Nick Pilkington, Unita, CTO discussed on Artificial Intelligence in Industry

Automatic TRANSCRIPT

So Nicholas glad to have you back on a making the business case episode. . We're GONNA talk about kind of ease of deployment at finding ways to integrate into existing enterprise workflows. . Before we do that, , you know your work, , a drone deploy vary some clients. . It's a much more robust kind of integration of the client's talent in your your talent of the clients tack and your tech others are plugging in playing it as a software. . Can you address that gradient for folks that aren't familiar with your? ? Business. . Absolutely I think when it comes to machine lending, , you've got this kind of spectrum of sophistication among our customer base. . There are some customers that have their own in house machine lending teams in training their own models they're got these models in hand and they're coming to join the boy saying, , Hey, , we need to plug this into our existing workflow. . And in those cases, , our role is a company is to make sure that we're providing the flexibility withdrawn deploy software that we can plug into accustomed existing workflow without disrupting, , everything and bring that kind of machine learning intelligence and efficiency gains to their whole workflow. . On the other end of the spectrum, , there are customers that less familiar with machine learning they still want to benefit from those those machining products and there are role is is actually take on the machine learning. . Task and to provide new machine learning based tools within drone deploy that just surfaced to those customers so where they would ordinarily just be performing vigilant SPEC with drums. . Now, , vision inspections are machine learning powered and they're much more efficient than they're highlighting issues automatically. . So we WANNA service on both ends of the spectrum there and I think the reality is still in twenty twenty production sizing machine learning tools is still a challenge is software that kind makes it easier and making. . Things more commodities, , but it's not just about training the model. . It's not this kind of fire and forget process. . We set something up once and then it just works forever. . It's really this kind of its ration- of getting something working, , adjusting retraining, , and it's all of that kind of integration work and the support for that whole model running in production that we want to provide to make it easy for our customers and I wanna go into what those elements are of making. . That production model work because that's the nitty gritty I think the business crowd who didn't go to Carnegie. . Mellon. . For Data Science they might have gone to you know working for business could really use some detail on but just just to flesh this out and put this in context drone deploy obviously you folks use drones is it primarily for Infrastructure Inspection Nacre? ? Should we frame your company in sort of a little bit of a wider sense here? ? So people know what you? ? Yeah, , we actually operate across industries. I . think Lodges Three Industries Are Construction. . Agriculture energy but there's this huge long tail of use cases in drums. . There is vehicle infrastructure this session rescue operations mining in aggregates just I think at this point, , we're finding that almost any industry in some way benefit from aerial imagery. . Got It in that context of training algorithms and deploying models in that world, , you know we we could talk about you know inspecting some kind of oil and gas equipment. We . could talk about inspecting crops. . Imagine you folks work the whole gambit there what are those real considerations for deploying models? ? Again, , we want to make this as easy as we can. . Especially, , you know you do <unk>, , you're selling the stuff. . So you WANNA be able to get this up in working for clients as quick as possible. . What are the factors that really have to fit into play that are sometimes quite challenging but you gotTa. . Work through them. . To actually make deployment work. . Absolutely I think the I is the is typically the machine learning problem that you're looking to solve how bespoke and how customers is there a bunch of very common patents in machine learning. . If you're looking for objects in an image, , you're looking to classify an image as good bad or exhibiting a problem or not exhibiting a problem. . There are kind of cookie cutter approaches to those sort of tasks in machine learning I. . think that's withdrawn. . Deploy can do an excellent job off the shelf by providing a bunch of machine learning technology that appliance can just plug into straightaway and reap the benefits of. . On. . The Mole bespoke side if you're digging deeper into these different verticals and more specific problems in more nuanced machine lending. . Solutions then I think it's much more of an iterative process and that's where. . John apply will typically deferred those customers to really understand that the main that trying to solve this problem and how to bring machine lending solution into production, , which usually involves getting a model up and running running on some type of infrastructure which drawn deploy can take care of, , and then working to improve what that typically means is letting it run assessing the outputs, , making corrections and retraining it and I think that's one of the pots of deploying machine lending solutions in production as typically overlooked. . A lot of people think that it's it's a one off process whereas it's an ongoing investments to build and maintain improve machine learning solution and I think that's where having let drone imagery in one place making it accessible to these customers. . Machine learning models plugged into the workflow from the beginning. . Makes that a lot more compelling a lot

Coming up next