I think we’ll get started now, more people will likely be joining soon, but just kinda have to get started now. So Hello, everyone, and thank you all for coming to today’s webinar. My name is Max, and I’ll be the host for this presentation. Please utilize the q&a function for any questions you may have to the course of the presentation and our guest speaker will do his best to answer them at the end. Mr. Tom funk is a strategic director of architecture at trace three. But he started his career over 25 years ago, becoming the youngest to Bell Labs to re engineer of Unix based voicemail systems is enthusiasm for his field comes from the enablement of new and exciting enhancements of the human experience. And since he started his career, he made an indelible impact on the industry for that very purpose. Today, he’ll be talking about how we can improve the foundations of humanity, like government, academia, business, and more, by utilizing the critically overlooked applications of IoT devices. Please welcome Mr. Tom funk.
Tom Funk [1:13]
Thank you for that intro max. I’m very glad to be a part of this. And again, we’ll do some Q&A. Because what I’ve been seeing in the industry and my experiences are kind of how I’ve drawn this out and of course, differ from other folks. Also, since I’m an engineer, not really officially a salesperson, my PowerPoints are not special or, or, or pleasant to look at or, you know, graphically designed. But the technology is really we’re at a crossroads where the velocity of this new emerging technologies is coming at us in a rate that I have, I have never seen before and to keep up with it is really a full-time job in and of itself, particularly to be solving business outcomes. Using this technology, not putting the cart before the horse and of course getting funding from your various organizations. You know, of course, the classic OT versus IT battles, and OT gets funding it doesn’t but now we have to converge all of this environment into a seamless and functional ecosystem. So, a little bit about me, I’m a first cloud, okay, not really a cloud, but back then they called it back end services. And yeah, we did use mainframes and cloud model. But interesting thing happened at some point backend services, they started calling it the cloud. Maybe it sounded fancier, and the engineers wanted to get paid more, because now they work on Cloud. You’ll note that IoT came from machine to machine from skater from ICS backgrounds, and we start calling it IoT hoping they’re going to pay us more. It hasn’t happened yet, so far for a lot of us. But I think what’s on the horizon, particularly with the enablement factors that I’m going to take you through relative to wireless networking to GPU enablement to training versus edge inference, those sorts of things I hope to touch on, and hope to make it interesting. So, my first application first computer program I ever ran was zork on a Commodore 64 back in 1980, something and of course, having been through AT&T Bell Labs loosened. You know, we touched on a lot of the system fives of Solaris legacy technologies that brought us about to where we are today. So Hyperlite always leads right. And this is just one facet, this is just the 5G aspect of of what it’s going to do to the various verticals, projections, through in this case, arcs. And most of the projections I see are Gartner, but you know, choose your own. And then I broke that down into today’s price of GameStop. Since I thought that was a little bit interesting. So so 20 billion shares 20 billion points, six years’ worth of GameStop is what this industry is worth. That’s GameStop today, not GameStop last week.
Tom Funk [4:09]
I’m talking about the hype, the hype cycle in intercourse, flavor, flavor, noted height man in the industry, that’s actually a screenshot of a cameo that I had him produce to, to insult my friends as a couple of no account onion heads, which is fairly accurate, but amazing, you know, where technology has brought us but looking at the cycle a couple of these I want to you know, we could spend an hour per session on any one of these points, but the very salient ones I think that are applicable to the industry and you know, the industry at large things like obviously security, the ability to deliver broadband to the device level, which you know, previously we had mobile LTE phones, but you know if you’re talking about An ICS, a SCADA system, very constrained devices that typically aggregated a gateway. So, you know, as 5G, and private LTE become more ubiquitous, which is still a few years off. But we’re in a good planning phase right now with plenty of opportunity, particularly in filling in the gaps of 5g and what’s going to amount to neutral host want to talk about that a little bit more. And IoT edge architecture is another very interesting one, to me, for the aforementioned reasons, you have constrained devices, usually a lot of limitations, maybe even security risks, right? Because they’re not built, like a, you know, over built like a mobile phone where you can, you know, check in and in present certificates where you can enforce encryption. So, I think the edge is going to be a lot more, it’s going to require a lot more resources at the edge. And whether that’s by way of GPUs, or FPGAs, or a lot of edge inference can operate just fine on CPUs, whatever that happens to be, I think we’re going to be driving towards this architecture, which allows for intelligent mesh, you know, abstracted hardware systems, and you know, your services layer are going to going to deliver, you know, your AI capabilities. Another thing I want to talk about, which, you know, typically at the data center core, you know, the requirements of model training, as you know, are, you know, very, very high, they require a lot of throughput on your network in order to saturate GPUs when you’re training your models, and that’s going to go wireless. So disruption, you’re going to hear, you know, 5g core. So right now, there’s a lot of 5g, but it’s it’s just overlay, it’s it’s writing on, you know, 4g, Daz and other you know, such systems. So we’re still a few years off, but what it’s going to allow a neutral host and and that’s going to be a monetization potential for people with with private LTE, in their its airspace relative to your elevation. So I’m here in Colorado, and we’re looking at use cases, some are up in the mountains, you might make an educated guess as to who, but they’re trying to get their their gear online, where we’re, while we’re LTE and Wi Fi doesn’t cover, so yeah, there’s gonna end there are white Wi Fi, gateways and backhauls. But a lot of this is going to be new territory. And when a company like, you know, who offers those 5g services, if they’re traversing your airspace, and you’ve kind of got a what they call a squatters rights claim on that cbrs network, then then it’s going to be something that you can build back for. And when I say squatters, rights, private LTE cbrs has a number of different categories. So obviously, you’re incumbent in your, your military, those are always going to be the top priority. And then you’ve got your cable out your MSO operators and companies who have bid on the spectrum through the FCC. And the last category is the rest of us. But the difference is now instead of just a wild west, you know, airspace we are going to be able to
Tom Funk [8:15]
offload, we’re going to be able to seamlessly transition with these networks. Really amazing stuff going on. And what that’s kind of leading is I think, one of the next enterprise gold rushes is going to be these large companies with first of all needs, you know, connectivity needs out in very rural areas, that that’s a great use case for CB Rs, right, you know, Wi Fi six is just going to be internal. So start with a need, because I think in my experience, when you start with the Hey, this is neat edge technology, it doesn’t get really far beyond the the lab and the conceptual in the science project phase, it’s really got to be driven, driven by business nice. But what that allows you to do, when you do get that network out there is, you know, again, you’ve got the claim to that, you know, kind of unofficial claim, you’ve got priority, I should say, at your range. And that’s going to be you know, broadband that just cannot be delivered today without massive cell phone towers. So a lot of a lot of mesh opportunities. Slicing. Very interesting here you’ve probably heard about, it’s the concept of, if you think of a VLAN just in terms of a logical division, you know, divider slicing is going to be not only that logical but also physical because these are actually scheduled transactions. So now everything that’s going on over this wireless airspace is going to be scheduled and has a lot better latency and work together. You know, the crowded other devices is not going to be so much of an issue because of density. And again, it’s going to depend on your your range, your penetration, your indoor versus outdoor, your atmospheric attenuation, and another thing, but we do definitely want to keep a mind to what’s going on in that airspace. So I work with companies like Armis, who is able to tap into, you know, an entire enterprises, OT IoT wireless network, and some of the things that are going on over wireless airspace, hey, maybe I have a, you know, super-secret factory, and we just got a drone Mecca dress show up on our w land controller, let’s treat that, or like probably most hospitals, today, there are a lot of their staff, a lot of the doctors and staff have smart cars, and those updates are going on over the guest network, probably something you know, should want to identify and treat. So networking, you know, a major enabler. And I think the reason why IoT is not as ubiquitous today because we just, you know, we still need wires to some areas and in battery life and so forth. So what you’re looking at here is something very interesting coming out in the Wi Fi six e space, this is an IEEE Working Group on w LAN sensing. And what that means is W LAN sensing eventually in two to three years from now, but the IEEE group is already working on the specs. So it’s, you know, it’s a work in progress. This is going to allow the millimeter waves that come from your WiFi six router to track gestures, objects, people without a video camera. So, you think of the data model modeling involved in a computer vision case today, right? Tons of reference data, do you want format, you know, tracking, OCR, character recognition, what is your use case, but it’s all going in over a flat, you know, video stream. So the W LAN capabilities is something I’m really keeping an eye on, because that’s it’s going to be very transformative for IoT, eventually, you know, IoT, at that rate will not need devices so much. It’ll still need, you know, gateway aggregation inference capabilities, but I’m talking specifically about sensor device, you know, camera type of devices that are trying to make heads or tails out of what’s going on.
Tom Funk [12:07]
Here’s one of the companies it’s their gear is actually Salona. So Salona works with the Roomba and they can already deploy CBR or they’re very close, I think this is this is already offered. But I’m deploying cbrs for again, the private LTE to backhaul other companies to get broadband to where you previously could not. And with that transition, I wanted to get into the applicability, you know, the applications really of AI, if you have not seen AlphaGo, Netflix, YouTube would probably a number of streaming services now, I would highly recommend it because the game of Go is just super complex, exponentially more so than chess, where there’s a, you know, a fight fairly finite amount of move of moves, the AlphaGo possibilities on that board or something to the power of something which I you know, I don’t know, they liken it to stars. So needless to say, they would expect that this game would not be conquered by AI for another, you know, 10 years from the 15 years from when this took place, you know, a decade or so ago, probably less than a decade. But the idea was that a game that requires human skill nuance critical thinking outside the box creativity is not something that you can leave up to neural networks but the long and short of it is they had a Google team that knew really nothing about the game fed at reference data and ended up beating you know the best players in the world. So, I don’t get all wrapped around the axle about AI ubiquity you know becoming all knowing all powerful this that because the use cases are very specific, very targeted require a lot of training require here’s a number of things that I think are keeping us in of course, my twisted sense of humor, I had to put in a Skynet logo, just because I get accused of helping build that. Although I that’s not exactly, you know, the end game here, right. The end game is the experience of business and, you know, interaction and the arts and all that, that important stuff. Um, but, you know, it’s going to be a lot of a lot of efforts to, you know, begin to deliver the sort of function functionality that that it promises, and, you know, data silos sources. standardizing a lot of edge device, a lot of device data that comes from the edge is perfectly useless. So, if you’re looking at maybe 2% of actionable data from a particular sensor, you know, that’s, I would guess that’s probably average. You’re constrained devices. We’ve talked about your tech company data monopolies, and I think there’s been a move towards you know, decentralization, and I’ve got a in my resources page at the very end, Tim Berners Lee’s projects through MIT, it’s called solid, it just describes a markup language whereby you store your own data, and you make it available and accessible as you want. But, you know, data is valuable. So, the idea is that it’s, you know, the first to market with these social media platforms were the exclusive bearers and, and retainers of all of the data being generated. That’s kind of been fading out of popularity, I think, I don’t know where it’s going to land. But it does seem to particularly with your network distribution. Now, it does seem to be leading towards decentralization, your prep work, data brokerages, another concept here that hey, there’s, you know, today very probably industry focused places where you can get some data you need. But I see that becoming more democratized as well, particularly as you think of a microtransactions site like Fiverr. Well, maybe now I want to commission somebody or offer some data that I’ve just been, you know, collecting that could be relevant to whatever model and use case they’re working on. And your AI networks, right now, in the data center core, the concept of cloud led to where you’re pushing that closer to the edge to become accessible more quickly, to your edge devices, is probably something that if I had to guess I would say is coming in the not-too-distant future. So
Tom Funk [16:28]
probabilities, all of AI is driven on probabilities. And if you get a model trained up to 90 plus percent, you’re doing great. Well, think of the fact that we’re letting autonomous cars drive that probably won’t get in an accident. Now there’s other mitigations and things that will come about but model training and and your accuracy percentage is a huge area of importance to data science, and everybody benefiting from from that study. So here, I need AI to tell me I’m ugly, it’s thinks I’m a dog. Whereas it thinks a dog is a teddy bear. So, you know, just a little fun thing I did on my Nvidia Jetson nano, you know, in tinkering, because that’s kind of what makes me interested in doing this for industry is understanding how it works myself. But this is really just to show, hey, an out of the box model, untrained using pytorch, or something is really not better for anything more than a science project. But as you know, you begin to train those models up, that’s when you’re going to see in again, that’s going to be also enabled by technology. GPU is really in hot order right now, for training, that now there’s alternatives. But right now, most of the gear I’m seeing out there, whether it’s by Cisco or whomever that’s using GPU, it’s usually ties back to Nvidia. As you know, GPUs can be very environmentally impactful. They take a lot of power; you’ve probably heard about cities using their entire power budget for mining Bitcoin and things like that. An interesting thing that’s being offered by Microsoft Azure at the edge, their edge box is FPGA, so they can give you different flavors do I want GPUs do I want FPGAs, which traditionally, the problem with FPGA is, is it was the code was tied to hardware, you couldn’t really you easily apply it outside of, you know, whatever you’ve tied together, Microsoft managed to abstract that through containers, and now, FPGA promises to have the same trainability and, and everything that the GPUs can do for your model training. Without the environmental impact, potentially, marketplaces, we’re just looking at, you know, again, a breakdown of, eventually, there’s going to be I think, maybe there’s cities that will be collecting weather data, and some of that will be available. One of a guest client power and gas clan I’m aware of is doing they call it pie for everybody. It’s a statistic that you’re able to get into their collection systems and see now not the proprietary ones, right. But there’s a lot of data that maybe doesn’t build to their proprietary model that they could be offered or otherwise sold. data models, just describing the, you know, relationship, one to one, many to one, as it traverses through its data journey, bringing on additional enrichment points and context, you know, you’re really going to get a purpose built and useful. And really, you know, probably the only way you can collect some of these new metrics, you’re going to get a whole readout on, you know, the journey of data. And there’s a concept of data threading, which describes a whole framework that you’re able to do that with, but just to give you some examples of what you know, pieces of data might be captured and brokered, which all of these will be leading to You know, perhaps somebody’s data model, and otherwise not really readily available. So, if you had such a marketplace, and of course, you know, I just gave a few little categories here in Jason, a data scientists would say, hey, that’s not enough. It’s never enough for data scientists, if you’ve worked with them, it’s never enough reference data, its never enough data points, and so on and so forth. But machinery, audio, right, because if you’re in a factory, and some bearings are, or machines are beginning to go out, maybe it’s not registered on your industrial control system, right? quite yet. But if somebody who was trained, was able to hear that, you know, they say, hey, there’s something wrong with it. So, then we would begin modeling around what are the ranges frequencies and normal operating What are anomalous. It’s just like a mechanic if they hear a good mechanic can hear certain sounds from the engine, and, you know, tell you exactly what’s going on, we’re just training networks to do that. I, I’m not a car guy. So when my wife tells me that her engines making a noise, I just tell her turn up the stereo. But this
Tom Funk [21:08]
transaction would be, you know, just one example of how, you know such a marketplace of this environmental data coming from built smart building cities, municipalities, were these are all coming about another way, Hey, your life doorbell feed, I’ve heard some of those operators. Use your your video feed anyway, even if they’re not supposed to be but think about the idea that maybe the next gen of video doorbell will have an open API. Licensing such that you are able to offer your your video feed to somebody who’s interested, maybe they’re modeling things like Amazon, porch pirates, or you could do facial rec to let yourself in the door. A number of reasons why, you know, this data should and will become more useful. Um, what about a custom data set, so if you want to, for instance, know, train your model to tell you the difference between certain trucks, and maybe there’s no digits on them, you know, for OCR, or they’re not readable, or what have you, we would look at a massing video of of sufficient video, right, because the more video, the better you train going all the way back to statistics, you can train a model, but if it’s only on, you know, a few 100 images, it’s not going to be very accurate, just just like the example I showed previously. But just again, a way to get a new view and a new handle into to data and making it useful. Maybe there’s going to be subscription services where somebody just offers their environmental or whatever statistics might be interesting. I always think of fleets. Because I’ve done some IoT work in fleet with Deborah and a few rolls back. The fleet has these new ways now of getting predictive and prescriptive analytics. And hey, maybe the tire wear is of interest to the tire manufacturer, but otherwise, you know, doesn’t is not entirely important to you know, the bigger picture and they could subscribe, you know, offer that as a subscription. So, Nvidia, again, one of the the supercomputers, I have my rack and stack cert for their 30 gx line. The a 100 is the most recent and I don’t want to get into tech and spec and all this good stuff. But just to show you the, you know, the iteration of where the AI architecture and enablement is going to be arising from and where it very well might be going. So there’s dg x, one dg X to dg x workstation. And these were all purpose built servers kind of for different different reasons, one would probably have T four cards to do inference. The dg x two is going to be the workhorse that trains those models, the dg x two network, the the interconnects that they use through mellanox have been benchmarked at being able to transmit all Netflix HD content in just in minutes. Because they’re so massively over built, that little piece of architecture will be moving over 5g to where we no longer need to train and do most of our AI heavy lifting in the data center. Now cloudlets is one current option that you know I triple E i think began talking about and is getting some, some more traction particularly as data center spend has gone down last year in COVID. I don’t know if that’s universal, it’s just across a certain client base but a lot more edge requirements a lot more cloud autonomous capabilities to the edge is where you know where the transport and the the the functionality is, is driving. Data is very recent, you can see that if you’ve you know, if you don’t archive some of your internet content, and you check again, and it’s gone, you know, maybe it lives on a cache somewhere, but a lot of what’s coming about is going to be, you know, new analysis, new data, points, news, new sources. And, you know, you’re all aware of what, what computer vision can do. And this is obviously ready for primetime This is today, and what things like w land sensing is going to do in two or three years, I think it’s really going to augment a lot of these, a lot of these use cases. And there’s just going to be plenty more that people haven’t thought of, and that, you know, come about by by necessity. So, a couple more resources, solid w LAN sensing dweet io, this was kind of interesting. It’s the social network of devices. So if you’re to go there, you can actually see people who have wired up maybe their wine cellar, or their agricultural project, and it’s streaming out these
Tom Funk [26:08]
metrics that you can, you know, you can see you can look at trending, you can do some very basic modeling, nothing, nothing intense there. But you can also stream out your own lab projects to to, to visualize them. wireless broadband Alliance, obviously, one of the drivers, and that is what I wanted to cover with you today. So, if there are questions, I think we’re going to open it up.
Right, well, yeah, thank you so much, Mr. Front for that in depth write down and thoughtful presentation. And again, yeah, if there any questions at this point that you’d like to be addressed, please don’t hesitate to throw anything in the chat or use the q&a function. If you want, you can maybe be unmuted for your question as well. Let’s wait a couple seconds or seconds. But I actually did have a question myself. With the development of smart cities, what do you personally feel about? Like how cities might increase surveillance? Right? or just general privacy issues? You know, how should they be? How should they be addressed? Cuz you discussed some pros, but
Tom Funk [27:05]
yeah, no, that’s a really good question. And you know, there’s ways to depending on on things like facial rec, if they’re banned in some areas, then I’ve seen vendors use LIDAR along with the cameras, which actually actually obfuscates features. So So LIDAR is is a perfectly acceptable, you know, for a personal identity. Now, as it applies to school, you know, to to public safety. Yeah, there’s a lot of a lot of ethical and and questions to be determined. My daughter’s a school shooting survivor, and I’m convinced that the answer toa that problem is, is going to be solved in technology, because you have to model anomalies with you know, which are on camera, or as we get to better sensing technologies, we’re gonna have the better capabilities. But you know, certainly that’s, that’s going to be it’s not like the Smart Cities are going to be going away from the use of technology to support you know, EMTs, maybe on a journey to the first responder is it. It turns all the lights green before they even get there. You know, today it’s working on just simple sensor and light technology, and eventually it’s going to be AI.
Well, yeah, I guess there’s no questions from the other attendees. But thank you again, so much for presenting this afternoon. And of course, very well done. And thank you as well to everyone else for attending today. Please have a great rest of your day and a great Friday as well. Thanks