>> Commentator: Live from Orlando, Florida It's theCUBE. Covering Microsoft Ignite. Brought to you by Cohesity. >> Hello everyone and welcome back to theCUBE's live coverage of Microsoft Ignite 2019. We are theCUBE, we are here at the Cohesity booth in the middle of the show floor at the Orange County Convention Center. 26,000 people from around the globe here. It's a very exciting show. I'm your host, Rebecca Knight, along with my co-host, Stu Miniman. We are joined by Francesca Lazzeri. She is a Ph.D Machine Learning Scientist and Cloud Advocate at Microsoft. Thank you so much for coming on the show. >> Thank you for having me. I'm very excited to be here. >> Rebecca: Direct from Cambridge, so we're an all Boston table here. >> Exactly. >> I love it. I love it. >> We are in the most technology cluster, I think, in the world probably. >> So two words we're hearing a lot of here at the show, machine learning, deep learning, can you describe, define them for us here, and tell us the difference between machine learning and deep learning. >> Yeah, this is a great question and I have to say a lot of my customers ask me this question very, very often. Because I think right now there are many different terms such as deep learning as you said, machine learning, AI, that have been used more or less in the same way, but they are not really the same thing. So machine learning is portfolio, I would say, of algorithms, and when you say algorithms I mean really statistical models, that you can use to run some data analysis. So you can use these algorithms on your data, and these are going to produce what we call an output. Output are the results. So deep learning is just a type of machine learning, that has a different structure. We call it deep learning because there are many different layers, in a neural network, which is again a type of machine learning algorithm. And it's very interesting because it doesn't look at the linear relation within the different variables, but it looks at different ways to train itself, and learn something. So you have to think just about deep learning as a type of machine learning and then we have AI. AI is just on top of everything, AI is a way of building application on top of machine learning models and they run on top of machine learning algorithms. So it's a way, AI, of consuming intelligent models. >> Yeah, so Francesca, I know we're going to be talking to Jeffrey Stover tomorrow about a topic, responsible AI. Can you talk a little bit about how Microsoft is making sure that unintentional biases or challenges with data, leave the machine learning to do things, or have biases that we wouldn't want to otherwise. >> Yes, I think that Microsoft is actually investing a lot in responsible AI. Because I have to say, as a data scientist, as a machine learning scientist, I think that it's very important to understand what the model is doing and why it's give me analysis of a specific result. So, in my team, we have a tool kit, which is called, interpretability toolkit, and it's really a way to unpack machine learning models, so it's a way of opening machine learning models and understand what are the different relations between the different viables, the different data points, so it's an easy way through different type of this relation, that you can understand why your model is giving you specific results. So that you get that visibility, as a data scientist, but also as a final consumer, final users of these AI application. And I think that visibility is the most important thing to prevent unbias, sorry, bias application, and to make sure that our results are fair, for everybody. So there are some technical tools that we can use for sure. I can tell you, as a data scientist, that bias and unfairness starts with the data. You have to make sure that the data is representative enough of the population that you are targeting with your AI applications. But this sometimes is not possible. That's why it's important to create some services, some toolkits, that are going to allow you, again, as a data scientist, as a user, to understand what the AI application, or the machine learning model is doing. >> So what's the solution? If the problem, if the root of the problem is the data in the first place, how do we fix this? Because this is such an important issue in technology today. >> Yes, and so there are a few ways that you can use... So first of all I want to say that it's not a issue that you can really fix. I would say that, again, as a data scientist, there are a few things that you can do, in order to check that your AI application is doing a good job, in terms of fairness, again. And so these few steps are, as you said, the data. So most of the time, people, or customers, they just use their own data. Something that is very helpful is also looking at external type of data, and also make sure that, again, as I said, the pure data is representative enough of the entire population. So for example, if you are collecting data from a specific category of people, of a specific age, from a specific geography, you have to make sure that you understand that their results are not general results, are results that the machine learning algorithm learn from that target population. And so it's important again, to look at different type of data, different type of data sets, and use, if you can, also external data. And then, of course, this is just the first step. There's a second step, that you can always make sure that you check your model with a business expert, with data expert. So sometimes we have data scientists that work in siloes, they do not really communicate what they're doing. And I think that this is something that you need to change within your company, within your organization, you have to, always to make sure, that data scientists, machine learning scientists are working closely with data experts, business experts, and everybody's talking. Again, to make sure that we understand what we are doing. >> Okay, there were so many things announced at the show this week. In your space, what are some of the highlights of the things that people should be taking away from Microsoft Ignite. >> So I think that as your machine learning platform has been announcing a lot of updates, I love the product because I think it's a very dynamic product. There is, what we now call, the designer, which is a new version of the old Azure Machine Learning Studio. It's a drag and drop tool so it's a tool that is great for people who do not want to, code to match, or who are just getting started with machine learning. And you can really create end-to-end machine learning pipelines with these tools, in just a matter of a few minutes. The nice thing is that you can also deploy your machine learning models and this is going to create an API for you, and this API can be used by you, or by other developers in your company, to just call the model that you deployed. As I mentioned before, this is really the part where AI is arriving, and it's the part where you create application on top of your models. So this is a great announcement and we also created a algorithm cheat sheet, that is a really nice map that you can use to understand, based on your question, based on your data, what's the best machine learning algorithm, what's the best designer module that you can use to be build your end-to-end machine learning solution. So this, I would say, is my highlight. And then of course, in terms of Azure Machine Learning, there are other updates. We have the Azure Machine Learning python SDK, which is more for pro data scientists, who wants to create customized models, so models that they have to build from scratch. And for them it's very easy, because it's a python-based environment, where they can just build their models, train it, test it, deploy it. So when I say it's a very dynamic and flexible tool because it's really a tool on the pla- on the Cloud, that is targeting more business people, data analysts, but also pro data scientists and AI developers, so this is great to see and I'm very, very excited for that. >> So in addition to your work as a Cloud advocate at Microsoft, you are also a mentor to research and post-doc students at the Massachusetts Institute of Technology, MIT, so tell us a little more about that work in terms of what kind of mentorship do you provide and what your impressions are of this young generation, a young generation of scientists that's now coming up. >> Yes. So that's another wonderful question because one of the main goal of my team is actually working with a academic type of audience, and we started this about a year ago. So we are, again, a team of Cloud advocates, developers, data scientists, and we do not want to work only with big enterprises, but we want to work with academic type of institutions. So when I say academics, of course I mean, some of the best universities, like I've been working a lot with MIT in Cambridge, Massachusetts Institute of Technology, Harvard, and also now I've been working with the Columbia University, in New York. And with all of them, I work with both the PhD and post-doc students, and most of the time, what I try to help them with is changing their mindset. Because these are all brilliant students, that need just to understand how they can translate what they have learned doing their years of study, and also their technical skillset, in to the real world. And when I say the real world, I mean more like, building applications. So there is this sort of skill transfer that needs to be done and again, working with these brilliant people, I have to say, something that is easy to do, because sometimes they just need to work on a specific project that I create for them, so I give data to them and then we work together in a sort of lab environment, and we build end-to-end solutions. But from a knowledge perspective, from a, I would say, technical perspective, these are all excellent students, so it's really, I find myself in a position in which I'm mentoring them, I prepare them for their industry, because most of them, they want to become data scientist, machine learning scientist, but I have to say that I also learn a lot from them, because at the end of the day, when we build these solutions, it's really a way to build something, a project, an app together, and then we also see, the beauty of this is also that we also see how other people are using that to build something even better. So it's an amazing experience, and I feel very lucky that I'm in Cambridge, where, as you know, we have the best schools. >> Francesca, you've dug in some really interesting things, I'd love to get just a little bit, if you can share, about how machine learning is helping drive competitiveness and innovation in companies today, and any tips you have for companies, and how they can get involved even more. >> Yeah, absolutely. So I think that everything really start with the business problem because I think that, as we started this conversation, we were mentioning words such as deep learning, machine learning, AI, so it's, a lot of companies, they just want to do this because they think that they're missing something. So my first suggestion for them is really trying to understand what's the business question that they have, if there is a business problem that they can solve, if there is an operation that they can improve, so these are all interesting questions that they can ask themselves their themes. And then as soon as they have this question in mind, the second step is understand that, if they have the data, the right data, that are needed to support this process, that is going to help them with the business question. So after that, you understand that the data, I mean, if you understand, if you have the right data, they are the steppings, of course you have to understand if you have also external data, and if you have enough data, as we were saying, because this is very, very important as a first step, in your machine learning journey. And you know, it's important also, to be able to translate the business question in to a machine learning question. Like, for example, in the supervised learning, which is an area of machine learning, we have what is called the regression. Regression is a great type of model, that is great for, to answer questions such as, how many, how much? So if you are a retailer and you wanted to predict how much, how many sales of a specific product you're going to have in the next two weeks, so for example, the regression model, is going to be a good first find, first step for you to start your machine learning journey. So the translation of the business problem into a machine learning question, so it's a consequence in to a machine learning algorithm, is also very important. And then finally, I would say that you always have to make sure that you are able to deploy this machine learning model so that your environment is ready for the deployment and what we call the operizational part. Because this is really the moment in which we are going to allow the other people, meaning internal stake holders, other things in your company, to consume the machine learning model. That's the moment really in which you are going to add business value to your machine learning solution. So yeah, my suggestion for companies who want to start this journey is really to make sure that they have cleared these steps, because I think that if they have cleared these steps, then their team, their developers, their data scientists, are going to work together to build these end-to-end solutions. >> Francesca Lenzetti, thank you so much for coming on theCUBE, it was a pleasure having you. >> Thank you. Thank you. >> I'm Rebecca Knight, Stu Miniman. Stay tuned for more of theCUBE's live coverage of Microsoft Ignite. (upbeat music)