 Our next presentation is by Noah Hassan Mustafid talking more generally about advancing Noah's mission with support of AR. As everyone knows, Noah has come early to the game and has a great story to tell. Hello everyone, my name is Hassan Mustafid and my colleague Greg Tosik from the same agency, the National Ocean and Atmospheric Administration. We are pleased here today to share with you how we are advancing AI in support of our organization or our agency mission. This is our outline. In the first slides we're going to talk about the big data, the Noah big data, then the Noah strategies and then we'll focus on AI strategic planning and I'll be sharing with you a few examples of AI implementation with Noah data and then to open a few questions to the audience or some discussion for the audience and then I will end with some enhancement. So talking about the Noah data or Noah big data, it's amazing how much data has been collected. Instruments are at work daily gathering ocean data from current speed to the movement of schools of fish and much more from acoustic, fisheries acoustic or site scans can send our multi-beam instruments as well as audio or video and from different platforms and crewed platforms, satellite. It's about 20 terabytes of data collected daily. Of course this represents a new frontier for understanding or opening the end-down of the oceans but also generating a larger volume of data that can overwhelm traditional data handling and visualization method. There is needs for more efficient processing of these novel data streams and needs for high performing computing communication. In this slide I put a few couple maps there. The first, the top one is the integrated ocean observing system environmental sensor map. This is basically representing all the all the sensors that are active within US and other parts of the world. Now you can see there is some inland ones there and then also because we are partnering with other agencies and they're representing their data in there too, their sensors. The map, certainly this is very interesting to explore such as Google I use environmental sensor map. You should be able to get there and the bottom one is representing the sensor, I mean the assets, observing assets there. There is some lines you can see in the left in the west coast. Those are gliders or encroached vehicles doing those surveys and then they overlap on a global current model. So talking about the big data, we have been talking about big data for a while now and think we are coming to come into an end as the focus shifts from how we collect data to processing data in real time. Big data is now a business asset supporting the next errors of multi-cloud support, machine learning and real-time analytics in other words. We are moving to emerging era of AI or artificial intelligence. That's why we are here today to talk about. So I'm sure agencies around the world are also trying to figure out how to deal with emerging technologies and NOAA is also doing the same thing. And here we are showing six NOAA strategies that were released last year, early last year. The NOAA cloud strategy, the NOAA data strategy and crude system strategy, NOAA citizen science strategy, the omics or DNA genetics strategy and finally the NOAA artificial intelligence strategy. All of it is in the science council website if you are interested to explore them more in detail. Talking about the NOAA AI strategy that is dramatically expanding the application of AI in every NOAA mission area by improving the efficiency, effectiveness and coordination of AI development in users across the agency. As I said it was released early last year and it's available in the science council website. The NOAA AI strategic plan basically this is making that strategy or moving transition that strategy to actions and this is a strategic plan for 2021, 2025. It identifies specific actions with leads and target completion dates and also synergies between or with other strategies that I mentioned to you in the previous slide. This one was released early this year and it's also available in the science council website. You have a link there in the bottom of this slide. So let's move to the goals of this strategy, the strategic planning. The goal one is to establish a NOAA center for AI, enable coordination of AI, research, algorithm development, data acquisition, applications, information exchange and awareness. Other functions would be maintaining the portal and open source of government application, hosting training event and workshops and facilitate new partnership and of course leveraging across strategies, especially the data and the cloud strategies. Second goal is to fuel the research and development component of advancing AI and assess gaps for AI expertise and asset across our agency and then identify solutions to fill the gaps, prioritize AI based approaches and support NOAA research grants and federal funding opportunities and other form of grants. Establish an annual research and development price competition and finally improve algorithms and evaluate model performance. A goal three is basically transition those R&D projects into operations at operational capabilities and you can see in the left side there we use a readiness level to move from O to R and from R to O. Strengthen and expand AI partnership as I mentioned earlier in my first slides is that partnership is a key and collaboration between different partners is a key to advance AI. We are partnering with academia with private sector such as Google, Microsoft, NVIDIA, Amazon and others. Then the last one is compete and assessment of training needs and create an online AI learning and support center. Basically this is to promote AI proficiency in the workforce through our internally or externally through different program that we have with academia partnering with academia and then advance of course education programs to address AI needs. Then my next slides is more about showing you or sharing with you there are some applications that we have for different type of data that we collect in our agency. This one is to address the need for climate predictions and uses an unsupervised learning to detect different regimes in the model dynamics. It's a neural network that classify regimes through training with surface data from a climate model or satellite observation useful for assessing things like heat transport, important for predicting climate variation. There is more details in the paper on the bottom there. There is a link so in Wally I'll 2019. The next example is the use of a neural network, trained at no linear ensemble mean to improve global wind and wind wave forecasting. EM stands for in the maps stand for standard ensemble mean and NN stands for neural network nonlinear ensemble mean. The red is over estimation and blue is under estimation compared to ultimate observations. You can learn more again from this last year paper by Campo Cial 2020. And the next one is to address the need for the beach gurus for safety reason and that is to detect the rip current through the constant imagery and this is used it's again a partnership between our agency and scientists from academia in this case here University of Santa Cruz and the machine learning that was trained here is a convolutional neural network to detect rip current and those images that's collected in the beach and this is to support rip current forecast model. Again there is a recent paper that was published this year by the Selva YAL for more details. This is an example of a tomated site scan sonar contact detection for safety of navigation surveys. It's an innovative image processing and machine learning technique designed to reduce the number of false alarms. This automated technique are directly applicable to port security preparations so we are able to get the dimensions and probability associated with with each contact as you can see in the in the one in the slides in the bottom or the figure in the bottom left right and left. A major challenge of coral reef monitoring is a huge data processing when we do 300 and 500 sites per year. Given the scale at which we work and the increasing threat to reef fish or sorry reef coral reef reefs we are placing increasing phases on imagery to reduce operational complexity and increase scope scale and scale but this necessitates development of artificial intelligence machine learning tools to extract data from all the new imagery. So this is again a collaboration between our agency and University of California San Diego that developed the coral net tool that is an online software package that automate analysis of bentic photo quadrants. This open access software is widely used by reef scientist and manager. It has the ability to generate fully automated or semi-automated annotations of coral reef. The next one is automated site scan or this is this one is more of sorry this is the image image inflow cytopods and it's an in situ automated submersible flow cytometer that generate images of particles taken from the aquatic environment and capture in situ images of phytoplankton. Again this is a collaboration between Integrated Ocean Observing System, Regional Partners, South Central California Ocean Observing System and Axiom Data Science. They developed the pipeline to improve the accuracy of phytoplankton species classification and submit submit the image flow cytometer data to inform about harmful algal bloom in a publicly accessible web portal. And so it applies the machine learning technique on the streaming data in the cloud environment that can increase the efficiency and the accuracy of data managers in the public for data managers public. Again there is links there if you want to learn more about this project. The next one is application to fisheries data. This is the case for scallop survey done by HAPCAM. This is the sixth generation of HAPCAM deployed on RIMAS 600 AUV and it collects a large amount of data. It's about the habitat or the HAPCAM bentic survey from the Northeast Fisheries Science Center in Woods Hole collected about five million image pairs per year. So manual annotation of sea scallop and fish from approximately two percent of the images are used to provide estimate for stock assessment and management. This is the application of artificial intelligence here is a convolutional NORA network CNN that was deployed on a VAMI software toolkit and it annotates the full image data but also detects additional species target and reduce manual annotator load. In other words most of these computers computer vision applications are the objective really most of them is to reduce the time of processing but increase the accuracy and precision of the detection of classification of the targets or the objects. Another example show you is the AU fishing example here this is the dynamic prediction system for illegal and reported and regulated activities in the Pacific and the objective here is to use the machine learning to identify any various vessel behavior across the Pacific as a function of suspicious AIS disabling and authorize the transshipment and overlap with protected or vulnerable shark and tuna species. So the benefit here is to provide low enforcement with data that need to look at and AU fishing activities. Again there is a link down there if you want to learn more about some of these applications. The last one is what you're trying to do with this one is during the surveys that we have in northeast fisheries science center for bottom throw surveys we try to collect imagery and data to build a ground fish image library to develop test and advance the application of machine learning algorithms in support of fishery electronic monitoring program. I think some of my colleagues from NOAA presented or are going to present on these applications for electronic monitoring. Then finally I want to talk a little bit open some challenges that we are facing with advancing AI in our agency and as I showed you those projects or those applications are very small and then the question is they are localized and not like everything is shared resources but they are specifically for specific data set or data type. Now when you talk about transitioning these all applications to operation that's something that of a challenge to to agencies so organizations. So we what I would like to do here is more of giving you a few options that we are thinking about we are this is still work in progress and the question there is what cyber infrastructure is needed to advance AI in our organization or your organization. The option option one would be to to make a centralized resource management of course there is pros and cons on that doing that the pros can have more resources in one network centralized and centralized maintenance and then the cons there is that when everyone want to have a space in there for himself then they it will generate some conflict or conflict may arise on task scheduling and work prioritization and the option is decentralized resource and management again there is pros and cons I will let you read through them and the last one is a hybrid between the two and so this is percentage of resource can be allocated. Again you can read through them and we can have a discussion panel discussion. With that I would like to finish with a few announcements one is the NOAA the first one is the NOAA workshop on leverage in AI in environmental sciences this is the third one it will be hosted at the NOAA center for artificial intelligence in Boulder, Colorado during September 13-17 of this year if you are as you like skiing this is the perfect time to be in Boulder, Colorado. The second one is the GPU hackathons this is a collaboration between NOAA and NVIDIA and this is the REN AI hackathon on NVIDIA cyber infrastructure and this event will be this year August 23, 29, 28 and September 1st basically two weeks first week to set up the teams and then do some applications and testing the data and then algorithms and then the second week is really doing the work this is a hands-on really a very good opportunity for someone who have already a team and they would like to train and test some of their applications into multi GPUs and NVIDIA provide basically there is mentors that help you to set up the system and get into the NVIDIA systems. I did one last year I hosted one team and we had really a good time testing some algorithms and training them on on multi GPUs basically testing them in how they work and how they perform in multi GPUs so I will encourage everyone to to look at this one all part of them thank you very much and if you have any question happy to answer and thank you have a good day bye bye. Thanks very much Hasan thanks for sharing NOAA's message and we note that NOAA has published the AI strategy that looks at improving the effectiveness and efficiency of work through the development and use of AI but NOAA seems to have a very strong commitment the agency seems to have can you give us some feeling of how did this commitment come to be from early 2014 we heard that they were investing in this and seem to have a lot of horses out of the gate and running what was it that brought this to such great attention in NOAA and maybe other agencies can learn from from your experience thank you Kim that's a good question there I think that if you look at how advocacy is well let's say stay within us there is the privates or private sector really it's it's advanced it's so far you know advancing and implementing AI Google Amazon all these private entities are are going so far but the government agencies like NOAA you know to that that produce the data and we know the challenges of the data of the data processing you know we we we kind of started you know thinking how we can move in the same level as these private you know companies and you know match exactly what we do in in terms of data processing but again the thing that moved NOAA to this level of producing these strategies and trying to you know to compete is that we have in a good timing we have a good leadership it's all to do with the leadership so team vice admiral team Gallaudet came into NOAA and said you know what this is what we are going to do we are going to move forward with this you know a new this emerging technologies it's not only AI but you can see we had to we had to look at all you know possibilities I mean citizen science came the last one and we had to edit as well but first time we we've been focusing on encrowed systems like because because we had a lot of already we had to figure out what to deal with that how to deal with them the encrowed system gliders all these platforms and manage systems we don't call it and manage anymore we called encrowed because that was something that we have to change but then AI also came into the mix because he was a meteorologist I mean sorry he was he was meteor but also oceanographer of a navy so oceanographer or navy coming into NOAA and saying you know we have to and he was an acoustic he was dealing with acoustic as well so hydroacoustic like sonars all these things and you know massive data from acoustics that comes into NOAA and how to process those data right now I can tell you I've been in survey for many many years many years in surveys doing acoustic but what we take we take only target species from those acoustic data what is the rest of a whole ecosystem that is in that data we didn't we cannot deal with it we just have like one percent two percent of data processing so so I think answering your question Kim is that a leadership is a key in moving things forward when he sees that there is a team behind him and supporting him so you have that team behind there but if you don't have a leadership you know you know pulling everything forward it's always been hard to do things in any agency so I think what you what you saw Kim it's it was fascinating journey and we have a new administrator now and we have a new new boss now for NOAA and he was he is fantastic Rick Spinrad he will again move things again forward I'm sure that all these emerging technologies are going to flourish as we move forward so Hasan it's it's great to hear about that foresight in leadership that was happening in the US but also your enabling private environment that enabling environment where there was a lot of skill sets and so on and the requirement coming through the data luckily innovation has been put forward as a major push for FAO under the new leadership that we have here under the general I'm going to turn over to Anton because we're talking more about policy questions and so on Anton do you have a question please for for Hasan yeah maybe because I know Hasan has a good background in in geo data so maybe he can bring some lessons from the geo people for managing geospatial data what would you recommend the AI people that maybe a few years behind a few months behind me in terms of data storage data sharing and data standards so maybe some quick reflection yes thank you thank you Anton I think that's a big question of course we are all trying to to wrestle with I mean I mean at the end of my of course my presentation was not presented all because it's maybe was long and we don't have time for that I I understand it but I had a lot of examples in there showing all those AI in action and and I think one of them well we're dealing with with models with with very very sophisticated models and we have to advance them geospatial models and all these things so the current situation with with advancing AI in NOAA is to do with a fact that we need a shared infrastructure we cannot have every office or every program have its own infrastructure or yeah we have to figure that out we have to figure out you know that first you need to organize your data okay you need standard for that we've been developing standard for for different type of data but also in other word we are calling it AI ready data and we have a whole working group on that right now is is is is been you know trying to figure that out and then since we we're we're establishing this center of artificial intelligence it will basically have a centralized in a role to make sure training and everything there and then making this data ready but again it goes to training data where what you are going to do with training data we are going to host it we're going to put it and so we have this the good thing about about NOAA is not taking only about AI I just one colon it's taking about it horizontally you are talking about cloud you're talking about big data in same time as you're talking about AI so when you are putting all this together you now you have the cloud ready you have the big data ready and then you have AI ready so that's how it should work it should have to go horizontally across all this all these you know columns whatever and so talking about answering your question about standards yes there is now enough standard for geospatial data that we can we can adopt or adapt but also it comes to when you're running these algorithms on top of these data you need a power okay now you're talking about power and speed and so the good thing what we've been doing is a partner with nvidia and we're using basically I I ran one in video hackathon last year and we brought a lot of it was how many images from coral reef because we've been trying to figure out coral reef how we can advance the coral reef segmentation and notation and then you know classification and to do that we had really to use the multi gpu system from nvidia and so we had teams coming all together in one place using multi multi gpu and you know using all that you know singularity docker all kind of that thing that nvidia provides you but also we had mentors from nvidia helping us to set all that so what I can say is that you need to partner this partner with with other private like nvidia or google or microsoft or arc gis or companies or or you know all that is going to help enable AI to be implemented in the right way and without that I don't think it's easy and in my presentation I have one slide that can give you a direction how to set up your hackathon within nvidia I mean it's just feasible to do it's not hard but I think if a you might probably think about how they can enable and facilitate you know this work but by making this partnership with nvidia and others and then creating these hackathons or code sprint whatever you can call them but those are going to help enable everybody to come and share and work together I hope that answer your question and on thank you very much Hasan that story about getting an AI ready ecosystem of all the parts is very insightful very helpful for everyone working in smaller teams thank you