SingularityNET: Deep Funding | Awarded Teams – December 22, 2022
All right, welcome everybody. Good to see you on this team meeting on the 22nd of December. Welcome everybody. We have at least two presentations for you today from the teams from VOTREC, but first from 0 to AI. And I am personally very much looking forward what they have to offer. So, Subir, can you tell us anything about it? I’ll let Barack write this for us. Sure. So, thanks a bit. So, you know, today, you know, we’ve been working on a milestone where basically we kind of provide an automation around ingesting the services that currently exist on the SNET marketplace and then automatically kind of, you know, populate them on the 0 to AI platform, which runs on the mobile phone and you can kind of, you know, consume those and make them, you know, much more accessible. So, again, today, this is working progress. So, we’re going to show you, you know, some of the services that appear on the 0 to AI platform and how do you kind of like, you know, access them, interact with them, and so forth. And then if time permits, we can kind of just give you kind of little, you know, what’s happening behind the scenes. So, looking forward to feedback on this, I’m going to turn over to Chiranji who will present the demonstration around this. Thank you. Cool. Looking forward to that. And yeah, take your time. We don’t have 10 presentations. So, you have some time. Yeah, Chiranji, whenever you’re ready. Okay. Let me know if you guys can see my screen. Yes, I can see. So, yeah, as for all the mentioned, what we tried here is to surface SNH services on 0 to AI app. And we classified it as SNH service. So, we can do a quick search. And it will list all the services that are under SNH. Yeah, I guess so. So, we can select any one of them and try it out. So, for now, we curated two services or rather we enable two services. So, I’ll select one of those. One is semantics segmentation. So, it will take us to the screen and I can click on try now, which will prepare the app, the service so that I can interact with the SNH service. And then we will do it in a minute or two. It should prepare the interface to do interact with the semantics segmentation. Yeah. So, yeah, as you can see, so it is showing inputs that were defined by the semantics service, so I’m going to just select an image, click on the visualized and submit. Yeah. So, right now we are using free token to speed up the response time. And you can see if it does work properly. And on top of this, we can now share this with other people, just generate a shareable link. Or we can edit and push it to open. Yeah. That’s pretty much it. Any questions, Parag, if I missed something? No, no. So, yeah, I mean, you know, again, on an app, you know, you know, you know, you know, you know, you know, you know, you know, you know, you know, kind of extracting information that, you know, all the services that exist on the SNH marketplace has been captured, translated. And now they are running on the zero to I platform, which, which allows now anybody having a mobile phone to be able to go try it, provide feedback. You know, you know, sort of personalize it. You know, you can kind of give it your own name, give it some color scheme for the app, you know, share it with family and friends, as well as the output that you generate. You can like, you know, very easily share it on a social media platform. You know, so this kind of enables the interaction collaboration and so forth. And the last thing that, uh, children, you mentioned is that you can, you know, also publish this, you know, actual running app on the NFT marketplace on the open seas today, where one can go look at it, not just as a pretty picture of the app, but actually a running app that by clicking on it, you can, you know, interact with it and provide, you know, like dislike comments and so and so forth. So it kind of opens up the accessibility aspect, you know, and as we go forward, we’re going to be kind of, you know, auto testing them auto ranking them, you know, collecting all this feedback on the blockchain, you know, kind of keep the trust in the, you know, the app of, you know, how well it works. You know, it works for, you know, uh, for certain images, it doesn’t work for well, you know, so the developers can go and improve and, you know, kind of, you know, continue to improve and. And for all if I made right so the mechanism in the back end is as follows, right. So we had with all of this programmatic, we collect all the metadata that exists from on the, you know, single-outed at market place, they’re 80 plus services, we fetch them automatically, we put them in the right structure, and we run this sort of incremental process where we take, you know, all the metadata, all this, all the program, we have a file set, etc, we generate intermediate representation, which is JSON from which we actually create code on the fly, both the front end code between calls, radio, the back end code, which is by turn wrapped and the D. R. So we see client side code, right, creating proper images for them to be runable and we are for specific is enough zero UI network, all of those things are done incrementally and programmatically, and the nice thing is that steady state, what’s going to happen is, if new application show up, new services show up on single-outed, they will also automatically show up on the zero UI network. Wow, that is not a good part. Yeah, I think this is a good illustration of the way we see our platform as a decentralized things, no, we have our marketplace and we have our platform, but anybody can build on the platform, you could also build your own marketplace on the platform, if you would like to. So this is actually very interesting and early use case for the service on our platform and it also feeds into a little bit into idea our ideas of having a kind of reputation system on services on our platform, so you’re already getting ahead of us there by doing it on a blockchain. But what is still a little bit abstract for me is exactly what will show up on open see. Especially. Great question, great question, should not be abstract, you can be sure something on open see, what the open see experience looks like, this is in context, so I think it will help people sort of consolidate all these ideas and point this. Yes, and for instance, also what would happen if the output is textual, for instance, or would be for more technical. And then you talked about the interaction on open see. Great, great, and then there are a bunch of things changing on a game, like one by one. Sure, so should I go with the open see publish activity first. Yeah, I can just show the process just show something that’s already published what the experience looks like. And then the second question that you had was this is an image right what about a text or something else, what is the input output look like. Okay. Open accessing a newer version. Okay, so yeah, this is the open see page where we host all the. An update that were published from zero to AI app. And this just gives a small or other speech out of how the UI or what the user has created. The user’s creation may be some background or how the user has personalized that and then there will be. Okay, so as it see there was a link which will then redirect user to zero to AI apps where. This app will be redeployed and user can play with that. Okay, that makes sense. So the interaction doesn’t happen on open see, but if you click on the created image, then you are redirected to the app and there you can. Do create some additional things and have the have the real experience right. Right. Yeah, yeah, makes sense. Okay. Yeah, I think this has huge potential and already seeing that overview of surfaces on your app and that being dynamic and updated. And I think this is a great extra front end and potential to reach out to new customer groups and make ourselves known. So I’m really, really, really glad with this initiative and I expect great great things from it. Thank you, right. So hopefully we will open up sort of you know you obviously have your own marketplace now everybody with a smartphone can live with an interact with it and provide feedback right. So we talked about the feedback and I think it dies in with the reputation things too right. So I know you guys are building stuff but I think what we’re doing we are recording all these things in the ceramic blockchain. The user interaction and I think we can feed into your reputation right maybe you will be the front end for the reputation will give you data for sort of computing the reputation. Oh yeah, reputation can consist of many many dimensions right many. And this will be one of them. We have users and users have DID apps have DIDs we have done some exercise in that respect I think we can supply those links together. Yeah, and I mean press that you’re already taking that so seriously that you are actually registering all that those interactions on the ceramic blockchain already. Which is yeah, amenable. Also the text text you know a text tab please. Okay. I didn’t dare to ask for another one I wasn’t sure what I was not available. We do we do have a text for. So you can right so even outside of SNF right we have our sort of phone services that we surface of hugging phase open window right and some custom ones. So these are all added tape right or everything from SNF also can added to the list of things. Yeah, so what I’m wondering now I think I asked this question before let me first look at this let’s first look at this at georgit. Okay. Yeah, so this is the text to text which summarizes block of text so I just copy a text from a site. This is circulated this was. So yeah, and now so now it’s actually making an API call to a service that is on our platform right. Yes, absolutely. Great. I hope this app when it’s finished will be downloaded so often and used so often will be very good. Yeah, that’s the goal. So again, you know the power is that you know the automation behind right tomorrow the 10 surface show up on the. The next day they will show up on the zero to our platform and you know being on the mobile you know naturally you know mix it you know much more accessible to lot wider community. I have one question every service on our platform has like currently like 15 three API calls but after that you have to call. And those services are from different organizations so yeah in worst case scenario you would have to make. Create an escrow for each of these organizations so how are you doing this or planning to do that in the future. Great question right so each organization will have its own sort of payment channel etc etc. I think for now we are planning to sort of use our part of our grant right. Yeah, because I think 15 may not be enough people may need to play with it longer so those things we will work out over time right and maybe there’ll be sort of a layaway plan people can sort of. Yes, so we will see how the how that goes once there’s excitement then I think we can probably figure out how to sort of you know make sure people are paying for these things right and the payment will be sort of mediated to zero to I will collect payment maybe not necessarily only through blockchain mechanism to be out and then translate and things like that. Yeah, so first of all this is a perfect way to use 25% that is reserved for API calls really perfect so that is what it’s meant for. I can also imagine that you can create some deals with the creators of those services saying yeah we create a perfect. Shopping window for you are you call it so maybe we as zero to I we would like some to have some more calls for free and then make an agreement that every user in your app will not have more than XX mouse because I imagine that whether whether I am on the app or somebody else is on the app. So that’s for the API call on the platform it’s all the same user correct. So I’m afraid your sound is really bad or who. Sorry. Can you please repeat that. So I call to a certain service then the service would recognize the endpoint the app as an endpoint and the app would be somehow registered to create that call but whether I am in the app or somebody else is in the app pushing the buttons is probably not visible on the marketplace. So that would mean that not every individual user could consume an additional 15 15 or whatever free amount of course. So I think we still need to work out a lot on the kind of payment side of things but our initial focus has been more around get more hands on it like put it out there where anybody wants to play with things it’s the accessibility is there. So there are work remains to be done on how we simplify things on the payment side because even we noticed that when you know here we are getting you demo with this free token aspects of it when you do a payment you know go through the payment channel it actually adds to the kind of even time it takes. So you know for it to kind of come back and render and things like that so there are there are areas that need to kind of iron out and work on it but but the goal would be is to we need to make this thing you know very simple and easy. So I think that’s what we’re doing is to make sure that we can use it for the users. Absolutely no I totally agree with this approach and I think what you’re showing is wonderful. So we can solve that and maybe also on our site we can help you with with thinking about proper solution. Yes yes whatever what I’m thinking one one direction would be that we would somehow propagate these 15 or whatever any any service provider will configure there. So services would propagate that to each of your users but that would require a little bit of different enhanced integration I would say and a certain amount of trust which I don’t see as a problem. But that would be one direction for now indeed by all means you can use the the 15 the 25% tokens for this that is not the part that is not used on hosting. And yeah I can also imagine agreements with individual individual service providers but but that might also if there if there are more and more service providers that would also might require some automation at some point. So there’s a direct and there may be down the line there may be tiered of services right the apps you know certain apps are you know free all the time and then there are some advanced features or app that can be you know that could be charges and things along the line. So that’s what I’m I’m thinking about indeed we could say that your app is free but on the condition that each of your users would get no more than X amount tries to avoid that it’s a loophole to create unlimited free. Absolutely right. All right interesting stuff but very very yeah interesting and helpful developments. No thank you I mean you know again as one emphasize this will continue to improve and get better you know but I think this whole notion of you know being automated is the key even the the UIS fact is completely automated right it’s kind of attached on the fly. Yeah that’s wonderful and it also shows a good how do you call it. Damn it. The UI of different apps are somewhat similar I can’t think of the word right now sorry and but that’s a good thing so people know what to expect and it’s easy to to handle. It’s always you know say three ways you know yeah and then this also will enable lot more developers right not every team has expertise on all the all area so people with the ML expertise they focus on you know building their models and let the platform to the rest. That brings you know huge you know kind of power of you know them being able to you know kind of you know give that usability aspects without anybody about. Yeah maybe you can stop the screen sharing now if you want. Sure thank you. I’m not sure either whether we should go into this park but I’d like to mention that we also had discussions on doing this the other way around and using the app as an own ramp for the platform and making the onboarding process easier for developers and of course. There will be a while and there will be complexities and then maybe we need to sort out the do some kind of feasibility study also there but. Yeah the idea behind it and the concept behind it is already fantastic and yeah if I think if we could make that work somehow that would be a huge asset and and add it to you to our platform for sure. Yes absolutely now we are we are excited about it and we are very confident we can make this happen we have to go through some small you know POCs as it goes with everything but I think they know the potential is you know. Reminders there wonderful wonderful thank you so much for this presentation and i’m very glad to have you on the team and looking forward to. See what what will come out of of this I think this is just the start. Thank you yes thank you. Alright okay can Rick from photograph you mentioned you also had something oh by the way there’s anybody else have some questions to brag a severe about this app. If not then we’ll just continue with with generic presentation. Yeah well nice to be able to join you all so photorec is getting close to launching its first service which will be. Risk profile assessment of machine learning outcomes and we’ve thus far submitted three reports that focused kind of on the development of the tool itself and we were progressing with the integration but having challenges with it we brought a new. Our team member Blake anderton who’s digging into the integration issues and then once we’re through with that he’s also going to do some work on improving our models for our second project. But I thought I’d take today to kind of introduce an interesting example of use of the risk profile. So what we’ve done is we did some analysis of the US House elections and I presented this at a call as at the Cardano summit in Washington DC so today I’m just going to show one slide from that work and how that will be one of our services. And then we’ll just go to the screen here so on the left you can see a diagram of the final forecast that was given by a team called 538 which does pretty rigorous statistical analysis. And we’re projecting that the Republicans would win 230 House seats that’s their mean here but they also had a wide distribution and so within the 80% range that included the possibility of Democrats winning and Republicans having what was referred to as a red wave where they would win a lot of seats. However, the news media kind of picked up on this and just prior to the election there was this big anticipation of a red wave of Republican wins. And then when the win by the Republicans was much narrower and even as seats were being counted the possibility of the Democrats still winning there was this then sort of counteractive narrative that the forecast had been very bad. However, I think what actually happens in these situations is that people who are trying to interpret the forecast don’t clearly explain or understand the uncertainty. And so one of our one of things we contribute here is to actually measure how the forecaster did in not just forecasting wins and losses which would be a classification percentage. So for instance, in this case in the end the classification the classification performance was 96.3% correct. It’s actually pretty good. But even more so than that what we want to do is do analysis on how good they were at assigning probabilities to that win. So for each seat they signed a probability that someone would win. And on this turn on the right here we show the actual ratio of outcome probabilities I’ll come back to what that means in a moment. And then on the x axis here we show the forecasted probabilities. Now what we’re doing is we’re creating bins of approximately 15 of these house seats into a bin and assigning an average forecast for each of those bins. And then within that being we’re calculating the ratio of the actual outcome of the net those that won versus the total. So that gives us a way of comparing the outcome versus the forecast on the probabilities and an ideal forecast everything would line up along this dashed line here. And you can see for the most part with a deception of two bins that’s all very close. And so overall 538 team assigned a probability of 85% to each of the winners. And so that says two things one this is actually it’s a much harder thing to do than the classification. And this is actually quite good performance but even within this very good performance what people are missing is that there is uncertainty there. So for instance this 85% kind of maps out to where they what they were predicting here that there was a 84% chance that the Republicans would win the house. So we’re looking forward to applying this to other applications. And in fact one very first application will be our own second project where we’re developing a machine learning algorithm that incorporates models of risk into how you train the algorithm. Now one thing here for this presentation to keep it simple I only included our accuracy measurement. But we also include in this analysis a measurement of the size of this which is similar to the classification performance and a measurement of robustness which kind of does a measurement of how you did in case you were going to get a measurement of how you did in cases where your performance was not good like these outliers here. So we’re looking forward to launching this soon and look forward to some feedback from the community about it to utility. And I will well I guess I’ll keep sharing for a moment if there’s any questions about the application that I’m showing here. So we can chat about this and other potential applications. That’s very cool that you were able to apply this already to such a concrete and high interest case. Basically your finding was that the forecasts were well done but maybe not entirely well communicated. I think that will always be a challenge together once the message out. But it’s interesting to see that it’s applicable and that it lines up so well. Yeah and we’re also going to be applying this to our work on learning models of images and looking at there the robustness of those models. So yeah and we’re working through the integration issues those the integration has been challenging for us but singularity net has also been providing good support we had a nice call with them this morning. So we’re hoping that the finding this very soon. Yeah so we are using all that feedback to improve our onboarding process are a number of initiatives going on there are also other priorities at the moment in the background happening so we can’t always do things as fast as we want but it is very helpful to get feedback from not only you but also other deep funding teams that are thinking or working on onboarding the services. Of course we have zero to AI who might be able to help us with that service and there are also other things that we’re looking at even to something simpler simple as improving the navigation and the the the documentation a little bit to make the whole process easier to understand. There are things we can improve definitely something will be complex and but far reaching and and something will be low hanging fruit as they call it which might actually also already have an impact so I hope that and I’m certain that one effect of deep funding will actually also be that will help us improve our tools. So we’re going to do this direct communication around it and it says not all down in a day but it has our attention and more than that we are working on it so we expect to see some progress there in the first quarter or two quarters of the next year. Right thanks a lot to Kenrick anybody else has some questions or remarks here. All right then Robert you had a very nice presentation last time I’m not sure so I assume that there’s nothing new yet. Quincy maybe you almost always have something to say and to present the so it would be fine if there’s nothing right now but. I’ve got a video but that’s fine. All right all right then thanks a Kenrick and a parac so be here and here and for your contributions and with that we’ll stop with the recorded part of this meeting.