Show Notes Wirelive #14
Andy Readman 0:16
Hello everyone and welcome to Wirelive number 14 talking all things, machine learning, with a very special guest, Adi Polack from Microsoft. Ad he works in the cloud scale analytics advocacy team. All Things distributed systems Big Data Analyst, analytics, machine learning at scale and has a particular focus on Apache Spark ML ecosystem as well. I think, welcome, welcome to wildlife. Thank you Andy lovely to be here. Yeah, so I guess we should probably start thinking we're gonna have lots of different people listening, listening along to the session or joining us that they live on YouTube. with a bit of a general overview I guess on what is machine learning, I know it's quite a broad topic but if we start started sort of maybe a beginner intro and then we'll work our way up to something a bit more complex as we go. What is machine learning to you, what's your what's your definition.
Adi Polak 1:09
I think it's a great question. So let me try and simplify it as much as I can. So, the way I see it machine learning at the end of the day it's kind of a statistical mathematical representation of different set of rules or different set of patterns that we might have seen in data, and it can help us to automate some process, or even drive innovations, and what is nice about it is that most of the times patterns are hidden, it's really hard when I'm working with analytics as a bi business intelligence and this LLS person. Sometimes I have to use a lot of personal or domain specific data expertise to extract this betters and machine learning is kind of like the next step for analytics.
Andy Readman 2:02
What's really cool. Yeah, I think the first thing that really surprised me when I got into into data and AI and machine learning is how it finds its own path to the answer for you, it's kind of taken all of the kind of questions we've proposed a business intelligence and things in the past and how big has been done in in business in particular and kind of industry. And normally we have some assumptions and we try and to try to prove what what the humans have decided is correct but I think, machine learning for me, opens opens up the opportunity like, like you're saying for it to tell us something new, and it brings that much, much more kind of fair, unbiased kind of opinion on how to segment customers or how to how to make those recommendations in new ways that traditionally we've been kind of, we've maybe had a blinkered approach as humans and we thought we thought we know the answer before we really known that, then send the data to prove it right.
Adi Polak 2:56
Yeah, I mean, you know, if you think about how people learn themselves, like you and I, how we learn, we might be looking at some reading a book or looking at someone's patterns, right, what they did, how they became, you know, where they are today, or if they did some of what the architecture that they build these are old patterns and we can identify if it was successful or not successful according to some, you know, business objective, personal objectives. So it's slightly like that, it's extracting patterns out of data that sometimes it's hard to detect as human beings that are based on statistics.
Andy Readman 3:39
That's cool. Okay so, Cloud Scale analytics advocacy team or the developer advocates, I think, for me, some of the kind of superstars of the Microsoft teams that we work with, I think, obviously, we know you guys go around the globe talking about the latest tech and very heavily involved in sort of building out the services as kind of an engineering team as well. Sounds like an ideal kind of role but what are the what is the day to day look like and what kind of does that involve that that role.
Adi Polak 4:08
This is a fantastic question and I think every person on the team is going to have a different answer for it. Just because every person comes from a different background and has different skill set and I think this is what makes it such a unique team when you bring a bunch of people that have different capabilities, and you make them build things together, so it can be, you know, either a meeting with the product team looking at the roadmap, thinking what the vision is what customers are struggling with. It can be a, you know, presenting at a conference of giving a presentation that really helps unblock existing and potential customers. It can be a presentation that it's more beginner intro to help more people onboard into, into the technology, or sometimes even into the field because we're seeing, especially now with COVID-19 We've seen a lot of people that are looking for new jobs and new career paths. And I personally am a big believer that tech is going to grow. So, if you know me as an advocate, I can help them. I can help them learn about this technology so they can create a future for themselves. So this is one of, one of the areas that I'm trying to invest as well so it can be giving a presentation. It can be working with a product team, it can be helping fix a document or some training. It can also be building tools so building integration tools so it will be easier for developers to integrate multiple open source technologies with the SAS services that we have on the cloud. Yeah, a bunch of good things. And every day is is different, really, every day is different.
Andy Readman 5:57
But I think it's a really important role as well, particularly the moment where we've got a bit of a disconnect between how many people know about these tools and services and are able to use them, and Microsoft to do a great amount of work towards kind of democratizing the technology for everyone to use open sourcing technology and the fact that you primarily work with a lot of open source technology as an employee of Microsoft maybe is unexpected if people know Microsoft from maybe 1520 years ago you know a lot, a lot has changed and I think considering. Very few people are actively using these technologies on a global scale across kind of standard businesses. Yet we've got so many people that are that are being consumers of machine learning from from brands and things, and social networks as obviously as a massive topic that we can get into I suppose so. But for people that maybe don't realize they're, they're kind of you, they're, they're using a service that's being driven by machine learning. Have you got any examples of social media and things like that that, that are leveraging that type of thing.
Adi Polak 7:08
Yeah, so we have a case study I can share the link as well. For grab. So grab is a ride sharing company that works in APAC. Maybe some of you use that in one of your travels, and they're expanding into food deliveries as well and what they're trying to do, they're successfully building, machine learning capabilities into their algorithms. So, I don't know exactly what specifically how but maybe we can think, if we would to build such a platform, how we can inject this, how we can use machine learning, in order to grow the business and make sure we're hitting these business objectives that we want to reach, like customer retention, maybe cost prediction, and giving better services in terms of recommendations as well. Right, so if you're using a food app and you want to order new foods, and maybe the application can learn what your what you like and recommend you the next delivery. Yeah, what do you think,
Andy Readman 8:27
yeah definitely grab I've had firsthand experience with as well. A few years ago in Singapore. If anyone has not traveled in Asia and needed a taxi, grab is the Uber, of the Asia Pacific region really and as much as Ubers available some parts I think grabbers is more popular in most, and it was the only option really to get to get a taxi in Singapore. Highly recommended really good, I think. Back then it was doing ride sharing better but and before Uber could and certainly that speaks to their level of innovation that they're working on Route, Route calculation and recommendations to be able to do that for drivers to know who they can pick up on the way and then the most effective ways of doing that because that's obviously not just standard mapping from A to B when you're starting to do sharing, which obviously machine learning plays plays a bit and I think. Yeah, I think lots of things people are exposed to systems and whether that's recommendations in retail, all the way through to, to things that are maybe a bit more subtle than that I think there's there's lots all around us, even ostrich intelligence and machine learning on our phones these days in the software is kind of understanding what apps we might want next before we've, we've opened them up yet. And we're becoming reliant on it as consumers, I certainly am not someone that's quite, I guess, quite aware of these things and I expect, I expect better I think a few years ago, I often bring this example up but I think Amazon used to recommend TVs to me because I bought a TV, and obviously I'm not a buyer and seller of TVs, I just wanted one for the living room, but nowadays, I think recommendations across all of retail are, are being made more accurately, I think, and more transparently which is really important as well so, because you bought this we're recommending this and that, that kind of approaches is really nice.
Adi Polak 10:17
Yeah. And you know it brings me back to. So there is an Uber article that the Uber engineering released a couple of months ago I also shared it on, on Twitter, a while back, and they refer to their specific specifically they refer to their machine learning pipeline of how they monitor the performance of the machine learning model. I think it's very interesting because it shows, innovation, and it shows progression in the way they're building their machine learning models so they build machine learning model to predict cost. So if I want to drive somewhere maybe I'll take one car versus the other type of car based on cost or, you know, maybe I'll do a different type of ride sharing. And then they kind of looked at together with what exactly that person paid, so they will have better accuracy in real time, and better monitoring in real time of their machine learning performance. This is something. Yeah, it's really cool this is something that I've seen, you know, people sometimes struggle because it. It's not in the books right. It's only my example, you have to try you have to be creative, you have to think out of the box to understand how you're closing that loop of observability and monitoring. And when I saw it I was like Yeah, exactly, that's, you know, this is how people should build their machine learning pipeline kind of end to end, quality.
Andy Readman 11:47
So, I always makes me the most excited and the one most passionate about with anything with within the kind of data and AI spaces real world examples and we talked about few already. What, what gets you the most excited.
Adi Polak 12:03
Well, what gets me the most excited. Maybe you'll show her a little bit about my journey with machine learning and data so I actually I started as a machine learning researcher I have a master's degree in machine learning and cybersecurity I worked in the lab. I worked in, it was a collaboration between Deutsche Telekom, and the University I studied at. So I worked there for three years, and during my last year, IBM just opened a new security lab, and I was going to board into, into a new project because they liked my thesis. And then I got exposed to big data, because they were having big data, and they wanted me to implement my thesis algorithm on their data which required, working with environments such as a dupe. Back then it was my MapReduce. And I think that really showed me that the difference between doing our own there very kept research where I have full control of the data and how the data flows into my machine learning algorithm, because back then I was designing the whole end to end experiment from collecting the data from real, real sources, all the way to the machine learning model and testing and evaluating it, versus working in an environment where the data is out of, out of my control specifically. So it was a lot of back and forward with the DevOps team and a lot of back and forth with the. We didn't call it data engineers back then but now we know it's data engineers, the backend engineers on what we can do. So I think that kind of brought me into the space of understanding that there is a bigger scope to it. And this is the area that I'm really passionate because I've seen there, I was there, I struggled with the tools for enough time. I found success I developed my thesis and my thesis algorithm running in Java, because back then. I thought in all the other tools weren't supported enough, there wasn't enough good support for the big data scale. And I think this is going to be the new tools that are now emerging for researchers that are actively doing the research inside the corporation, the need to deal with this scale of data is going to really help unblock and also greater impact in greater scale in their organization, because they're going to drive better innovation, they're going to find that pattern. Patterns and their data. They're going to close the loop as well, because they will be able to have all the, all the stakeholders on board and all the tools on board, but I think this is specifically to an area that I'm very passionate about,
Andy Readman 15:23
I really like that it really resonates well I think. Yeah, the fact that you've experienced the pain points that you have now are helping people address, and you've kind of been in their position I think it makes it makes it a lot easier to kind of convey that with a greater understanding and maybe more of a kind of see things from the customer's perspective, which is always really key. Okay, so let's talk a little bit more about Apache Spark conscious that we've got varying levels of kind of technical ability watching and following along today but what does Apache Spark ML library bring that wasn't there before or what what are the differences with using something like that.
Adi Polak 16:06
That's a great question, maybe I'll start with what is Apache Spark. The best place to start. So it's a general engine for running big data analytics. So it started in Stanford University by mateesah Korea. It was his race research his PhD did a PhD on distributed systems. Any develop that tool that enables you to do faster MapReduce processing that, that is, has more capabilities and abstracts, a lot of the complexities of distributed systems. So it's great tool for analytics is the generic tool, so we can use it for machine learning. We can also use it for basic analytics if you want to work with or just build a what we call data pipeline or an ETL to transform the data from one state to the other. So it gives us a lot of capabilities. Specifically, the machine learning part this is the library that it's still growing. So it has a lot of kind of pros and cons and need to be conscious about what you're using. But they're implementing the distributed machine learning, because the whole engine itself is distributed. So they were backing in and using a lot of because it's based on statistics, a lot of statistic into building, giving people API's, out of the box, run their machine learning at scale. So I think it's, it's interesting to see that tool, it's really interesting to see how it's going to grow with the years and how it's going to connect better to other tools as well. For example, we've seen TensorFlow. Right, so we have TensorFlow on Spark project tool for people they wanted to tensor, Spark, we've seen tensor frames, which is an API directly on Spark, that the community created. And we're seeing it grows more and more into kind of cooperation between these two spaces. And the reason for it is because there is a need for deep learning technology as well some of the algorithms inside Apache Spark ML has there, they are based on deep learning, but it's not, they're not covering everything yet, so this collaboration and this ecosystem that builds up gives more tools to developer to data scientists and developers, and also make it a more cohesive conversation between the data engineers that most of them today already work with tools such as Apache Spark, and the data scientists. So I think it's a, it's growing. It's going to be very interesting tool for data science to use.
Andy Readman 19:13
So, but for anyone that doesn't already appreciate it. Spark is completely open source as well so it's not Microsoft's intellectual property that they're writing and and selling as their own, but they are heavily involved in something that's completely open source community community built and yeah that's what he's talking about is that is that collaboration with the open source community which is it's fantastic to to see, from my perspective. Got a question, so if you're if you're following on live, pop some questions in the chat and in the comments on YouTube. Annabel has asked, What do you think is the most promising thing about machine learning is, and its influence is already significant but do you think it's likely to become more significant.
Adi Polak 20:00
That's a great question. Um, so a couple of things. What the blockers that we've seen with machine learning. I think this is kind of the topic that we should address first. So, what happened in the industry, at least from my opinion. And, you know Andy I know I know you've seen a lot too, so. So in here as well. A lot of companies seem to hype around machine learning and what it can do for the business and how it can help drive innovation and drive business growth. Right. So a lot of the executives said hey I need a new data scientist so they start hiring data scientists, but they didn't provide them with the tools, the data science would come to the organization sit in their chair in nice office, you know, or working from home, and they would ask where's my data. Where are my tools, how can I work, am I going to run it on my laptop, they need to go somewhere and find data, how do I get started, And this is when we've seen it a little bit. Now, going towards transformation in the essence of how we can prioritize this. How can companies prioritize this as organizations in order to drive that innovation so we're not drifting away from shooting machine learning is still the focus in machine learning also helps with being data driven because you can extract patterns, etc. But we're understanding companies around spending that they need better tools they need bigger teams, they need someone that executive sponsor that process inside the organization, and be finding the team that's going to build the pipeline. So you can think about my quarter flows and the pipes, in order for the machine learning experts fish nurse data scientist to do their job when building model and also like we talked about a little bit earlier, is closing the loop, so they will know what's happening in production with their machine learning, so if they need to deploy a new machine learning, they would have this process already, to me, if so, the way I see it, some of the tools already exist. Some of the tools or requirements. And I think kind of an opportunity for a lot of people who wants to build those tools. But I think it's going to grow, and it's going to be more heavily used. And NT, maybe you can share maybe even more carefully regulated. What do you think,
Andy Readman 22:37
yeah, definitely. I think that's a really good point. So in terms of what's promising about the near future, I think, is allowing businesses to catch up with the with the technology so that there's so much advancement that's gone on in the last maybe five 510 years, and businesses now need to maybe take a step back and sort out their data state and sort out the culture of their business before they can actually just deploy HD Insight and then start using Apache Spark so I think yeah it's, there's a, there's an element there, making sure the data is in a reliable enough position to start feeding it into something that that's making recommendations for you or something like that, that's this is a huge, hugely important thing to tackle that we, we do a lot of work on on data strategy here at y hive and that's often the first hurdle, whether you're doing, whether you're just trying to get to some business intelligence improvements or you're looking to go that next step into machine learning I think sometimes tackling it in that in that kind of logical order is a great stepping stone so sorting out the data and how the company manages data and that's not just cleansing your your SQL database that's, that's really understanding what third parties you work with and do you do govern what access you get to the raw data if you're working with SAS tools and online services, things like that, all the way through to Okay, well where do we store this data so make sure that we're handling it in the best way possible to allow future experimentation in ways that we haven't discovered yet to really come in and adapt to that data nicely. And then once it's there, it's doing the business intelligence it's using human intelligence to kind of get insights from that data, and then start to post questions of what if we just knew about this or were understanding the value really so I think where the where there's friction where there's challenges for businesses. It's a huge opportunity to apply some tried and tested and very mature machine learning systems out there, applying those to friction and challenges is ultimately where businesses are going to get the most value from a business perspective, I think it's, it's understanding that you can, you can have your horizons kind of wide open all these amazing tools, but it's understanding what you need to do and taking the steps to get yourself there is. For me, one of the most exciting things I guess some for some people that's, I just want to get to the end of this, maybe the boring part is sorting out the data but I think that's like the foundations to the house. Nobody really shares pictures of their foundations they share the pictures of the house being built or the extension that's been done right, but, but the foundations are so important and yeah the doing it well, really opens up huge opportunities. We don't know what, what services are going to be out there in two years time, but if we do the data in the, in the best way possible we design that architecture for the data. So we've got a copy of all the raw data we've ever had. Then if something comes along, new and we have to rebuild all of them the machine learning or the business intelligence practices, we can go back and we've still got some data there that we can start again, things like that lots of, lots of principles that come into it but yeah so certainly what gets me excited and I think that the rate of change, like you said it yourself where you can't be that can't have been long ago, you're working and didn't have the tools that you needed and didn't have data engineers and things are changing so quickly. I think keeping up with that, or at least doing things the right way to allow us to adapt with the kind of the pace of change.
Adi Polak 26:05
Yeah, absolutely. It's uh, yeah, I really like your analogy of building a house when you have the foundation and, you know, against everyone takes pictures of the house and are proud of how the house looks like but, you know, have to have the bank foundations, otherwise it's not going to work right. So I really love that I'm curious, when you specify data strategy. Are you looking into the organization structure or specifically the technologies,
Andy Readman 26:39
yeah absolutely so it's way more than technology so we talk around people, process and technology, like, like most kind of strategic kind of consultants would come in and talk about those that's that's fairly obvious, but I think it goes all the way through to culture and core, core strategy of the business and in aligning the business strategy with a technology strategy or a data strategy that helps achieve the goals. So, as there's even a bring up a little background so we talk around jellyfish. I'm totally derailing this but they. Yeah. So hopefully this is within the context. We took around the be more jellyfish. This concept of having a business, business data that's so reliable they're so well prepared and trustworthy and validated that you can almost act like jellyfish, as employees of that business so it's democratized hand jellyfish if there's any jellyfish watching or listening along jellyfish don't have brains and that's a fact. So we're not trying to offend anyone that's, they just they just instinctively react to what's what's going on around them so their bodies change, without any kind of brain getting in the way and trying to make decisions it's all instinctive, and I think we talk around. Can you get your data state as a as a business into a place where you can be more jellyfish you can you can trust that data instinctively and react to it. And then there's another concept because I love the kind of parallels with the animal kingdom is be more dolphin so again totally hijacking this. But, being more dolphin is a concept where dolphins work in pods, and they have this incredible hunting mechanics and they have the echolocation sensory information if we consider that to be the for this analogy the data so the data is the signals from the echolocation. All of the dolphins have that echolocation system, it's not there's one emperor a dolphin. And he or she has the has the tools and then tells people where to go and where to go and hunt. And I think yet they still work in pods. So I think really key thing there is, understanding democratizing the data, making those insights, whether they're just bi or maybe they're machine learning insights, making that available to the entire business as a culture, getting your team working like intelligent dolphins that all have the tools and the equipment to do their job the best they can, is really where we see the best results and yeah absolutely a culture thing as well as I thought it was a kind of a technology thing.
Adi Polak 29:09
Cool. Yeah, it to know that a lot of times I know they have some of your abilities and it's kind of how they map the ocean. Yeah, interesting.
Andy Readman 29:22
I've got a few data books behind me but at least two of them are books on the animal kingdom facts and features about I've tried to find even more kind of parallels and self insights into interest data use within the animal kingdom. But yeah, if you can think of any more, let me know after the session. I'll do. Speaking of books, you're, you're working on a book I hear the next couple of months, potentially, to what what's the book in, and what's it about.
Adi Polak 29:52
Yes, so it's going to be about machine learning with Apache Spark ecosystem. So we touched a little bit. it's about how you can design and build your this stone with different tools that connect with Apache sparks and working on big data and having distributed machine learning running on top of put up on top of large scale data. And it's really interesting because TensorFlow and also pipe torch and all those loves, you know, very love libraries and open source tools, distributed strategy. So how they can run their machine learning, training in a distributed manner or how they can serve it in a way to support large scale serving a lot of requests and need to answer, or to provide insights into a lot of requests. So, yeah, it will take a couple of months to be in, in public preview it's usually you get the kind of a tough variety you get the baby. Third, you get the animal. But you know once I have the animal, I'm going to go back to you, Andy and ask you what is that we definitely
Andy Readman 31:08
can't wait to see it, I think that's fascinating. And, yeah, me personally, I have not had a massive, massive kind of hands on with, with, like Spark, in particular, yet so it certainly inspired me to, to take a closer look especially with the machine learning, machine learning libraries built into it as well and the TensorFlow integration and API that's pretty interesting to kind of do all that in, in, kind of one package or one one framework so that's cool. Okay. Yeah. Any more questions to find then we've got a little bit more time left. So, I guess. What do you think, around, without getting too deep into it. What can anyone talking about machine learning and artificial intelligence can always kind of lean on the kind of the fear side of things. For me I think there's lots of scare mongering in the press, but what can typically go wrong with machine learning, done in this context of business, use of this kind of artificial intelligence.
Adi Polak 32:12
I think money goes to waste. If you don't put the infrastructure first correctly you might be hiring, you know, people with many years of experience, and build a data science team but you're not giving them the right tools, then we will see, you know, thinking, wearing the hat of an executive, the ROI is kind of those very low. So, you would either close the project or say machine learning is not for me, it failed me without giving you a good, you know, a good start. So, this is what I think is might be an issue for some companies it's like they say, Yeah, we tried it, it didn't work, we hired the best data scientists that money can pie right. So I really think that's changing the mindset and the culture. First, in making sure to stakeholders are there, the tools are there going to make it a success, I you know I can make it or break it. So if you don't have it, it will be very hard for the better scientists to to deliver and bring it back to the organization.
Andy Readman 33:20
No, I agree I think the fact you didn't even talk about terminators and how it's ridiculous to worry about these things kind of really sets the point now I think so. Anyone, anyone that hasn't had the experience or hasn't come to come on to one of my talks, I have to terminate a slide and just to say that we're not here to talk about this kind of sentience intelligence that some people sort of associate with artificial intelligence and certainly the press, and lots, lots of coverage out there makes it scary in terms of the what's possible on that scale I think when we're working with maybe narrowly focused AI and machine learning, just to just to essentially run, not to demystify it too much and simple oversimplify it but essentially we're doing calculations and advanced very very advanced calculus, mathematics, often with with the algorithms. But that's pretty much it so it's it's a safe to a business as using a calculator. Very, very scientific, of course, it's not going to take over your business at all run it into the ground and do things you don't want it to do, it's just going to give you much better insights with a massive amount of data that is not feasible for humans to kind of make those calculations that quickly or that accurately. But yeah. Often, the worst, the worst to happen from a project like this is money money has been wasted because it was maybe set up in an incorrect way it was discovered badly or or not done in a way that maybe understands that there's ultimate value that in investing into to get that value back. I think focusing on focusing on the ROI and customer again customer challenges and kind of the frictions there of the businesses we work with really ensures that we don't start doing very expensive machine learning project on something that that isn't actually adding any value to their business This essentially just, just a bit of a fun project to do I think that's, that's really key to, to understand. Okay so, thinking about the, maybe the future of machine learning, I think we talked around that, there's been some massive advancements really quick growth and adoption has to kind of catch up a little bit but do you think the technology, kind of, has plateaued or is looking to plateau are we still going to keep growing at the same rate, Despite kind of, despite how amazing it's got to at this stage it is that slowing down, do you think or where do you see the future of machine learning.
Adi Polak 35:50
Good question. Um, I'm thinking we'll see, for supporting tools for it, Because people are by now understand many companies and people are understanding that they need to support the process not only develop the machine learning model and implement the algorithm but also support the overall process so monitoring observability versioning, all of that. So that, that's, that's the first thing. Second thing, we'll see it grows. Once we have the tools we'll see grows more into the space of distributed machine learning and extracting patterns out of big data. And, which was a challenge because there wasn't a lot of tools so that was sort of somewhat restricted the data wasn't available, etc. And I think we will see a lot of SAS or machine learning serverless solutions coming up to support big data as well. Just because the infrastructure, you know, might be a little bit more complex, or DevOps or you know, requires some expertise that can be kind of replicated in a managed service. So I think we might see more of that as well. In maybe the start-up ecosystem or maybe even in the cloud. And I think there's a lot of opportunities, because there are a lot of challenges and we always kind of say when there's challenge, there's an opportunity for everyone who wants to kind of learning about that topic and maybe build that saw solution. Yeah,
Andy Readman 37:38
that's good to know. I think the more kind of sass, and the more work that that's being put into it and kind of managed for you, I think great example of that is the code of services. So machine learning algorithms that are pre packaged pre built and maintained by Microsoft and available as API's to simply add into any kind of existing or new, new software that's out there. For me that they're some of the most exciting things to talk about with customers because really opens the eyes of kind of amazing life changing in some cases experiences that you can, you can add, and you don't even need to worry about the kind of the DevOps side or the machine learning, maintenance yourself but yeah if we if we can simplify the not even take it all the way to the cognitive services, kind of serverless example but is there some middle ground there I think would be really nice if if there can be some improvements to do that.
Adi Polak 38:32
Yeah, I think it divides into when we're looking at the machine learning world and it's great that you mentioned, we can kind of divided into two pre build models that we can utilize and start working with the big companies are producing, and then tailor made models, we want to create specifically for our business for very specific, very unique business case and the to interact so we can use them together, right so we can use cognitive services to extract some information out of unstructured data, for example, and then use that data to fit it into our machine learning algorithm to extract the specific machine learning model, like a very tailor made machine learning model that we need. So I think it's a very good point, you know, Let's, and we'll see the both grows, both together, especially in the space where you can find kind of a generic requirements for machine learning like identifying face or kind of transferring sound into audio into text, etc.
So it's interesting.
Andy Readman 39:43
Yeah, there's more and more It's frightening how, how quickly they're even the cognitive services are being updated with new, new features and other tools in that area, things like the custom neural voice, have you seen that you experienced that at all. Yes, I did. That's really awesome. I think it's awesome. So if anyone's interested, this is a speech to text or text, text to speech tool that is completely different to anything you've ever seen before that it in the same way that kind of, I'm going to say Alexa off in everyone's living in the same way that Alexa has that kind of quite unnatural tone consistent, really, really decent voice font across everything that Alexa is saying on Amazon devices. Thank you. Any brand out there now has the opportunity to build their own personalized voice, that can be the kind of the face of their brand for any kind of text, text into speech, and the results are staggeringly good, I think it's, it's into it's out of preview now and it's available in kind of the public. It's out of the private preview list, and, yeah, there's some demos on the Microsoft as your site I think Joe the Australian one is my favourite. So, have listened to that after, after the session. It is, yeah, they've got a clip of the original person that narrated the, the text and then the output of the custom neural voice, which is, yes so good, I think you very difficult to tell which one, which one is the human talking so just something like that that's, it's already been solved. You don't need to, to delve in and start writing your own machine learning to do something like that and I think it's a really nice point of enriching what you're trying to do and not reinventing the wheel or something that's already been done if you need to use text to speech as part of something that you're building from scratch. Don't write your own text to speech engine, use the custom neural voice. If you need to do some face detection, lean on lean on the Face API, something like that. And then, anything bespoke is where, where things get really exciting with with Spark and data bricks and then various services out there that you can you can play with. Yeah, absolutely. Looking forward, cool. So yeah what's maybe going back to the real world examples cuz that's my favorite thing, what's, what's maybe the most exciting thing you've, you've been able to work with, or things that you can talk about obviously I'm conscious when we have mics, I guess there's always this is a really exciting project, you can't talk about expect. But, I think, yeah, I've got one that I think some great examples if you want to talk more generally obviously around the maybe the AI for Earth but if you have one other than that or feel free to pick pick some of that stuff out and talk us through some more real world examples would be great.
Let's go, let's go with the AI for Earth. What did you see there to kind of capture your attention.
So I just before lockdown we run ran a data and AI workshop and envisioning session at the White House offices, and we've run various ones that kind of Paddington machacas headquarters as well. And yeah, we had someone that came along that was we're already working with trying to track hotter dolphins in the in the North Sea, of the, of the shores of the UK, and they saw on my slides on Earth and how people out there are trying to do individual animal recognition from from sort of fins that reaching the surface and things like that, and yet the fact that someone else has already kind of done the machine learning to help with that and there's already a project that of course is open source of course a very collaborative, charitable project, they were able to get in touch and kind of obviously work work with them on that so bringing, bringing people together for similar causes. Obviously, amazing, and our purpose is really something that I've seen, if this is quite special, I think understanding obviously that tracking up on the dolphins that the important thing to understand there is that you can infer the kind of the ecosystem and the marine marine habitat for the rest of the agriculture in that area and obviously the rest of the marine life, expanded kind of on a global scale you can infer changes and what's happening there, which is a really important topic and AI for good is something that makes me really happy something I'm really passionate about, obviously.
Yeah, same here and one of the projects that I've seen was with the white pepper, but you can identify them, and it's very hard to find whichever just because of their colors. Their color to nature, but you still want to make sure they're not an extended time or treat them in a way that people will protect them. So I think it was one of the exciting project that I've seen. And I think this is going to grow in that space, especially now with Microsoft is invested into heavily invested in green. So reducing co2 developing link greener technology, we thinking about how we do processes and maybe run our VMs and machines are going to really help in that space. So I think machine learning is going to play a big role there as well. And I think there's going to be a lot of interesting, more tools and more kind of opportunities for machine learning to do good for the world. So, yeah, it's all start with effecting patterns I mean if you can detect patterns and include them and have them better, and this the same kind of business objectives. It's all good.
Yes, this one's a great example because that was, that was even on primetime TV advertising here in the UK which is anything that's what's about AI machine learning that's extremely rare for it to just be showing off as we are Microsoft and here's some cool things that you can do with artificial intelligence and this is how we're changing the world and making the world a better place. I think that's actually fascinating to see that advert and yet, obviously that there's no left one is a great example, because, yeah, sifting through images and taking every five seconds on the webcam, as humans, a lot of resource you would need to do that and it's not the most effective way obviously of seeing them, but yeah it makes, makes things that pretty previously just weren't feasible to do possible. And that's what really exciting for me they're the best, best case study or the best example for me currently is premonition, where if you if you Google premonition maybe free three or five years ago. There's a picture of a human with a, with a net trying to catch mosquitoes so they can do the analysis, and things have changed a little bit. So, in the last year, you've now got collaboration with with some advanced engineering teams to build these, they look like smart speakers or sono speakers kind of thing but they, they say the mosquito or pathogen tracking and testing devices that can capture things somewhere between 10 to 15,000, mosquitoes per night, which then the data is then analyzed and then sent out to the cloud. The mosquitoes are unharmed or that they're safe cell free to go and roam around and, and other people on the holidays or whatever they normally do mosquitoes and the data in the cloud and all this was technology we're talking about more of a technical level how quickly it's evolving and the things you can do real world what it means is that these 10s or 15,000 Mosquito samples per device around the world and be sampled and analyzed against something like 3 trillion different genomic sequences to identify animal DNA that's in the mosquitoes blood that it's been extracting from its from its environment, which means, yeah we can we can understand that that mosquito took blood from a cow that cow had particular disease that is spreading in that kind of species, but essentially we can stop outbreaks and pandemics from happening. Once these types of services are rolled out globally, which I think is obviously very fitting at the moment and yeah have this Coronavirus pandemic happens maybe a few years in the future, something like that might have been able to kind of might be able to stop it or give us early warnings to stop things getting so bad across the world. So,
yeah, it's really interesting look into that project, though. Now it's now curious to see what they're building you know how it's evolved. So thank you, thank you for mentioning it.
I don't see very very relevant at the moment.
Yes, very much so I think when the pandemic started one of the, one of the things data scientists to work on in the community everyone from every companies, they collaborated created a place where research articles are shared in the eyes share the kind of work together to find solutions. I think this is a good power that we have as a community when we come together to do good. First one,
completely agree. What a fantastic kind of point to end on, again, my personal favorite part is is the real world examples and how the technology can completely change the world around us for for good. So, yeah, fantastic, fantastic topic obviously very vague being machine learning, but I think got some great real world examples in there a bit of tech bit a bit of something for everyone really. Abby thanks for joining us, you've been a fantastic guest, and then I know that we'll be checking out the book in a few months once that was ready and yeah stay in touch more.