>>Hi, Maya. Mr England. And I am an associate professor of electrical engineering and computer science at M I T. It's been fantastic to be part of this team that Professor Yamamoto put together, uh, for the entity Fire program. It's a great pleasure to report to you are update from the first year I will talk to you today about our recent work in photonic accelerators for machine intelligence. You can already get a flavor of the kind of work that I'll be presenting from the photonic integrated circuit that services a platonic matrix processor that we are developing to try toe break some of the bottle next that we encounter in inference, machine learning tasks in particular tasks like vision, games control or language processing. This work is jointly led with Dr Ryan heavily, uh, scientists at NTT Research, and he will have a poster that you should check out. Uh, in this conference should also say that there are postdoc positions available. Um, just take a look at announcements on Q P lab at m i t dot eu. So if you look at these machine learning applications, look under the hood. You see that a common feature is that they used these artificial neural networks or a and ends where you have an input layer of, let's say, and neurons and values that is connected to the first layer of, let's Say, also and neurons and connecting the first to the second layer would, if you represented it biomatrix requiring and biomatrix that has of order and squared free parameters. >>Okay, now, in traditional machine learning inference, you would have to grab these n squared values from memory. And every time you do that, it costs quite a lot of energy. Maybe you can match, but it's still quite costly in energy, and moreover, each of the input values >>has to be multiplied by that matrix. And if you multiply an end by one vector by an end square matrix, you have to do a border and squared multiplication. Okay, now, on a digital computer, you therefore have to do a voter in secret operations and memory access, which could be quite costly. But the proposition is that on a photonic integrated circuits, perhaps we could do that matrix vector multiplication directly on the P. I C itself by encoding optical fields on sending them through a programmed program into parameter and the output them would be a product of the matrix multiplied by the input vector. And that is actually the experiment. We did, uh, demonstrating that That this is, you know, in principle, possible back in 2017 and a collaboration with Professor Marine Soldier Judge. Now, if we look a little bit more closely at the device is shown here, this consists of a silicon layer that is pattern into wave guides. We do this with foundry. This was fabricated with the opposite foundry, and many thanks to our collaborators who helped make that possible. And and this thing guides light, uh, on about of these wave guides to make these two by two transformations Maxine and the kilometers, as they called >>input to input wave guides coming in to input to output wave guides going out. And by having to phase settings here data and five, we can control any arbitrary, uh, s U two rotation. Now, if I wanna have any modes coming in and modes coming out that could be represented by an S u N unitary transformation, and that's what this kind of trip allows you to dio and That's the key ingredient that really launched us in in my group. I should at this point, acknowledge the people who have made this possible and in particular point out Leon Bernstein and Alex lots as well as, uh, Ryan heavily once more. Also, these other collaborators problems important immigrant soldier dish and, of course, to a funding in particular now three entity research funding. So why optics optics has failed many times before in building computers. But why is this different? And I think the difference is that we now you know, we're not trying to build an entirely new computer out of optics were selective in how we apply optics. We should use optics for what it's good at. And that's probably not so much from non linearity, unnecessarily I mean, not memory, um, communication and fan out great in optics. And as we just said, linear algebra, you can do in optics. Fantastic. Okay, so you should make use of these things and then combine it judiciously with electronic processing to see if you can get an advantage in the entire system out of it, okay. And eso before I move on. Actually, based on the 2017 paper, uh, to startups were created, like intelligence and like, matter and the two students from my group, Nick Harris. And they responded, uh, co started this this this jointly founded by matter. And just after, you know, after, like, about two years, they've been able to create their first, uh, device >>the first metrics. Large scale processor. This is this device has called Mars has 64 input mode. 64 Promodes and the full program ability under the hood. Okay. So because they're integrating wave guides directly with Seamus Electron ICS, they were able to get all the wiring complexity, dealt with all the feedback and so forth. And this device is now able to just process a 64 or 64 unitary majors on the sly. Okay, parameters are three wants total power consumption. Um, it has ah, late and see how long it takes for a matrix to be multiplied by a factor of less than a nanosecond. And because this device works well over a pretty large 20 gigahertz, you could put many channels that are individually at one big hurts, so you can have tens of S U two s u 65 or 64 rotations simultaneously that you could do the sort of back in the envelope. Physics gives you that per multiply accumulate. You have just tens of Tempted jewels. Attn. A moment. So that's very, very competitive. That's that's awesome. Okay, so you see, plan and potentially the breakthroughs that are enabled by photonics here And actually, more recently, they actually one thing that made it possible is very cool Eyes thes My face shifters actually have no hold power, whereas our face shifters studios double modulation. These use, uh, nano scale mechanical modulators that have no hold power. So once you program a unitary, you could just hold it there. No energy consumption added over >>time. So photonics really is on the rise in computing on demand. But once again, you have to be. You have to be careful in how you compare against a chance to find where is the game to be had. So what I've talked so far about is wait stationary photonic processing. Okay, up until here. Now what tronics has that also, but it doesn't have the benefits of the coherence of optical fields transitioning through this, uh, to this to this matrix nor the bandwidth. Okay, Eso So that's Ah, that is, I think a really exciting direction. And these companies are off and they're they're building these trips and we'll see the next couple of months how well this works. Uh, on the A different direction is to have an output stationary matrix vector multiplication. And for this I want to point to this paper we wrote with Ryan, Emily and the other team members that projects the activation functions together with the weight terms onto a detector array and by the interference of the activation function and the weight term by Hamad and >>Affection. It's possible if you think about Hamad and affection that it actually automatically produces the multiplication interference turn between two optical fields gives you the multiplication between them. And so that's what that is making use of. I wanna talk a little bit more about that approach. So we actually did a careful analysis in the P R X paper that was cited in the last >>page and that analysis of the energy consumption show that this device and principal, uh, can compute at at an energy poor multiply accumulate that is below what you could theoretically dio at room temperature using irreversible computer like like our digital computers that we use in everyday life. Um, so I want to illustrate that you can see that from this plot here, but this is showing. It's the number of neurons that you have per layer. And on the vertical axis is the energy per multiply accumulate in terms of jewels. And when we make use of the massive fan out together with this photo electric multiplication by career detection, we estimate that >>we're on this curve here. So the more right. So since our energy consumption scales us and whereas for a for a digital computer it skills and squared, we, um we gain mawr as you go to a larger matrices. So for largest matrices like matrices of >>scale 1,005,000, even with present day technology, we estimate that we would hit and energy per multiply accumulate of about a center draw. Okay, But if we look at if we imagine a photonic device that >>uses a photonic system that uses devices that have already been demonstrated individually but not packaged in large system, you know, individually in research papers, we would be on this curve here where you would very quickly dip underneath the lander, a limit which corresponds to the thermodynamic limit for doing as many bit operations that you would have to do to do the same depth of neural network as we do here. And I should say that all of these numbers were computed for this simulated >>optical neural network, um, for having the equivalent, our rate that a fully digital computer that a digital computer would have and eso equivalent in the error rate. So it's limited in the error by the model itself rather than the imperfections of the devices. Okay. And we benchmark that on the amnesty data set. So that was a theoretical work that looked at the scaling limits and show that there's great, great hope to to really gain tremendously in the energy per bit, but also in the overall latency and throughput. But you shouldn't celebrate too early. You have to really do a careful system level study comparing, uh, electronic approaches, which oftentimes happened analogous approach to the optical approaches. And we did that in the first major step in this digital optical neural network. Uh, study here, which was done together with the PNG who is an electron ICS designer who actually works on, uh, tronics based on c'mon specifically made for machine on an acceleration. And Professor Joel, member of M I t. Who is also a fellow at video And what we studied there in particular, is what if we just replaced on Lee the communication part with optics, Okay. And we looked at, you know, getting the same equivalent error rates that you would have with electronic computer. And that showed that that way should have a benefit for large neural networks, because large neural networks will require lots of communication that eventually do not fit on a single Elektronik trip anymore. At that point, you have to go longer distances, and that's where the optical connections start to win out. So for details, I would like to point to that system level study. But we're now applying more sophisticated studies like this, uh, like that simulate full system simulation to our other optical networks to really see where the benefits that we might have, where we can exploit thes now. Lastly, I want to just say What if we had known nominee Garrity's that >>were actually reversible. There were quantum coherent, in fact, and we looked at that. So supposed to have the same architectural layout. But rather than having like a sexual absorption absorption or photo detection and the electronic non linearity, which is what we've done so far, you have all optical non linearity, okay? Based, for example, on a curve medium. So suppose that we had, like, a strong enough current medium so that the output from one of these transformations can pass through it, get an intensity dependent face shift and then passes into the next layer. Okay, What we did in this case is we said okay. Suppose that you have this. You have multiple layers of these, Uh um accent of the parameter measures. Okay. These air, just like the ones that we had before. >>Um, and you want to train this to do something? So suppose that training is, for example, quantum optical state compression. Okay, you have an optical quantum optical state you'd like to see How much can I compress that to have the same quantum information in it? Okay. And we trained that to discover a efficient algorithm for that. We also trained it for reinforcement, learning for black box, quantum simulation and what? You know what is particularly interesting? Perhaps in new term for one way corner repeaters. So we said if we have a communication network that has these quantum optical neural networks stationed some distance away, you come in with an optical encoded pulse that encodes an optical cubit into many individual photons. How do I repair that multi foot on state to send them the corrected optical state out the other side? This is a one way error correcting scheme. We didn't know how to build it, but we put it as a challenge to the neural network. And we trained in, you know, in simulation we trained the neural network. How toe apply the >>weights in the Matrix transformations to perform that Andi answering actually a challenge in the field of optical quantum networks. So that gives us motivation to try to build these kinds of nonlinear narratives. And we've done a fair amount of work. Uh, in this you can see references five through seven. Here I've talked about thes programmable photonics already for the the benchmark analysis and some of the other related work. Please see Ryan's poster we have? Where? As I mentioned we where we have ongoing work in benchmarking >>optical computing assed part of the NTT program with our collaborators. Um And I think that's the main thing that I want to stay here, you know, at the end is that the exciting thing, really is that the physics tells us that there are many orders of magnitude of efficiency gains, uh, that are to be had, Uh, if we you know, if we can develop the technology to realize it. I was being conservative here with three orders of magnitude. This could be six >>orders of magnitude for larger neural networks that we may have to use and that we may want to use in the future. So the physics tells us there are there is, like, a tremendous amount of gap between where we are and where we could be and that, I think, makes this tremendously exciting >>and makes the NTT five projects so very timely. So with that, you know, thank you for your attention and I'll be happy. Thio talk about any of these topics

(upbeat music) >> Hey welcome back everybody, Jeff Frick here with theCUBE. We're in our Palo Alto studio this morning for a CUBE Conversation talking about content marketing, attention economy, a lot of really interesting topics that should be top of mind for marketers, that we're in very interesting times on the B2C side and even more, I think, on the B2B side. So we're excited to have Randy Wootton, he's the CEO of Percolate. Randy, great to see you. >> Thanks very much for having me. A real pleasure to be here. >> Absolutely, so for those who aren't familiar, give us kind of the quick and dirty on Percolate. >> Percolate has been around for about seven years. It started as a social media marketing platform. So helping people, helping brands, build their brands on the social landscape, and integrating campaigns to deploy across the different social channels. Over the last couple of years, it's been moving more into the space called content marketing, which is really an interesting new area that marketers are coming to terms with. How do you put together content and orchestrate it across all the different channels. >> And it's interesting, a lot of vocabulary on the website around experiences and content not a lot about products. So how should marketers think and how does experience and content ultimately map back to the products and services you're trying to sell. >> Well, I do think that's a great point. And the distinction between modern brands, who are trying to create relationships with their consumers, rather than pushing products, especially if you're B2B, or technology pushing speeds and feeds. Instead, you are trying to figure out what is going to enable you to create a brand that consumers pull through versus getting pushed at. And so I think the idea around content marketing is that in some ways advertising isn't working anymore. People aren't paying attention to display ads, they're not clicking, they aren't processing the information. But, they are still buying. So the idea for marketers is, how do you get the appropriate content at the right time, to the right person, in their purchase journey. >> Right, and there's so many different examples of people doing new things. There's more conversations kind of, of the persona of the company, of the purpose, purpose driven things, really trying to appeal to their younger employees as well as a younger customer. You have just crazy off the wall things, which never fail to entertain. Like Geico, who seems to break every rule of advertising by having a different theme every time you see a Geico ad. So people are trying humor, they're trying theater, they're trying a lot of things to get through because the tough thing today is getting people's attention. >> I think so, and they talk a lot about the attention economy. That we live in a world of exponential fragmentation. All the information that we are processing across all these different devices. And a brand trying to break through, there's a couple of challenges, one is you have to create a really authentic voice, one that resonates with who you are and how you show up. And then, I think the second point is you recognize that you are co-building the brand with the consumers. It's no longer you build the Super Bowl ad and transmit it on T.V., and people experience your brand. You have this whole unfolding experience in real time. You've seen some of the airlines, for example, that have struggled with the social media downside of brand building. And so how do control, not control, but engage with consumers in a way that feels very authentic and it continues to build a relationship with your consumers. >> Yeah, it's interesting, a lot of things have changed. The other thing that has changed now is that you can have a direct relationship with that consumer whether you want it or not, via social media touches, maybe you were before, that was hidden through your distribution, or you didn't have that, that direct connect. So, you know, being able to respond to this kind of micro-segmentation, it's one thing to talk about micro-segmentation on the marketing side, it's a whole different thing with that one individual, with the relatively loud voice, is screaming "Hey, I need help." >> That's right, and I think there are a couple of things on that point. One is, I've been in technology for 20 years. I've been at Microsoft, I was at Salesforce, I was at AdReady, Avenue A, and Quantive. And now, Rocket Fuel before I came to Percolate. And I've always been wrestling with two dimensions of the digital marketing challenge. One is around consumer identity, and really understanding who the consumer is, and where they've been and what they've done. The second piece is around the context. That is, where they are in the moment, and which device they're on. And so, those are two dimensions of the triangle. The third is the content, or in advertising it's the creative. And that's always been the constraint. You never have enough creative to be able to really deliver on the promise of personalization, of getting the right message to the right person at the right time. And that now is the blockade. That now is the bottleneck, and that now is what brands are really trying to come to terms with. Is how do we create enough content so that you can create a compelling experience for each person, and then if there's someone who is engaging in a very loud voice, how do you know, and how do you engage to sort of address that, but not loose connections with all your other consumers. >> Right, it's interesting, you bring up something, in some of the research, in micro-moments. And in the old days, I controlled all of the information, you had to come for me for the information, and it was a very different world. And now, as you said, the information is out there, there's too much information. Who's my trusted conduit for the information. So by the time they actually get to me, or I'm going to try to leverage these micro-moments, it's not about, necessarily direct information exchange. What are some of these kind of micro-moments, and how are they game changers? >> Well, I think the fact that we can make decisions in near real time. And when I was at Rocket Fuel, we were making decisions in less than 20 milliseconds, processing something like 200 billion bid transactions a day, and so I just think people are not yet aware of the amount, the volume and the velocity of data that is being processed each and every day. And, to make decisions about specific moments. So the two moments I give as examples are: One, I'm sitting at home watching the Oakland Raiders with my two boys, I'm back on the couch and we're watching the game, and Disney makes an advertisement. I'm probably open to a Disney advertisement with my boys next to me, who are probably getting an advertisement at the same time by Disney. I'm a very different person in that moment, or that micro-moment than when I'm commuting in from Oakland to San Francisco on BART, reading the New York Times. I'm not open to a Disney ad at that moment, because I'm concentrating on work, I'm concentrating on the commute. And I think the thing that brands are coming to terms with is, how much am I willing to pay to engage with me sitting on the couch versus me sitting on BART. And that is where the real value comes from, is understanding which moments are the valuable ones. >> So there's so much we can learn from Ad Tech. And I don't think Ad Tech gets enough kind of credit for operating these really large, really hyper speed, really sophisticated marketplaces that are serving up I don't even know how many billions of transactions per unit time. A lot of activity going on. So, you've been in that world for a while. As you've seen them shift from kind of people driving, and buyers driving to more automation, what are some of the lessons learned, and what should learn more from a B2B side from this automated marketplace. >> Well, a couple of things, one is the machines are not our enemies, they are there to enable or enhance our capabilities. Though I do think it's going to require people to re-think work, specifically at agencies, in terms of, you don't need people to do media mixed modeling on the front end in Excel files, instead, you need capacity on the back end after the data has come out, and to really understand the insights. So there is some re-training or re-skilling that's needed. But, the machines make us smarter. It's not artificial intelligence, it's augmented intelligence. I think for B2B in particular what you're finding is, all the research shows that B2B purchasers spend something like 70 or 80% of their time in making the purchase decision before they even engage with the sales rep. And as a B2B company ourselves, we know how expensive our field reps are. And so to make sure that they are engaging with people at the right time, understanding the information that they would have had, before our sales cycle starts, super important. And I think that goes back to the content orchestration, or content marketing revolution that we are seeing now. And, you know, I that there is, when you think about it, most marketers today, use PowerPoint and Excel to have their marketing calendar and run their campaigns. And we're the only function left where you don't have an automated system, like a sales force for marketers, or a service now for marketers. Where a chief of marketing or a SVP of marketing, has, on their phone the tool of record, they system of record that they want to be able to oversee the campaigns. >> Right, although on the other hand, you're using super sophisticated A/B testing across multiple, multiple data sets, rather than doing that purchase price, right. You can test for colors, and fonts, and locations. And it's so different than trying to figure out the answer, make the investment, blast the answer, than this kind of DevOps way, test, test, test, test, test, adjust, test, test, test, test, adjust. >> You're absolutely right, and that's what, at Rocket Fuel, and any real AI powered system, they're using artificial intelligence as the higher order, underneath that you have different categories, like neural networks, deep learning and machine learning. We were using a logistic regression analysis. And we were running algorithms 27 models a day, every single day, that would test multiple features. So it wasn't just A/B testing, it was multi variant analysis happening in real time. Again, the volume and velocity of data is beyond human comprehension, and you need the machine learning to help you make sense of all that data. Otherwise, you just get overwhelmed, and you drown in the data. >> Right, so I want to talk a little bit about PNG. >> I know they're close and dear to your heart. In the old days, but more recently, I just want to bring up, they obviously wield a ton of power in the advertising spin campaign. And they recently tried to bring a little bit more discipline and said, hey we want tighter controls, tighter reporting, more independent third party reporting. There's this interesting thing going now where before, you know, you went for a big in, 'causethen you timed it by some conversion rating you had customers at the end. But now people it seems like, are so focused on the in kind of forgetting necessarily about the conversion because you can drive promoted campaigns in the social media, that now there's the specter of hmm, are we really getting, where we're getting. So again, the PNG, and the consumer side, are really leading kind of this next revolution of audit control and really closer monitoring to what's happening in these automated ad marketplaces. >> Well, I think what you're finding is, there's digital transformation happening across all functions, all industries. And, I think that in the media space in particular, you're also having an agency business model transformation. And what they used to provide for brands has to change as you move forward. PNG has really been driving that. PNG because of how much money they spend on media, has the biggest stick in the fight, and they've brought a lot of accountability. Mark Pritchard, in particular, has laid down these gauntlets the last couple of years, in terms of saying, I want more accountability, more visibility. Part of the challenge with the digital ecosystem is the propensity for fraud and lack of transparency, 'cause things are moving so quickly. So, the fact, that on one side the machines are working really well for ya, on the other side it's hard to audit it. But PNG is really bringing that level of discipline there. I think the thing that PNG is also doing really well, is they're really starting to re-think about how CPG brands can create relationships with their consumers and customers, much like we were talking about before. Primarily, before, CPG brands would work through distributors and retailers, and not really have a relationship with the end consumer. But now as they've started to build up their first party profiles, through clubs and loyalty programs, they're starting to better understand, the soccer mom. But it's not just the soccer mom, it's the soccer mom in Oakland at 4 o'clock in the afternoon, as she goes to Starbucks, when she's picked up her kids from school. All of those are features that better help PNG understand who that person is, in that context, and what's the appropriate engagement to create a compelling experience. That's really hard to do at the individual level. And when you think about the myriad of brands, that PNG has, they have to coordinate their stories and conversations across all of those brands, to drive market share. >> Yeah, it's a really interesting transformation, as we were talking earlier, I used to joke always, that we should have the underground railroad, from Cincinnati to Silicone Valley to get good product managers, right. 'Cause back in the day you still were doing PRD's and MRD's and those companies have been data driven for a long time and work with massive shares and small shifts in market percentages. But, as you said, they now, they're having to transform still data driven, but it's a completely different set of data, and much more direct set of data from the people that actually consume our products. >> And it's been a long journey, I remember when I was at Microsoft, gosh this would have been back in 2004 or 2005, we were working with PNG and they brought their brands to Microsoft. And we did digital immersions for them, to help them understand how they could engage consumers across the entire Microsoft network, and that would include X-Box, Hotmail at the time, MSN, and the brands were just coming to terms with what their digital strategy was and how they would work with Portal versus how they would work with other digital touchpoints. And I think that has just continued to evolve, with the rise of Facebook, with the rise of Twitter, and how do brands maintain relationships in that context, is something that every brand manager of today is having to do. My father, I think we were chatting a little earlier, started his career in 1968 as a brand manager for PNG. And, I remember him telling the stories about how the disciplined approach to brand building, and the structure and the framework hasn't changed, the execution has, over the last 50 years. >> So, just to bring it full circle before we close out, there's always a segment of marketing that's driven to just get me leads, right, give me leads, I need barcode scans at the conference et cetera. And then there's always been kind of the category of kind of thought leadership. Which isn't necessarily tied directly back to some campaign, but we want to be upfront, and show that we're a leading brand. Content marketing is kind of in-between, so, how much content marketing lead towards kind of thought leadership, how much lead kind of towards, actually lead conversions that I can track, and how much of it is something completely different. >> That's a great question, I think this is where people are trying to come to terms, what is content, long form, short form video. I think of content as being applied across all three dimensions of marketing. One is thought leadership, number two is demand gen, and number three is actualization or enablement in a B2B for your sales folks. And how do you have the right set of content along each of those dimensions. And I don't think they're necessarily, I fundamentally think the marketing funnel is broken. It's not you jump in at the top, and you go all the way to the bottom and you buy. You have this sort of webbed touch of experiences. So the idea is, going back to our earlier conversation, is, who is that consumer, what do you know about him, what is the context, and what's the appropriate form of content for them, where they are in their own buyer journey. So, a UGC video on YouTube may be okay for one consumer in a specific moment, but a short form video may be better for someone else, and a white paper may be better. And I think that people don't necessarily go down the funnel and purchase because they click on a search ad, they instead may be looking at a white paper at the end of the purchase, and so the big challenge, is the attribution of value, and that's one of the things that we're looking at Percolate. Is almost around thinking about it as content insight. Which content is working for whom. 'Cause right now you don't know, and I think the really interesting thing is you have a lot of people producing a lot of content. And, they don't know if it's working. And I think when we talk to marketers, that we hear their teams are producing content, and they want to know, they don't want to create content that doesn't work. They just want a better understanding of what's working, and that's the last challenge in the digital marketing transformation to solve. >> And how do you measure it? >> How do you measure, how do you define it? And categorize it, so that's one of the challenges, we were chatting a little bit before, about what you guys are doing at CUBE, and your clipper technology and how you're able to dis-aggregate videos, to these component pieces, or what in an AI world, you'd call features, that then can be loaded as unstructured data, and you can apply AI against it and really come up with interesting insights. So I think there's, as much as I say, the transformation of the internet has been huge, AI is going to transform our world more than we even can conceive of today. And I think content eventually will be impacted materially by AI. >> I still can't help but think of the original marketing quote, I've wasted half of my marketing budget, I'm just not sure which half. But, really it's not so much the waste as we have to continue to find better ways to measure the impact of all these kind of disparate non-traditional funnel things. >> I think you're right, I think it was Wanamaker who said that. I think your point is spot on, it's something we've always wrestled with, and as you move more into the branding media, they struggle more with the accountability. That's one of the reasons why direct response in the internet has been such a great mechanism, is because it's data based, you can show results. The challenge there is the attribution. But I think as we get into video, and you can get to digital video assets, and you can break it down into its component pieces, and all the different dimensions, all of that's fair game for better understanding what's working. >> Randy, really enjoyed the conversation, and thanks for taking a minute out of your busy day. >> My pleasure, always enjoy it. >> Alright, he's Randy, I'm Jeff, you're watching theCUBE from Palo Alto Studios, thanks for watching. (digital music)