 afternoon. I hope everyone has had a great day. We're getting towards the end of the day so I'm sure we're on information overload which is part of the reason why for this team for this particular session we decided to change it up a bit instead of talking at you, having more of a panel and a discussion so that way it could be a little bit more lively than a single person up here probably be talking. So I've got two of my my partners and colleagues that are going to come up and speak so I'll introduce Lindy first because she's ready. Lindy sprinted a hundred yards. Yes, yeah from the Edge booth all the way here. So Lindy would you like to introduce yourself to the group? Absolutely. Good afternoon everyone I'm Lindy Sabloff on the COO of guys AI. Guys is focused exclusively on the Far Edge and we're gonna get into it a little bit why we feel that that is so critical but we're solving real-world problems for enterprises and we're doing it in a way that is a more efficient system that is more effective and it's so beneficial for organizations. Excellent and then Darren, Darren would you like to introduce yourself to the group? Sure. Hi my name is Darren Blue I work in for the Intel Corporation and our Network and Edge Group specifically in the Federal Industrial Solutions Division. We focus on manufacturing and energy as well as some federal applications so what we'll be talking about today is when we focus on we talk about the Edge we talk about all the way down to the machine like Lindy's talked about here a little bit and where AI plays in the control and the data plane and how all that comes together and how that's actually gonna move towards the future with we've all heard of an industrial 4.0 and the digital transformation that's happening in manufacturing. Awesome so I'll start with you first Lindy with a couple of questions that we have so when you're working with manufacturing customers what do you see some of the biggest challenges that they have? There are a number of challenges with manufacturing and I think it's funny to think of a large manufacturing plant as being a resource constrained environment but it really is and it can be resource constrained in a number of ways one you can have bandwidth issues two you can have network outages or three you have to be much more cost efficient and you can't do that by sending all your data to the cloud the better way to do it is by doing inferencing at the edge where data is actually created. And then Darren I know that you've spent we've talked a lot about this you spent a number of years looking at the challenges that manufacturers have so when you think about kind of how they're using AI into their factories what problems do you see that solving? Yeah so specifically in North America and Europe there's a labor shortage that they have difficulty keeping workers so AI bringing AI in and trying to replicate some of the knowledge where they've had workers that have been in jobs for 30 plus years and they can't find someone to replace that trying to extract some of that knowledge into an AI model is very important trying to not even be able to hire someone to do say an inspection so Lindy has an inspection use case that she's showing over in her booth and some of those kind of things as well as just being able to make things run more efficiently and then with the pandemic that happened they put things together with bailing twine and duct tape and now they want to put some kind of enterprise solution in place because they you know they realize hey this was a better way to do it how now do I take that to the next level and make it more permanent. Yeah I couldn't agree more about you know the issue of attrition in the workforce and while AI isn't you know directly replacing people I think the way we think about it is we're empowering field engineers right and that AI can be the Iron Man exoskeleton that allows them to do five to ten jobs rather than just their one and then have a huge gap. We see a lot of that as well with a lot of the automation that we're starting to put in place and many of the manufacturers so it definitely gets to the skill challenges the talent shortage that they may be having. I'm really curious on how you see and I'll point this at you. Absolutely. So with the AI space I think in the past we've seen a lot of kind of bolt on solutions that are providing kind of data into manufacturing establishments but now we're starting to see where AI may be starting to interact with the control systems or in the actual loop and so I'm curious if you can kind of share what you're seeing from Intel's perspective on what the manufacturers are doing there. Well and let me tie this into the demo a little bit that we have and we're showing with Kelly and her team so the if you look at the control systems that have been in place for and worked very well for long term they weren't really necessarily built with the instantiation of AI in the future they weren't necessarily built as data systems so when you start thinking about data and the IT types of solutions you've got to come up with an infrastructure that can handle and flow that data and at the same time as you start adding these devices in and the factories get bigger there's more and more devices you've got to find a way to integrate those things and manage those things so that virtualization technology that data flow those containers being able to put all those together and then have more remote manageability more remote lifecycle management is what's coming into play as this whole transformation is beginning to happen where it's going to it's going to be a some time till this happens obviously because the manufacturers are not going to replace it till they you know see the benefit and need to do it but that's what's kind of happening that's what we're showcasing in the the demo and that's what's the value prop that's going to happen over time as people start making that transition they start seeing what the art of the possible is with AI and then want to feed that back into their control type solutions I think that's really interesting so I know like Lindy your team is predominantly focused in the computer vision space but I know that from discussions that we've had you're starting to see just like the natural interaction of just AI in connection with even computer vision and the data coming from control systems so how is guys kind of preparing for these changes that are coming to manufacturing that's a great question I think we need to take a little bit of a step back I think they're still unfortunately is this persistent notion that AI models are large that they're cumbersome and that they're expensive and I think really when we need to start talking about the edge there are a couple things we need to get across one you can have these incredibly powerful but lightweight algorithms right I like to joke around it's it's Robin Williams genie in the 90s version of Aladdin where he pops out of the lamp and he goes phenomenal cosmic power living space and that's really how we have to start thinking about you know edge enabled AI and that these models are available and they are pardon the overuse of the term transformational but they really are transformational and I think from an adoption standpoint it's going to help really start accelerating adoption to be able to have you know these different algorithms for specific use cases that can run on you know very lightweight but you know small compute it is absolutely critical and especially you know I warn the reasons we've partnered with Intel you know is the ability to optimize on open Vino right and that you know when you're doing quantization you can drastically increase the FPS and then you're lowering you know the memory consumption and so it makes it so these models can run can run on small compute and can run very efficiently but the other thing I think to think about an edge is not just how AI has shifted to fit manufacturing to you know to fit energy and to fit you know other industries it's also to say that you know it's about scalability right and by having lightweight models it is more scalable but lightweight models that are agnostic and can run on different devices and also take input from all you know the proliferation of IOT sensors that are out there I'll stop it yeah actually I would love Darren for you to kind of chime in on this because I know we've spent a lot of time talking about embedded system developers and how different they are there's actually some of them in the audience because I know some of them the way that they operate is a little bit different in the way they think about how to build the applications or even the AI models will be different so I'm really curious for you to kind of share what Intel is trying to do in this space with especially with the use of open Vino to help with some of that optimization yeah I mean I think that this this transformation this whole thing we've been talking about actually opens up space for the developers and one of the things I actually hear from the manufacturers is the systems of the past because they were pretty vertically integrated because that was necessary when you start opening it up to more of an IT type of environment you can get developers and so the complaint from the manufacturers I get innovation only from one company when I can open it up and create a more open system I get innovation from all the developer community I get I have my own people that can write I can buy stuff from the developer community I can hire contractors I can even sell my own back out to the developer community and so we as Intel try to encourage that as a solution space so that you know it advances everything much faster you get more eyes on the whole thing you get more innovation happening and then that drives the whole transformation forward and do you have I know that we may not be able to talk about specific logos like specific customers but do you have an example of how this has worked with in the past with you and the work that you've done with some of the manufacturers with the manufacturers well it it's still very early in this transformation so the developer community is still kind of spinning up but it you know like she mentioned open Vino has had a lot of success and so there are a lot of companies out there using like you know you've got I've got one sitting here like we would consider her as a developer right and so she's taking that open Vino space and moving it forward we've got something that expands upon that with a data bus called Edge Insights for Industrial that is also being used by some system integrators to deliver some video applications as well so those are actually continuing to move forward and and be evaluated and taking more so right now it is more in the video inspection space but I think you're going to see that expand into more and more use cases as it goes forward okay that's great and so Lindy on your side with your team as you're looking at some of these new use cases and the interaction with customers can you describe for the team like what that engagement model really starts to look like because I know that you have a starting point of a model and then that's where your team really kind of engages to the customer environment yeah it well there is some you know one size fits all in in terms of AI models you're always going to have you know a certain bit of specification for the customer right and especially to maintain accuracy you know AI is about iteration and so as you go in you know and you're working with the customers you know to understand you know their needs from a hardware perspective power consumption you know what is the bandwidth available and so you know going in and really being able to very rapidly it's not a completely bespoke solution but at least customize so that you can achieve high levels of accuracy depending on the customer okay so now I'm going to ask you the thing that because I work for Red Hat I should ask you this so let's talk about how we've been working together and so in particular let's talk about the work with Armada and the system that has been built and the role that you see some of the open source technologies that Red Hat provides brings to the AI space oh I am an unabashed fan of Red Hat you know I think you know prior to working with Red Hat we looked at some solutions that you know gum and shoelace right that we made it work but it wasn't it wasn't efficient and it certainly wasn't scalable right and at the end of the day the customer needs something that is as close to turnkey as possible they don't they need something that they don't have to add IT personnel and that doesn't take you know in tens of hours of training they need something that they can install turn on and use that is going to make a difference in what they're doing you know be it manufacturing be it oil and gas and so where Red Hat has been critical to us you know especially Red Hat device edge I think you know we had looked at you know different Kubernetes solutions but when you're talking about the far edge and you're talking about some very small compute Kubernetes is just a little too heavy and so we were introduced to Ansible a little over a year ago and it's made a significant difference and I think that opened our eyes to say we can actually create an entire edge ops platform around it so we've actually integrated Ansible inside you know doing the callbacks and orchestration inside our edge ops platform so that's one two you know especially rel 9 has been absolutely marvelous so using rel 9 and pod man and then you know to replace Docker and then also using quay as our image registry so we consume a incredible of Red Hat products but it really has made a significant difference and I think you know Kelly you and I've had a number of discussions you know as we go out you know getting back to that idea that you have to provide a turnkey solution to a customer you really need a best of breed partner ecosystem because I think as we all go out if we don't have unified messaging and we can we can't come in and say here is your solution and a solution isn't just the software the solution you know you know is the hardware it's the processors you know it is you know the ability to to optimize with with open vino to come in and say here you know xyz corp this can you know can go right into your manufacturing plant this is this is going to be your base for how we begin automation and it is critical to get those components right but to have you know a a glove in hand kind of go to market with all the partners so there isn't so much noise when you go in and so red hat not just from a product standpoint is has been fantastic and and Intel as well you know working with a number of different departments within the company but everybody goes in and says how are we gonna get this done how are we gonna create that turnkey solution and it is such a difference when we sit in front of the customers and we all say here's here's your sheet this is how we're gonna do it here's your solution you know let's begin and it's it's been incredible yeah well I think I love the power of the the user interface that you guys have built that allows for the seamless kind of not only cataloging of the devices and the locations that you're pointing specific models to in the cameras but how seamless it is for the deployment so I'll shamelessly plug red hat with that because the deployment of all the models within their platform is they use Ansible for that and then it obviously runs on on the podman container system with rel now with that because we've been doing a lot of this work what's great is that when I look at the work we've been doing with Intel and we look at the control systems which becomes like the happy marriage of bringing the control along with the data and analytics together we start to see a lot of commonalities in the architecture and so if we look at as an example Darren the the infamous water tank demo that is with Schneider electric in our edge booth I know that we see a lot of similarities in some of the technology that's used but do you want to share with share for the group just in case they haven't been to our booth how that's all working with the water tank demo absolutely so this is a journey we've been on for a little bit we piece some Frankenstein some stuff together in the lab initially and then over the last year or so we've worked with red hat and Schneider electric to put together commercial software actually to bring in more to reality so we have a simulating a process control demo on the edge area on the floor using general purpose compute so some edge devices they use to core l car lake and tied to a server that uses red hat edge linux along with ansible and a runtime from Schneider electric to control the level of two tanks with one of them draining into the other constantly and it controls a variable speed pump to keep a constant level on the top tank so the and it can do bumpless failover which is very important when you're a control environment so that you know nothing ever lose and then ansible detects that something failed spins a new one up and then you you get the redundancy back into place and so if you haven't seen it please go over and see my friends Rob and Jeremy they'll be happy to take you through all that but the other plus to that then is you've got general purpose compute right at the machine level right out on the floor that allows you to do other things that you might want to try so you can deploy a vision case or you can deploy a machine learning that takes time series data and does something you know if you're if you're pulling information off of a welder that's just with arc and electricity etc to do different things so that allows you to do a lot of different things you can and so you know that general purpose compute on the floor allows you remote manageability allows you that to do those other things as well right there on the floor and then you know the things that we still have to to do is you know make sure that all of that creates the determinism that creates the reliability as we move this forward and there are some deployments being tested actually in real factories in real areas that we'll get some feedback on for that so it whenever that whenever that moves forward and it gives then that gives the sort of the the sandbox for people to explore and what they can really do when you get that kind of opportunity right at the machine I think that's a great point so what we're starting to see in the factories is people putting more general compute in their factories some would call it like a mini data center but the intent is then for it to be able to take a lot of the data that may be coming off of the control systems place it within their historian which is basically a time series database and be able to do advanced analytics on it but in addition that because they have the general compute there you can run additional inferencing applications if you need something that's a larger compute or potentially you can get away with not having it directly connected to the camera on that exact same box so we start to get into more general purpose where in manufacturing today I think many of them are are much more accustomed to everything has one specific purpose right lots of small computers and so that you've got a ton of compute that requires a ton of hardware refresh a ton of software updates when they can get to more of this general compute and start some of the I don't really like the word but it's used all the time the workload consolidation they're not really workloads to me if they're running your business because they are literally deciding whether you make money but I know that workload is a pretty common key term and it you know obviously that's very important for us at Intel and to move towards that general purpose compute but I'll steal a phrase from I work with Dr. Henning loser from Audi and been working for a while he was on stage at an ARC conference and his point was today I do have to buy a dedicated box in the future I just want to buy an application I can load on then I can even try it without having to you know go buy a whole box install all this stuff and if it doesn't work I just you know take the application back out and I can try something else and I can try three from different places and decide which one I like the best so not only does it give him opportunity to try to manage his whole factory a lot better but it gives him the opportunity to try different options from different places as simple as maybe not quite as simple because it is a manufacturer environment is downloading an application on your phone that is very interesting we've had a lot of discussion actually in the edge booth around can we just put QR codes on everything and then people can easily scan the the hardware as it comes in with all the software they need and just plug it in so there is definitely the concept of making things simple especially manufacturing which I know that you brought up as well Lindy and so when we think about how we may build AI systems for manufacturing in the future like if you were going to give the audience a little bit of advice what would be kind of the three things that you would tell them to think about that they have to keep in mind when they're going to do AI in the factory it's a great question I think there are a couple a couple things there you know one you know how are you going to scale right you know especially when you're thinking about a factory or not just one factory but multiple factories across multiple geographies so how are you going to scale and you touched on that earlier it's kind of why we built our Armada edge ops platform is to make you know scaling simple you know across so that you can either run you know in the cloud you know or actually on the edge server itself you know I think the other thing you know there as well is how are you going to keep your data secure and so we've added an encryption layer as well that you need to encrypt at the device level and you need to make sure that those divide you know edge is much much more difficult than cloud when it comes to security so you have to make sure that you have the right system in place so that somebody can't just come up and walk away with that intel nut right that if that happens then you don't have any compromises and you know you need to think about you know what are the parameters in the manufacturing plant itself you know is energy consumption a critical factor for you is cost you know a critical factor for you I think the power of the edge is that you can have hundreds if not thousands of heterogeneous devices but you know I think this is why we've partnered with Intel is that Intel really gets that in a way that other companies that we work with have not and that you know you really can run you know on small compute and the edge is a volume game and you need the right partners who understand that who can build you know the right hardware and the right infrastructure along with you as you go out and as you start building you know the different ai use cases Darren same question for you well I'm going to just build on what she said and expand on it so absolutely the security the intrinsic security I think is comes across is very important the other thing I'm going to say about the scalability is what you see is people will put in they'll test something over here and they'll test something over there and they'll test a third thing and only start putting a fourth thing and then the plant manager the owner or the engineering manager come in and go like well we need a data platform and none of these things connect to it now I've got all these disparate systems that I want to connect I've got to start all over and so they get upset so you have to think about as you start growing you need to think about your data platform and how that goes and then the other piece of when you start installing ai is are you going to close the loop so what I find is okay well I know what's wrong now but I how might what am I gonna do about it and so because in some cases we've we've done some you know POCs and stuff with vision and if and we can tell you whether something's going good or bad in milliseconds but all we can do right now because the systems aren't quite there yet is stop the machine and have somebody come over and go oh okay let's do what we can't so the next question is well how do we connect all these things together and close that loop how do I make it so that it's becomes more and more autonomous which we see as moving forward I'm smiling because I feel like you set that up for me perfectly with our edge ops platform or Armada does you know especially with with IoT devices and cameras you know there's a whole layer in there that actually allows you to start stop processing you know look at camera angle you know set your ROI so you know if you're doing optical character recognition you can say you know focus in on on this component of the so you can there's some amount of self-correction built into it again making it more simple for the customer to use excellent so I want to thank both of you for coming and talking about the work that you're doing it for AI and manufacturing I think that we've got a little bit of time so we'll definitely like let's stick around see what questions that the group may have because I have a feeling there may be some like this gentleman right here seems to be a very interesting question yes it's much more about physical security right because if you're moving data to the cloud then you only have one issue when you're talking about the edge you're talking about tens if not thousands of devices that somebody and these are small devices that somebody could literally come unplug put in their pocket and walk away with and if you're talking about computer vision depending on the use case and depending on the privacy level if you're oil and gas you have some very proprietary information on that device if you haven't encrypted at the device level and then attached that you know kind of glove in hand to the network properly you big problem and so it's not like somebody's coming in and you know lugging you know some big server out of there that somebody's going to notice when you're talking when you're talking about a you know an Intel knock or an SDM or Raspberry Pi right these are things that are very easy to tuck you know tuck into a pocket and walk out with but I think there's also a piece in the the cyber piece like if you're talking about the control system that is very true um when I talk to manufacturers the intrinsic security I'm mentioning is their security today is an air gap like if you would if you wanted to mess with a refinery you'd have to run through the door and turn a handle on a valve right if it now becomes connected everyone outside the cement box that the control system resides in today everyone kind of has the somebody hacked into target through the air conditioner or something if if somebody does that to our refinery it's a bad day for the neighborhood right I mean in something in that case or and so they're obviously very sensitive to making sure that that something like that cannot happen right right those are very important things and they want to make sure that as the system is built security is thought and built in as it's created so I think you have to think about what maybe is taking place within a factory so if you think about whether it's a refinery or it's an actual you know physical factory like making automobiles there is human life that is at risk most of the time when you're thinking about what happens in these facilities whereas when you're thinking about kind of endpoints off of the the ingress or egress from the cloud that's just application and data while privacy and data security are still really important there's a slightly different dynamic when you're talking about the risk of human life and it's not just the the risk of human life of the people that are working in the factory but as I kind of joked with some people when they are looking at the demo around the water tank what happens if this you know the water tank overflows well in real life things could go boom and that also means that you know neighborhoods and you know people outside of the working condition could be at risk so many of these systems have to be designed very differently from a security standpoint because while they may not be a government entity I think many may consider them to be like national infrastructure on thinking about how these particular industries may run right so in it's more than securing the actual hardware yes yeah yes Hello Steve it's nice to see you again industrial-centric context or you know they have all lived very very uh temperature environments humidity environments, power constrained environments and so like in this the whole thing about the way we are but if you have a question I can forward a lot of the answers I wanted to give from the table or give the your context to this so broad once you put a device in manufacturing compliance or in a bus stop that device will be missing from what walks out of the plant over there how critical is the data? GDPR compliance yep yeah the GDPR compliance is is critical if the second you step foot in Europe but I want to piggyback off something you said as well that I think what's critical to think about in terms of the edge is what you don't want to do when when you're talking about security is create you know a very tedious environment as well so you need kind of a good soft handshake between the edge device and and the edge server that you're running so that you can make it so that these things you know are are are automated they are repeatable you don't want something having to go and you know you don't want to unlock the door every single time you have to run an application if you take the device and leave I can decrypt everything and you add on the thing I can try to pull your language model links off of the device and have it reference something in the server for security department and at that point this is securing the zero of the edge but then I think of the edge you can do it right when the zero is just architecture and you segment your network correctly and you have the correct thing to place that the edge would probably be the place you could start carving the most that's one aspect of it you know but you know let's take what they're doing with with Schneider Electric right in terms of that demo like when you start automating you know those systems you have to make sure that somebody can't take them over externally as well and I think it's there are a lot of different things you know going on with that you know especially also not from the security aspect you know but the other thing you know and I'll get back to an earlier point I made AI is about iteration right you can't just have a set and forget for a model otherwise your accuracy is going to drop over time and so you do have to have that you know a good easy to use management layer so you can push updates so you can manage you know one of the things we've added to it recently and I'm excited to demo this for you soon is we've actually added a data layer to it so we've created a data store so now you can pull you know either the inferencing the JSON data or the image data pull it back to the edge server and so you can actually do retraining at the edge so you've made the system actually a little bit more secure because you don't have to bring any data off of the edge in order to do retraining so slightly different aspect of security but critical nonetheless any other questions I'm we're what's stopping people from being able to have a drink so yeah yeah I fully appreciate that so thank you all very much for coming really appreciate your time and if you have any other questions for us you will find us in in the edge booth in the the exhibit hall if you haven't seen it we have lots of really cool stuff that you actually get to physically see it's not just a display screen so that's booth yeah so come on down thank you thank you everyone thanks