Conversations on Applied AI - Stories from Experts in Artificial Intelligence

Erik Zwiefel - Asking the Right Questions of Artificial Intelligence

December 06, 2021 Justin Grammens Season 1 Episode 26
Conversations on Applied AI - Stories from Experts in Artificial Intelligence
Erik Zwiefel - Asking the Right Questions of Artificial Intelligence
Show Notes Transcript

The conversation this week is with Erik Zwiefel. Erik is is a Director and Technical Specialist in the Artificial Intelligence and Machine Learning group at Microsoft. He helps customers build out Advanced Analytics and Artificial Intelligence solutions leveraging Microsoft Azure and open source data science packages. Prior to that, he was a Senior Lead Data Scientist at Target as an operational Lead for the Enterprise Data, Analytics, and Business Insights Testing team.

He holds a master's degree in business intelligence and has taught graduate-level courses in the Master's of Science Business Intelligence and Analytics program at Saint Josef’s University.

If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events! 

Resources and Topics Mentioned in this Episode

Enjoy!
Your host,
Justin Grammens

Erik Zwiefel  0:00  
There was an article on popular science and how it can solve these interesting problems with evolutionary algorithms. And I remember the writer then said in the future computers will be able to answer any question we have, then it will be up to us that the real challenge will be asking the right questions. And so it's kind of as we continue to move forward, that seems to becoming more and more true.

AI Announcer  0:22  
Welcome to the conversations on applied AI podcast where Justin Gammons and the team at emerging technologies North talk with experts in the fields of artificial intelligence and deep learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real world problems today. We hope that you find this episode educational and applicable to your industry and connect with us to learn more about our organization at applied ai.mn. Enjoy.

Justin Grammens  0:53  
Welcome everyone to the conversations on applied AI Podcast. Today we have Eric weevil. Eric is a director and technical specialist in the artificial intelligence and machine learning group at Microsoft. He helps customers build out advanced analytics and artificial intelligence solutions, leveraging Microsoft Azure and open source data science packages. Prior to that he was a senior lead data scientist at target as an operational lead for the enterprise data analytics and business insights testing team. He holds a master's degree in Business Intelligence and has taught graduate level courses in the masters of science, business intelligence and analytics program at St. Joseph's University. Thank you, Eric, so much for being with us here today.

Erik Zwiefel  1:29  
Yeah, thanks for having me.

Justin Grammens  1:30  
Awesome. Well, I gave a quick sort of like rundown. I know, that's what you've been doing in your career. I guess recently. Maybe you want to give a little bit of a background? You know, maybe start from the beginning, I guess, how would you get to where you are today?

Erik Zwiefel  1:43  
Yeah, I've had a interesting path to get here. I, I have a bachelor's in biochemistry. And I started right out of school at Medtronic and clinical research, worked a lot with clinical sites, kind of running the tests, and so on. And then Medtronic got a site license for Spotfire. And it was all downhill from there started playing with Spotfire, and doing reporting and everything. And that just really got into, I've always been a computer nerd. But this really ignited the passion for data. Awesome.

Justin Grammens  2:13  
Yeah. And this was well before the term data scientist, I think really meant anything to anybody. I think you graduated from college, undergrad, I guess, around 2000 or so. Right? Yeah. Oh, has some of the field I guess changed? And how would you maybe define what Data Science and Artificial Intelligence is today? Yeah, so

Erik Zwiefel  2:31  
I've seen a big shift from just in my time, from more closed source to more open source. So the big players, before SAS, MATLAB and so on, seem to have been usurped in a lot of ways by Python. And R, although our seems to be a little bit of fallen by the wayside, and kind of practical day to day work that I've seen. I know, I'm gonna get a lot of hate thoughts and comments directed about about that. We've had a good audience. Okay, excellent. Python has become kind of the de facto, it seems and leveraging open source kind of everywhere, even to the point that at Microsoft, we try to be very non opinionated on even the languages that you're using for data science, and embrace the open source packages and so on.

Justin Grammens  3:13  
Seems like in general, Microsoft has moved into that direction more and more recently, with that probably true statement.

Erik Zwiefel  3:20  
Yeah, absolutely. I think you have seen a big shift to Microsoft, not only embracing open source, but contributing heavily. Even in the machine learning space, we have packages for interpretability, that we've open sourced for model fairness that we have open source for differential privacy, and so on. So really trying to contribute back to that open source community.

Justin Grammens  3:39  
Awesome. Well, one of the things I'd like to ask people is there on the show is how would you define AI? If you have a short elevator pitch, or people ask you, what do you do during your workday?

Erik Zwiefel  3:51  
Yeah, absolutely. So I try to take a very broad thought process when it comes to AI. AI, for me is any way that we have computers try to mimic human intelligence. So the most ubiquitous AI to me is like a password reset bot, where there is no real intelligence here, but it's trying to mimic human intelligence. And then it will obviously go very deep into neural networks and can grow from there. But just at the core, it's a computer trying to mimic human intelligence.

Justin Grammens  4:22  
Awesome. Do you believe that humans will be able to then get into the emotional side, emotional intelligence is kind of out there. But it's another sort of like tangential question.

Erik Zwiefel  4:33  
Yeah, that's a really, really great question. I think it will be, you know, artificial emotions, just like computers. I don't think we'll ever really think in the sense that we do, they'll never truly feel in the sense that we do Never Say Never obviously, but I think that you will start seeing fake emotional algorithms masking as emotions. And I think just as we're seeing advances in like the language processing space with GPT, three and so on, Like, it's it's some amazing stuff that it's able to do already. I can only imagine, as you know, well, but the pace of things like a year from now, we're gonna be saying I can't believe it could do that.

Justin Grammens  5:11  
Yeah, it has been insane, that it gets the ray Kurtzweil talks about, you know, just that, that sort of, essentially things happening faster and faster, right, this exponential growth. Yeah. And that, you know, it took 1000s of years to invent the wheel and hundreds of years to invent the sword and, you know, 50 years to invent the gun, and all of a sudden, you know, 10 years and we have a cell phone and everything right. So, you know, everything's been going more and more faster. What what, I guess, you know, the other more on like, the personal side? How would you describe yourself? What are a few words? Maybe that, that describe you or some of your strengths and weaknesses? Yeah. So

Erik Zwiefel  5:43  
I think the biggest description for me is father, five kids live outside of the Minneapolis area, kind of in the country. So when I'm not doing artificial intelligence work, either playing with the kids or have a woodshop like to go out there and use a lathe, turn pens, that type of thing. Oh, nice. You build any furniture at all or anything? When my wife tells me that I need to build something for her then yeah, that we build it.

Justin Grammens  6:10  
That's good. That's, that's good. Yeah. My, my, my stepfather built a bed. Okay. Once and it was, you know, of course, he didn't have all the equipment, but he had a friend who had all the all the equipment to be able to make it and I was pretty impressed at the end of the day, what what you can do and you know, somebody who knows what they're doing? Yeah, absolutely. Not. For me, I deal in zeros and ones, I tell people that I have a lot of respect for people that can build stuff like that.

Erik Zwiefel  6:34  
No, I totally understand that. I think that's why I get drawn to, you know, working with my hands like that is at the end of the day, the zeros and ones sometimes it doesn't feel like you've actually done anything or accomplished anything. You've done a lot of StackOverflow searches, but it feels like you're not moving the ball forward, at least a bit physical, I could see what I've done.

Justin Grammens  6:52  
Okay, well, I have some progress for sure. Well, what what is the day in the life of a person in your role at Microsoft? Yes, so

Erik Zwiefel  6:58  
I'm part of a team called the Global Black Belt team. And I get embarrassed every time I have to say that. But it's a fancy Microsoft way of saying I don't get assigned to a specific customer group of customers. Instead, I work with all customers in the US, Canada, Latin America, but I narrowly focus on machine learning on Azure. And so I'm helping those customers on lock that machine learning on Azure, and it's very flexible day to day, one day, I may be talking to a company that's working on autonomous vehicles, the next I may be talking to a manufacturer who wants to do some quality assurance with some vision scenarios, or potentially even collecting IoT data, building models around that. Or I may be talking to someone who's just kind of traditional using tabular data, and they want to start looking at customer churn and so on.

Justin Grammens  7:50  
That's awesome. Yeah, well, when you talk about IoT data, that's, that's my bread and butter. That's something I've been working in. And we do it at lab 651 fairly often, and actually uses your a fair amount and a lot of stuff in Power BI for the display analysis side of it. But then, yeah, once you're talking about large, large datasets, that's where a lot of the power of the AI can come in. Are you solution architecting things where then the internal team within those organizations are actually going to be the ones to implement it.

Erik Zwiefel  8:16  
Yeah, oftentimes, that's what we're doing with them kind of solution architecting. Or I might be working with customers on a follow up capacity to say, Okay, we are implementing now we're running into this challenge. And so I can help troubleshoot that can even come in at the earlier stages before solution architecting and saying, Here's what, for instance, the buzzword right now seems to be ml ops, like how can we take software development practices and bring them to machine learning, and make sure that machine learning is repeatable and so on? And so we may be in the early discussions with senior leaders to say, here's why you need to go down this path. So yeah, that's kind of where we're at play in those spaces.

Justin Grammens  8:54  
Cool. I had you done much on Microsoft platform before you joined? Were you guys doing a lot of it in print prior places? Or was or is it because there's so many different platforms? So many different technologies out there? I'm just more or less curious to know if they kind of recruited you away from your current employer, as you knew it, or you kind of like picked it up when to giant?

Erik Zwiefel  9:14  
Yeah, absolutely. So for me, it was I hadn't done much on Azure at all. I'd done a few little side projects, building websites for for someone, most of the work I did beforehand was Python on my laptop, you know, and then trying to find a random Linux server somewhere, I could run a cron job on to production analyze it. There's probably still a few sitting out there that I wrote, almost six years later, got interested in Microsoft, because my brother works here and convinced me like, Hey, you should come work here. And so I had a chance to join him and we've been able to do some work together.

Justin Grammens  9:49  
That's awesome. That's cool. Well, I mean, yeah, part part of the discussion that I typically have with people that are on the show, and we usually talk about it more at the end, but we can talk about it now is is that how do people break into this? Are there certain tips or tricks? Or I guess we talked a little bit about your path. But yeah, if someone is brand new coming out of school, and they're, you know, they're graduated maybe, you know, computer science, you know, they, they like math, you know, they're interested in more, you know, analytics and machine learning and stuff like that. Where do you suggest they go? What are they start looking at?

Erik Zwiefel  10:18  
Yeah, great question. Right now, I think a path that if I were to start over again, that I wouldn't be investing a lot of time is platforms like Kaggle, to say, let me go and solve real problems, I'll get that kind of experience under my belt, and then be able to create a GitHub examples that I can show to potential employers. And then, you know, networking whenever you can, understanding who had you know, a particular vendor, if it's a vendor, you want to work for a particular company, networking with them through, you know, monthly meetups, through conferences, and so on, and just getting to know them, each organization is going to be different on how you want to get into that organization. So I think it's important to tailor your approach specific to that. But I think for just data science in general, yeah, just get a lot of projects under your belt, just work through them. And sometimes what I like to do is create artificial constraints for yourself. So I was a more of a SAS Type user before, or Excel even to do some of the statistical analysis. But I had a particular project that target that I said, You know what, I'm going to use Python for this. I know it's going to take me twice as much time, but I'm going to use Python for it. And I'm very happy I did. And I had a look back. But yeah, that project took a little bit longer to complete. But that artificial constraint helped me grow into, you know, developing with Python.

Justin Grammens  11:46  
Yeah, that's awesome. That's awesome. Yeah, it's been hard with COVID, I think with some of these, getting a chance to have conferences in person and meeting people. But on the flip side, I guess I would say just sort of thinking out loud here is, you know, now everything is online. And you can attend things that happen all over the United States, or even maybe even across the world.

Erik Zwiefel  12:03  
Yeah, absolutely. And I also had a lot of people either reach out to me on LinkedIn, or I've reached out to them to just say, could we connect, I've never met you before. I see you are maybe solving a problem that I'm interested in solving. And just Can I take 30 minutes of your time to ask about your career. I've only had I think one person not respond to a query like that, that I've sent. Most everyone else would love to talk and help other people along the journey.

Justin Grammens  12:29  
Yes, awesome. Well, at the end, and in the liner notes of the podcast episode, we'll be sure and put your your LinkedIn information or whatever other sites you want to share, so we can can reach out to you. But yeah, I'll probably have you give it here at the end as well. What are some interesting applications? You know, the the podcast is really conversations on applied AI and AI, there's, there's probably some things related to your business line of work that you can't talk about. But I am just kind of curious if you're sort of reading the news, or you know, other things. It can either be personal, it can be professional, it can be just, you know, sort of like, stuff that you've seen over time. But yeah, what are some interesting ways you've maybe been seeing AI being applied?

Erik Zwiefel  13:07  
Yes. So the the one that recently has been just every time that I work with it, it's very interesting to me is GitHub co pilot, my God in the beta for that. And so as you're developing, if you're not familiar with this, it is a natural language model trained on programming code. And as you're working in Visual Studio code, and kind of developing out, I'll start by declaring a function. And as I give the function a name, the model will start suggesting and completing the function for me based on the name that I've given it. And it also appears to look at how I've written my code before and the variables that are available to me in my code, and we'll plug in those variables. And those, as I'm developing it, it still gives the user complete control over the eventual code, it kind of just suggests is a type ahead and you hit Tab to complete it, or you can cycle through a bunch of different suggestions. But it has been incredibly accurate as I've built it out, where I'm like, wow, I had once where it defined one function it was doing like, create read update type things. And it suggested about eight other functions all at once. And I just hit Tab, and boom, they were all completed, and I had just go back and tweak them a little bit. So that's been a really, really interesting one to see how that is playing out.

Justin Grammens  14:27  
It's awesome. Yeah, you know, it's it makes me think about people that always sort of said in years past, oh, all these creative areas AI will never be able to do and it feels like we've been able to knock those barriers down. Right now. AI is writing music and GPT threes. You mentioned it can write poetry. And now it's going to be able to potentially write code. You see any, you know, the damage to the way that like our livelihood, our productivity, like what we do as humans being threatened in some ways?

Erik Zwiefel  14:55  
That's a tough question for me. My initial response is yes, this worries me Then it's going to impact our jobs. At the same time, I look at technologies that have come in the past, and I'm sure people at that time thought this is really going to impact our jobs, perhaps that typists were first really worried about the computer. And they thought that now we're never gonna have to type anymore. And maybe in the short term, it does impact some specific jobs. But I think it also frees you up to focus on other things. It is an interesting one, as you start looking at different use cases, are we going to have to have changes to the way that economies function even? I don't know the answer to this, but definitely an ethical and philosophical, interesting area.

Justin Grammens  15:38  
Yeah, for sure. We actually had somebody at one of our applied AI meetups last month was talking about this thing called AWS code guru, which was, I don't know, it was, applies more security and coding best practices, really. So it tries to write code that maybe has taken a look at, or reviews code, I guess, for security concerns. And I know there's lots of security code reviews that, you know, you can pay consultants, 1000s and 1000s millions of dollars to come through? And would it be great if you could have an AI do this, but it feels like still today? In both of those examples, your example and that one, there's still a human doing the bulk of the work. This is just more or less assistive technology in some ways, would you would you agree with that?

Erik Zwiefel  16:21  
Yeah, absolutely. I think that's what we're seeing now. And these technologies that are out there, they're really assistive, but it is still incumbent upon the human to kind of guide that and shape that and make sure it's appropriate. I think it was even the late 90s. There was an article on popular science around generational or excuse me, evolutionary algorithms, and how it can solve these interesting problems with evolutionary algorithms that I remember the writer then said, in the future computers will be able to answer any question we have, then it will be up to us that the real challenge will be asking the right questions. And so it's kind of as we continue to move forward, that seems to becoming more and more true.

Justin Grammens  17:03  
That's totally true. Yeah, right. Now, you can plug anything you want into Google, and it'll give you an answer. But have you actually written the right question that you're looking for the answer? Yeah, exactly. Cool. Do you read a lot about AI or science fiction? Do you have any, any books or anything like that, that you might might suggest? I mean, you might just be busy as well, and just want to relax and read other stuff. But always curious to find out what people are reading or audio books they thought you're listening to today?

Erik Zwiefel  17:30  
Yeah, so I read all the time. I absolutely love reading. And one of the books that as you're mentioning this comes to mind is I don't know if you're familiar with these a fictional writer, Brandon Sanderson does a mate like my favorite author by far. But he has a book in there. Or in his series, it's a young adult book called skyward. And actually, it's a series there. And it's got some interesting AI character, or one interesting AI character in it that the, I'd love to see that. But yeah, I do really enjoy reading about AI. The other one that was really interesting to me, I read this about a year ago, I can't remember the author's name, but it was a fictional book called after on, it was interesting, because it starts to get at the intersection of quantum computing and AI. That is an inflection point. For me, that's going to be interesting as we start to see quantum computing become real, and AI being applied on top of quantum computing, that we could start to see generalized artificial intelligence take off faster than we can imagine, just based on the way that quantum computing works. Now, I don't know for sure, I'm not making any predictions here. But it was a very interesting look at that book on, you know, what happened when basically artificial intelligence became self aware, and an interesting thought process on what might happen

Justin Grammens  18:50  
there. Your background is in you said was in biochemistry, is that right? That's right. Yeah. I don't know a ton about quantum computing. But there's definitely some chemistry involved. I think in that, I don't know if you can explain or talk a little bit about that at all. I was way off the subject of AI. But I wasn't sure if you'd read about that.

Erik Zwiefel  19:06  
Yeah, so quantum computing, I think at its core is you know, currently computers are based on binary, it's either on or it's off. Quantum Computing adds at least one more state where it's both on and off at the same time. And so now, that completely changes the way that computers can function as a result, the way that coding needs to be done. And as I understand that, I'm not an expert here. But as I understand it, you start to write code that's more based on probabilities. And your code has to deal with quantum uncertainty and quantum probabilities. And that's how it is assessing things again, if someone who was more versed in this is going to listen to this say, Yeah, you're You were almost right, but not at all right. But that's my understanding of it.

Justin Grammens  19:51  
We had somebody at one of our meetups. This was months ago, but she has done a lot of work in quantum and sort of like run the local quantum computer. You grew up here in the Twin Cities. And she was talking a lot about just the power of large datasets, right? And that the fact you'd be able to analyze a lot of data at simultaneously. And just the fact that yeah, you're right, I think it seems to marry very well with regards to probabilities. Because that's really what you're doing at the end of the day with models, right? You're training things, showing it a bunch of stuff, and then understanding understand where things will land and just the vast amount of data that you can process through a quantum computer was going to sort of be a right sort of, like, turn the tables it felt like Yeah, from my recollection. Yep,

Erik Zwiefel  20:34  
absolutely. Is what I've heard as well. Well,

Justin Grammens  20:37  
awesome. Awesome. You mentioned GPT. Three, though, you know, I was wondering if you have you explored or play around with that? Are you guys using that much at at Microsoft at all? Are you seeing some interesting ways that that's being developed?

Erik Zwiefel  20:49  
Yes. So definitely, in use a lot of Microsoft with injecting it into the different services we provide. And so I think you'll continue to see that I've seen a lot of interest from customers and understanding how they can leverage this more in their day to day, I haven't had a chance to work with any customers yet specifically on implementing that for their use cases.

Justin Grammens  21:10  
Do you know if it's open to everybody? Or is it still in more of a closed beta testing?

Erik Zwiefel  21:15  
If I recall correctly, I think there's a open area where you can kind of play with it on their website, but I'm not quite sure where it's at in terms of, you know, actually developing it, injecting it in your own models.

Justin Grammens  21:27  
I think you're right now that I'm sort of talking out loud here, there was an open API that you could actually pay to use. I did have on the programs a little while back guy that did this thing called the AI dungeon. And it was basically a dungeon master. So every sort of, like adventure you went on was completely different. It was completely made up. I use a GPT. Three, but they were one of the first people to be able to use it. Okay. Basically, I think they actually weren't had access to the model itself. But yeah, there was it was under very, very tight constriction. And most people using GPT. Two, which actually wasn't that bad anyways, to begin with, I mean, there was a lot of good things that had proven and shown a lot of a lot of power width. So, you know, it's just you're right, who knows? What could be 84 is gonna be like, and all that sort of stuff, someone's gonna get better and better.

Erik Zwiefel  22:14  
Yeah, absolutely. Well,

Justin Grammens  22:16  
you want to give a little, you know, information with your eyes to how people would reach out to you and connect with you online. Which social media channels or whatever you use, typically?

Erik Zwiefel  22:26  
Yeah, absolutely. Typically, I'm on LinkedIn. That's the only social media platform that I currently use. I think I may be, I don't feel old, but I think I'm a little too old to understand Twitter, I still don't really get it. So I just stay off of it. But instead a You can reach me on LinkedIn.

Justin Grammens  22:43  
Well, there are other things you wanted to talk about or share, I guess, with people that might be listening to the conversations on applied AI podcast and any other

Erik Zwiefel  22:52  
tidbits. Yeah. So I, I would just say, I don't want to turn this into a Microsoft commercial, but you know, happy to help out in that space. But I love working with companies in the Twin Cities, because it means no travel for me, but open to working with anyone would just love to explore if their particular use cases.

Justin Grammens  23:10  
No, no, it's it's great. I love having people share what they do. And regardless of you know what company you're working at. And so a lot of the whole point of this is obviously scattered, you know, talk about specific applications and connect people in the in the community. And it's very focused in the Twin Cities today. But we get listeners from all over the world. So I think it'd be definitely, definitely good to have people. So yeah, like I said, I have liner notes and stuff that we've talked about, you know, with regards to Brandon Sanderson and the Kaggle competitions and stuff like that, I'll have some some links for people to check out as well. I appreciate being on the show, Eric, for sure. And I'll be sure to keep in touch with you. You know, with regards to Microsoft, I was gonna ask you offline, but now that, you know, since since since you're on here, do you work at all with John Cohen's familiar with him?

Erik Zwiefel  23:57  
Yeah, actually, John just joined my team. So now we are direct counterparts. So he's going to be focusing a lot on the AI and ML space, particularly with a focus in IoT and how you unlock what we've been calling a Iot.

Justin Grammens  24:13  
Yeah, absolutely. Oh, that's, that's awesome. Yeah. So because he, when you said you were a global black belt, I think he that's his title as well, you know, made me think about him. And I've known John for for many, many years. And just to sort of like focus on the AI IoT speed and you know, speak that's, that's really where, where my head is at, because we do a lot of work on the Internet of Things at lab 651. And I also have another company called recursive awesome, where we focus a lot on machine learning and AI. And so, you know, this whole idea of IoT and AI sort of overlapping. It seems to be a new term. I write a lot about it, and I seem to find it, but I'm not sure if I'm just in my own little echo chamber. What's your thought on that term? And is it becoming more industry adopted? Is it too broad too generic? Yeah, I don't know. just be curious for to like maybe what what you're thinking about when you hear AI IoT.

Erik Zwiefel  25:04  
So I do think it's become more industry adoptive, where I've seen it in a few places, I think it's still in the early phases of that term. I think the use case itself, though, to really get value out of IoT data, you need AI, because the volumes are just too great. I mean, you can still get value if you're just doing simple dashboarding, and things like that. But if you're collecting this much data, you can really accelerate it by looking at, you know, predictive analytics and things like that. In addition, when you look at AI, as models get more and more, use case, specific, deploying to the cloud may not be an option. So we've done work before on vision scenarios, especially that if you have a vision model, and you want to use it in a factory, your networking folks that are responsible, your factory are not gonna like you sending, you know, 30 images per second over their network. And your users aren't going to like the latency. So really then figuring out how do we deploy down to the edge and run these models on edge devices and gets to be a very interesting exercise there in terms of making the model smaller, yet still continue to get the performance you need. And so there's just very interesting scenarios that you get in as a result that I think that AI and IoT are kind of like perfectly aligned there to help support each other in their individual endeavors.

Justin Grammens  26:29  
Totally. Yeah, that's interesting. You used you touched basically on edge, compute there as your as your as you're talking about it. And I think sometimes people think about just sort of this overlapping, but they don't actually think about maybe where the processing is done. Everyone's head initially just says, Well, this is do it all in the cloud. But really, this idea of moving more and more intelligence to the edge is something that has been talked about for a period of time, but now it kind of has to be done. Like you're saying, there there are these offline scenarios, there's even times when it's just not connected. I mean, I do a lot of cellular and so you know, while people want us to think that so it was connected all the time, it really isn't. I mean, I was actually out running the other day on my phone was just I was in the middle of city and my phone had no connectivity for no reason. So there's just a lot of things that you're going to need to make sure that happens sort of offline.

Erik Zwiefel  27:13  
Yeah, absolutely. That kind of offline processing. And then the other thing that is interesting to me, and I think it's one of the next key areas when it comes to machine learning that I think you'll see a lot of work on is federated machine learning, how can I train a combined model without ever moving the data to a centralized place? Because this gets around, you know, maybe you have data residency requirements that I have to work around, I have privacy requirements that I have to work around, and so on. So I think you'll start to see a lot of work on instead of moving the data around, can I move the model around and have the small compute just slightly update my weights and biases on that subset of the data move back or however that might work? I've seen a few research papers, but I think that's kind of an area that's ready for a lot more work to be invested.

Justin Grammens  28:01  
Awesome. Yeah, for sure. We've done some dabbling, and looking into Azure our toss right there. Do you guys have their own have your own real time operating system? Kind of starting to live at the edge helping people build out models there? Is that true?

Erik Zwiefel  28:14  
I believe so. But yeah, that's where you're kind of over the over the tip of my skis in that area of Azure.

Justin Grammens  28:20  
Okay, maybe you know, I'll have John on the show. I should reach out to him. Absolutely. Yes. If he wants to join, because I know he could talk about that type of stuff. More and more specifics around. There are toss and stuff, or what you guys are doing there? Yeah, well, awesome. Cooler. No, I appreciate all this. Great. This is great. And yeah, I can say for sure. I'll be sure to list you and make sure to have people reach out if you have any questions and I appreciate you being on the show today and look forward to keeping in touch.

Erik Zwiefel  28:46  
Yeah. Thanks for having me. It's been awesome.

AI Announcer  28:49  
You've listened to another episode of the conversations on applied AI podcast. We hope you're eager to learn more about applying artificial intelligence and deep learning within your organization. You can visit us at applied ai.mn To keep up to date on our events and connect with our amazing community. Please don't hesitate to reach out to Justin at applied ai.mn If you're interested in participating in a future episode. Thank you for listening

Transcribed by https://otter.ai