing voice, but when we want to apply machine learning and algorithms to the volume of sensor data, so that every manufactured item, and not only every item we manufacture, but every factory that can be fully instrumented with machine learning understanding how it can be optimized. And then, what of the business processes that are feeding that factory? And then, what are the overall economic factors that are feeding that business? And instrumenting and having this learning, this unblinking, unbiased eye examining to find those hidden correlations, those hidden connections, that could yield a very much more efficient system at every level of human enterprise. >> And the data's more diverse now than ever. I'm sorry to interrupt, but in Voice you mentioned you saw Siri, you see Alexa, you see Voice as one dataset. Data diversity's massive, so more needles, more types of needles than ever before. >> In that example that you gave, you need a domain expert. And there's plenty of those, but you also need a big brain to build the model, and train the model, and iterate. And there aren't that many of those. Is the state of machine learning and AI going to get to the point where that problem will solve itself, or do we just need to train more big brains? >> Actually, one of the advantages of deep learning that you don't need that much effort from the domain experts anymore, from the step which was called future engineering, like, what do you do with your data before you throw machine learning algorithm into that? So they're, pretty thing, cool thing about deep learning, artificial neural network, that you can throw almost raw data into that. And there are some examples out there, that the people without any knowledge in medicine won the competition of the drug recognition by applying deep neural networks to that, without knowing all the details about their connection between proteins, like that. Not domain experts, but they still were able to win that competition. Just because algorithm that good. >> Kirk, I want to ask you a final question before we break in the segment because, having spent nine years of my career at HP in the '80s and '90s, it's been well-known that there's been great research at HP. The R&D has been spectacular. Not too much R, I mean, too much D, not enough applied, you mention you're bringing that to market faster, so, the question is, what should customers know about Hewlett Packard Labs today? Your mission, obviously the memory-centric is the key thing. You got The Machine, you got the Memristor, you got a novel way of looking at things. What's the story that you'd like to share? Take a minute, close out the segment and share Hewlett Packard Labs' mission, and what expect to see from you guys in terms of your research, your development, your applications. What are you guys bringing out of the kitchen? What's cooking in the oven? >> I think for us, it is, we've been given an opportunity, an opportunity to take all of those ideas that we have been ruminating on for five, 10, maybe even 15 years. All those things that you thought, this is really something. And we've been given the opportunity to build a practical working example. We just turned on the prototype with more memory, more computation addressable simultaneously than anyone's ever assembled before. And so I think that's a real vote of confidence from our leadership team, that they said, "Now, the ideas you guys have, "this is going to change the way that the world works, "and we want to see you given every opportunity "to make that real, and to make it effective." And I think everything that Hewlett Packard Enterprise has done to focus the company on being that fantastic infrastructure, provider and partner is just enabling us to get this innovation, and making it meaningful. I've been designing printed circuit boards for 28 years, now, and I must admit, it's not as, you know, it is intellectually stimulating on one level, but then when you actually meet someone who's changing the face of Alzheimer's research, or changing the way that we produce energy as a society, and has an opportunity to really create a more sustainable world, then you say, "That's really worth it." That's why I get up, come to Labs every day, work with fantastic researchers like Natalia, work with great customers, great partners, and our whole supply chain, the whole team coming together. It's just spectacular. >> Well, congratulations, thanks for sharing the insight on theCUBE. Natalia, thank you very much for coming on. Great stuff going on, looking forward to keeping the progress and checking in with you guys. Always good to see what's going on in the Lab. That's the headroom, that's the future. That's the bridge to the future. Thanks for coming in theCUBE. Of course, more CUBE coverage here at HP Discover, with the keynotes coming up. Meg Whitman on stage with Antonio Neri. Back with more live coverage after this short break. Stay with us. (energetic techno music)