 Okay, so hi everyone. Today I will talk about our activity in Linux Foundation AI. This is a new group in Linux Foundation, a relatively new group. My name is OfferMoney. I am a director of product strategy in the Andox, city office. And I'm also the chairperson of LFAI, technical advisory committee, or TAC. And my goal today is that you will share my excitement about what we are doing and if you would like to contribute or have something to contribute either by coding or by contributing a project or joining, it would be great. So what is LFAI? LFAI is what we call an umbrella project that has one governing body or one founding body and many technical projects. All of them are in AI domain, AI, machine learning, deep learning domain. We started the foundation in March 18 last year and with one project and now we have already five and we are still growing, new companies are joining and a lot of traction to the project that we are hosting as part of the foundation. This is the mission that we define to ourselves. We would like to build and support an open AI community and drive innovation in AI, machine learning and deep learning and enable collaboration between the companies, the projects, the users and of course create opportunities to the members and to the entire community. And we broke the mission into a few goals. First of all, we would like to allow, provide a neutral environment to host those projects. We would like also, and you will see at the end of the presentation how fragmented the open source AI landscape is. There are hundreds of AI projects in open source and one thing that we would like to do is to harmonize and to have interoperability between the different projects. And this is something that we focus a lot of our time. The third thing is all the fairness and ethics in AI. I assume that many of you who know AI know about those challenges and we believe that it is very important to have a fair and non-biased machine learning solutions and I will touch on it a little bit later. We are building also some marketplaces. We have a marketplace for machine learning models and I will show one of our projects that this is the headlight of this project and also a marketplace for data. And of course, I said earlier that there are so many projects we would like to understand which projects are the most promising ones and to support them and grow them as much as we can. So these are our goals and this is the structure of the foundation. So as I said, we have a single governing board, a single founding organization that allows the projects to focus on what they do best to create code in AI domain. And in this way, the governing board takes all of the activities that are related to marketing, all of the activities that are related to legal and other stuff and allows the projects to work. Under the governing board currently we have five subcommittees. We have an outreach committee. This committee is organizing all the outreach activities or the marketing activities. All the LFI days. Yesterday we had one day here in the summit or before the summit started. This is one example and we are doing days in different places in the world. For the rest of the year, we have one day in China, India, Europe and other places. For this year, I think that's it. Maybe two in Europe actually. And this is the outreach. We have a legal subcommittee. We have a budget subcommittee of course and recently we started two new subcommittees, the strategy subcommittee that tries to re-address the mission and goals and understand that we are providing the right value to our members and the community. And the strategy committee tries to define the value or will define the value proposition of this foundation. Also we have the fairness subcommittee. In this subcommittee we address all the fairness and ethics related to AI. We are trying to define or we will not define what ethics for AI are. There are plenty of organizations and countries that already define that. But what we would like to do as part of being an open source organization is to bring fairness projects and maybe create some certification program which different AI projects will be able to certify themselves into the ethics that we will have. And this is what we are doing as part of the subcommittees of the governing board. The technical advisory committee, our consul, the TAC, is the committee that I am leading. We have bi-weekly meetings. Every two weeks we have a meeting. In those meetings we sometimes discuss the different projects. Sometimes we bring external projects to give us presentations so we learn what is going on outside. We discuss the collaboration between the projects and how we meet the mission and goals of LFAI. All of our calls, the TAC, are open and recorded. And if you are interested, I will have some links at the end of the presentation. And you are very, very welcome to join and contribute and be part of the discussion. The technical projects, I'll touch each and every one a little bit later and explain what they are doing. But the idea is that there is a separation between the governing board and the TAC and the projects. Each project defines its own governance. So we are not dictating anything on the projects. The only thing that we ask is that the governance of the project will be open. We ask for an open governance and that's it. Other than that, each project in this list, you can see the five that we already have and others are coming soon. I assume at least two, three more in the coming few months. So each project has its own governing procedures and the way they are doing their stuff. Okay, let's move on. Any questions? Membership. We have actually three tiers for membership. We have the Premier in general. The Premier is of course more expensive and gives the members a seat in the governing board. And we have the general. Just yesterday, IBM joined as a general member and I hope they will switch to Premier soon. But many companies that are leading forces in AI and many Chinese companies leading in AI. So we are only like 18 months old and we already have quite a lot of companies with us. I wanted to mention another tier, the third tier, which is Associate. Associate is a free tier for universities and non-profit organizations. And if you are part of such an organization or you know an organization that would like to join to such activity, we will be happy to get you into the foundation. Questions? Okay, let's move on. Some milestones that we already achieved. So we started this foundation in March. Back then it used to be LFDL, Linux Foundation Deep Learning. It was not the smartest choice of name. I was on vacation for one week. I come back, LFDL, okay. So we started in March 18. We had the first project in the foundation when we started with Akimus and then we had the Akimus Challenge. It was an open challenge for developers to contribute machine learning models and the one that... If I remember correctly, the one that won the first prize was about analyzing data to identify cancer. It was very cool. We had a few days during the year we got two projects from Tencent and Baidu, Angel and EDL in August, actually in this conference last year. Oravod, a project coming from Uber and then Pyro later. And in May, we changed our name to LFAI and now we have... Yesterday we had the LFAI day and we have a lot of other stuff going on. Okay, as part of the foundation, we defined three tiers for project maturities. The first one is incubation. It's for new projects coming to the foundation. They are still growing. They need a neutral environment to get more adoption to their project to grow. And for that project that would like to join LFAI, the only thing they need is the technical advisory committee, the TAC approval. And there is a process. You go to GitHub and you submit your project. Of course, we will work with you to do that. You bring it to one of our TAC meetings and we have a vote and approve those projects. Until now, we didn't reject anyone. But we are still growing and maybe at some point we will. Then there is another tier, the graduation tier. This tier is for the flowers that have a lot of traction from butterflies. Those projects that have a lot of contribution, a lot of stars, a lot of traction, users, committers coming from different companies, et cetera, et cetera. And then we have a TAC vote and a governing board vote in order to move the project from incubation to graduation. The reason that we need a governing board vote is that the graduation projects can ask for a dedicated budget to run their project. So this is the second tier. And the third tier is Emeritus for projects that used to be big and great. And now there is less traction to those projects. Maybe the technology is getting a little bit old and there is new technology coming and replacing this project. Less contributors, less committers. And then we moved those projects into Emeritus status. For now we have four projects in the incubation status and one project in graduation. But two of the projects are going to move from incubation to graduation. Okay, so let's talk about the project themselves. As I said, we have five projects. You can see the names of the projects and the company that contributed those projects. Achimus came from AT&T, Angel from Tencent, Edel from Baidu, Orvod and Pyro from Uber. And let's spend a few minutes to discuss each and everyone. But something that I wanted to say is that we would like to achieve a landscape of projects that will allow the users to build an end-to-end machine learning solution. And for that we will have some integration between the projects and we will have to identify where the projects are in the, let's say, machine learning workflow. And I'll discuss it a little bit later. So let's start with the first project. First project, this is the project that we started the foundation with. This is Achimus AI. It came from AT&T. The idea of this project, the highlight, is to have a marketplace for machine learning models. Companies can bring, or companies or individuals, anyone can bring a machine learning model into the platform. And then other users, other companies, can take the model and use it. When we started it, it was only for open-source models. But now we are working on the third release and we are already implemented some licensing so companies can share models with licensing associated with the model. So the platform is still open-source, but there is an ability to commercialize your models via this platform. Maybe I'll touch the different steps or the activities that can be done on this platform. So number one on the left is to onboard the model and we support many, many languages for onboarding. The second activity is to train the model so a model can come trained or untrained to the platform and if the model is untrained, we can train it as part of the platform. And this allows also some interesting commercialization opportunities because sometimes I'm the developer but I don't have data. So I can share my naked model, the non-trained model via the platform and other company or other organization that has data can train it for me. And then put it back in the platform. The third thing is the marketplace stuff so you can rate, you can do some other things to the model and also you can change different models so you can take one model that does one thing, another model that does another thing and change them into a full solution. And the fourth step is to run the model what we are doing is to produce a Docker file. So this is the first project. The second project, Angel, is coming from Tencent and this project came a year ago, exactly a year ago and it was a small project, not a lot of traction. Yesterday we had a presentation from the Angel team, from the Tencent team and they presented... they managed to take the platform so quickly, so far, it was amazing and I will touch it later when I'll show you the location of this project on the workflow but they managed to move very quickly, they have more than 4,000 stars in GitHub, they have more than 2,000 commits and many contributors and committers and more than 100 organizations using this solution. So this project is going soon to graduation. Another project, EDL, Elastic Deep Learning, coming from Baidu and I'll explain a little bit about the... you know what, let's say something about the solution so I'm not sure. Okay, so what Angel is trying to solve or at least what I knew until yesterday that Angel is solving is training in scale. So you have your model, you built your model, you have your data ready, now you need to train your model but you have a huge model and yesterday in their talk they presented billions of parameters in their network, machine learning network or model and they use parameter servers in order to train huge models and this is what Angel is doing. EDL, Elastic Deep Learning is in the same area distributed training but EDL is trying to solve a different problem, it tries to solve the server utilization problem so in many cases when you train a huge machine learning model either you exhaust your resources or you don't use anything because of the structure of the model and they're trying to solve it based on Kubernetes and some other technologies. Horavod is coming from Uber, this project has also a lot of traction in the market, many users, many companies contributing and using this project, more than 7,000 stars, many committers, contributors and this project also touches the distributed training and let's say the training phase and the distributed or training at scale and I will show it soon and the last one is prior, prior is probabilistic, help me, language, programming language based on PyTorch and this is another project coming from Uber and I like the number of stars two days ago so I wrote it exactly there, 5555, okay, so we just discussed five projects and I tried to explain what those projects do but it's a little bit, you know, it is not very clear and when we are trying to explain what we are doing or what the project are doing, it's a little bit unclear when you're talking out without a lot of context so we decided that we would like to build or define the ML workflow to define all the steps that one needs to take in order to build a machine learning solution and then to identify what our projects solve and what we are missing so we took a known flow, we took something that we have seen from CubeFlow, CubeFlow is a project coming from Google for machine learning pipelines so we took their model or their stack and we extended it a little bit and we ended up with this stack so we have three layers, we have the data preparation layer, we have the model creation layer and we have the rollout and the next exercise and maybe I should mention a little bit more and on the left we have a marketplace and on the right we have all the workflows and the orchestration and we also defined some additional layers, we defined the ethics management and the security management and now we have a good understanding what we have, if we have all of the components here or we have solutions that solve all of the components here we can build a system that solves the problem or can provide us a solution for machine learning problem and then we took our project and we put them on the map and what we identified is that these are the areas that we cover and now we know that for example we don't have open source as part of the foundation we don't have open source that touches for example the monitoring and some other things and also we understood that it doesn't help or it doesn't solve the fragmentation problem if you have different projects solving different aspects or activities if we don't have integration between them so what we started as part of this exercise we built that and now we are working with the different projects in order to build an integration between them so it will be easier for a user to build an end to end solution using our projects we also add some existing only some existing other projects on this map to see what other open source projects can help us in order to build a full solution some of what you see here I hope we'll join the foundation soon but it will take time any questions? yes so this is a good question we hope that the services that we give and the interaction that we can create will bring the projects and as I said at the beginning having an umbrella project or an umbrella foundation allows the project to focus on the project itself and the marketing, the legal, the governance can be done by someone else so I assume that this is why the project that came yes so yes I personally think that we have the responsibility to make sure that they are working together you know we cannot force okay I cannot say as the tech chair or the government board chair cannot say anything about the technical solution I can suggest, I can recommend I can try to create collaboration if one project doesn't want to collaborate it will not happen but I think we share most of the projects share the same vision and I hope that we will manage to do that we're still young one last exercise that I want to share with you is the LFAI landscape and it's a little bit similar to the MLflow but the landscape is a little bit older and it looks at the looked at the problem from a different point of view so what you see here on the right is the open source machine learning solutions there are 163 projects on our map not every project not each and every open source project in the world is on the map we have we define some minimal criteria to get in but we have already 163 projects with like almost one million stars actually there are a few projects waiting to get inside and we will cross the one million stars 67 organizations 11 universities and this is available online and I would like to spend one minute to show you let me I hope it will work okay you see okay so this is the landscape and actually when we started with it we actually took the landscape from CNCF if you know CNCF it's another umbrella project in Linux foundation focusing on cloud and we took their code and implemented it for us with our projects or AI projects and there is you can play with it so you can see the landscape you can move to card mode there are different categories, different projects the LFAI project are the big ones here you can see the five of them and you can play for example take different licenses or take different organization if you want for example to see what open source projects are coming from Google you can do something like that and you see all the open source projects that Google is contributing and this is basically what I wanted to share with you today so questions so the question was if I heard correctly what to do if I want to contribute what should I do so you can see here all the information you need in order to get involved there is the email and the website and the wiki and the github and the landscape this is the landscape and the website so all the information is here if you would like to get involved send either me or the info email and we will be happy to have you part of this activity I personally really enjoyed great people great environment to work with and you are very welcome more questions so there is in our github there is a process of what a new project needs to bring in order to show in order to get to apply to be an LFAI project and we would like to see what is the scope of the project what is the roadmap who is contributing to the project what is the traction of the project and recently I personally asked the new projects to map themselves on the landscape not the landscape, sorry the ML flow so we will be able to see if this is an area that we are struggling with or this is a crowded area so this is basically what we will ask more questions and if you have a specific project in mind that you would like to contribute come after the talk I'll say more questions ok, thank you very much for your time and see you next time thank you