Using Data Analytics for Good – Mike Flaxman, HEAVY.AI

This is texturung TV. Welcome to Tech strong TV today. I'm delighted to have Dr.

Mike flaxman product lead at heavy AI joining us today. You may be wondering how organizations can leverage Ai and big data to address complex challenges related to natural disasters emergency response and sustainability all while making a positive impact. Well, Dr.

Flexman has the answers for us. He's going to share his insights Dr. Flexman's a pleasure to have you on today.

Great to be here. Great. ai?

Sure, in terms of my background. I started out in the sciences and I ended up with a PhD landscape ecology looking at Fire and Fire behavior and along the way got more and more interested in public planning aspects of that. And so Taught in a number of planning programs eventually at MIT teaching and their planning program using GIS.

For a variety of planning short-term tactical stuff, but also my favorite topic was simulating sustainable Futures that I taught a course on that. It's really sounds fascinating and I myself my background is in meteorology. So I'm very fascinated by how you're combining Geographic analysis with machine learning or geoml so for those of us that aren't into this field.

It really interesting. If you could kind of break it down what it is how it works and how you're using this at heavy day AI. Sure.

So the it is an interesting topic to me because it's a combination of a lot of things people been trying to do for years. That was very difficult or impossible. But the Geo side of things is obviously keeping track of the variety of geographic information.

Ml is more famously applied at the moment to things like large language processing. That's that's all over the news, but there's an aspect of it that directly impacts the ability to do. Predictive modeling and to do that better and with higher resolution and more accuracy than previously loud.

And also I think a lot of people ignore the fact that a lot of data comes that's increasingly coming quickly is very messy data. It has a lot of errors in it as it comes in off of a raw sensor and so machine learning can be used to clean up the data as it's streaming in and that is absolutely a great use for machines because it's horrible work for humans. So we have kind of increasing large amounts of raw sensor information coming in and you know, you might have a satellite image that has a cloud shadow in it.

And so the the ml can help remove the shadow so that you get normalized data and then when you run a model on top of that you get it more accurate result. Can you explain how these analytics are used to tackle sustainability challenges? Sure.

So a lot of sustainability challenges are involved big data in space and time. And so yeah climate change, you're you're looking out multiple years and you're looking backwards multiple years to calibrate a model. So that's a huge box.

If you think about spatial dimension in one one access and temporal on the other that's a lot of data. and so I think what's new here, is that the Machine learning models and GPU analytics are starting to to learn how to deal with data volumes like that. Right?

So that's that's a nice consequence of all the stuff for learning from machine learning includes how to handle extraordinarily large data volumes and that gets applied then to sustainability planning. Because we need to understand, you know non-stationary environment. And so, you know, I guess Quick example, so historically flood planning was always done assuming that the climate was constant, right?

So you have a this idea of a hundred year floodplain. Well if the and that's based on the historic record in that case usually 30 years of historic data. But if you're in a period where the climate is actually changing not so easy to figure out what what a floodplain should should look like right because the answer to where the floodplain is is you need to know what time you're predicting it for.

Right? So we've gone from a world where we could use a static prediction to one where when you're going to go build something. you need to be able to have a dynamic model run specifically for the thing that you're trying to build so that you understand, you know piece of infrastructure putting in what environment it's going to experience the next 20 years or 50 years or whatever your time period is How can organizations ensure that they are collecting and utilizing data in an ethical and responsible manner particularly in the context of AI?

That's an increasingly profound and complex question particularly from the public planning side. Now that we have machines that know how to write and can write something that looks like a letter coming in from the public. You know, I'm concerned about some of our public involvement systems getting overwhelmed by by false information.

I think we take two approaches to it one is that the data provenance is ever more important? And so, you know, it's it's true today has been true for a number of years that a Hollywood special effects Studio can make an image that you can't tell from reality, right? But what's now true is that pretty much anybody can create such an image.

And so that means that you need data provenance in order to to make sure you understand where your data is coming from. That's the kind of first step in the gate and then You want information systems that can handle personally identifiable information in a sensitive way. And so for instance in our system we deal with a lot of personally identifiable information individual cell phone pings, but we aggregate them up into a level that's guaranteed to protect privacy.

And so that requires a little bit of user sophistication and input and a little bit of stuff on the software side to be able to make that guarantee, but we want to be able to use this very grainyular data but to use it responsibly in that case we need to aggregate it. It's the same ideas the census they've been doing that for Century, but it needs to be done. Dynamically on data that's coming in in the moment.

And so that makes it you know considered a little more complicated to do it. Right but I think we have the technology to do that. Right and we need to choose to apply it basically can you walk us through an example of how data analytics has been used to address a specific sustainability challenge or master Sure.

So my favorite topic is is the interaction between utilities and fire and fire risk sitting here in the Berkeley Hills of California it cuts close to home, but I've long been interested in this topic. and what we're doing with a with a major Southern California utility is to look at their weather model data and its implication on fire hazard, but then also implication on customers and so it's a it's a really complex set of policy decisions and calculations that go into that but in a tactical sense, they want to be able to preposition Assets in the right place. So if you see a you know, big wind event coming up you want to look at where you where you can put people in an equipment, but also there's de-energization of lines in the short term and then ideally obviously in the long term we make our whole utility Network much more robust.

And so then you have a series of long term planning activities. Such as vegetation management that you have, you know a schedule of doing vegetation management you want that to be risk-based and traditionally, it's just been scheduled based. So just the utilities making that turn to risk-based management to me is a really important thing for them and socially so it's economically more efficient it's safer, but it's a big transition because traditionally The Regulators just said you have to do all of this every four years.

They haven't said, oh you need a risk-based approach you need to make sure that the science is solid on the risk basis. You need to run your operations based on that risk. That's what the Regulators are starting to do, but it's taken, you know a decade to get to where we are now, and I think there's more work to be done.

For sure with the rise of edge Computing and The Internet of Things. How do you see the collection and use of data changing in the context of sustainability and climate change? I guess the big news there is that all of the infrastructure that's out there that we're used to being very very dumb basically concrete sitting there for 50 years is all being designed these days with sensors embedded and so I think that the the data streams coming from infrastructure are going to be very very rich and some of this infrastructure extends all the way across our natural environment.

And so there's a kind of two-way relationship. But for example, the utilities just mentioned they're all building out weather station networks on their transmission lines and with hundreds of stations. So that gives you a lot of detail where they're traditionally weren't weather stations like up in the mountains and mountain pass and so that can be then use flick in mind with satellite information to give a really Rich and detailed view of the current weather for instance the wind hitting the power line.

i. That's a great question. And there's a lot of them and so I guess it's it's hard to pick favorites but we've already talked a little bit about flooding and fire.

Those are obviously two things that need much more intensive management. I'd say water in general right drought as well as flood is going to need much more intensive management. And that's a really good use of this new accelerated technology beyond that.

I guess I want to bring in the social Dimension. So one of the things we can do now much much better. It's Target interventions for different communities.

And so for instance if you want to pick wildfire and you're trying to do, you know evacuation Basin wildfire. we know from the information technology where there's a community that's you know, predominantly Spanish speaking or that has limited access to to vehicles or whatever the issue may be and so I think we can do a lot better job at response if we know who we're trying to serve and getting more granular about that is a big topic even just within Emergency Management, right? So if we're going to adapt to climate change, we're gonna have more emergencies, but how do we treat those?

And then ultimately do proactive measures in advance. Let's such as you know Outreach in Spanish. For example, where does that go?

Yeah. Yeah. That's a good point.

I think meeting these goals for adaptation adaptations itself, you know big topic within the sustainability goals, but we want people to be more informed and to have more current and more locally specific information, right? So not only do I know flooding is potential issue, but I know hey the creek I live near is likely to flood and I know that in Spanish if that's my native language, right? Yeah, and it's it's more helpful and informative if people it's in contrast a better communication absolutely if we can get it in different languages and and make sure the messaging is is relayed just to kind of build on that for other organizations that are watching this and they're thinking, you know, I would like to leverage data analytics to drive positive change and promote more sustainability in our own operations.

Do you have any advice for how they may be able to incorporate that? Sure, there's always a kind of crawl walk run strategy. So the the crawl part is making sure that your your existing analytics can handle the data you already have access to and so a lot of organizations end up with these big data silos these days that are often some department or another and not integrated with their overall decision making so that's that's the easy first win.

The intermediate is to look more broadly beyond your traditional organizational boundaries and see who else is collecting data that's relevant. And so um For instance. We work a lot with telcos.

It turns out to be very relevant to telcos vegetation and it's the configuration of education and the moisture in the vegetation turns out to be what block cell phone signal and so they don't need to go out and collect all that data. There's lots of scientists out there collecting that there's entire public programs collecting that data. It's not traditionally something that they've built into their RF planning, but it's something that our tool brings into the mix.

So basically, you know you the telephone already know where your towers and antennas are but use public information to get the best available information on the current state of meditation. And so that's that's kind of the Walk phase. And then the Run phase is Predictive Analytics.

So then you can get ahead of things and say all right. What if we have a heatwave what happens then? What are we going to do to respond to that or a drought Etc.

So you can you can run the clock forward with these tools once you've got the information base in place. And you can do predictive modeling and then have your organization have a plan ready for those various contingencies. Sounds great.

Well, Dr. Mike flaxman product lead at heavy dot AI we really appreciate your Insight and your knowledge. Thank you so much for joining us today on techstrong TV.

It was pleasure. Thank you very much. Great.

Well stay with us. We're gonna have more coming up.