# AI Show – Deep Learning vs. Machine Learning

>>You’re not going to want to

miss this episode of the AI Show coming to you because it’s

special from Microsoft Ignite. Francesca Lazzeri tells us all about the difference

between machine learning, deep learning, and what it means

for AI. We’ll see you then. [MUSIC]>>Hello and welcome to

this special edition of the AI show where we’re live. Well, we’re not live, because

you’re watching a recording, but we’re live right now at Ignite. I have my esteemed dear

colleague, Francesca Lazzeri. We actually work together, but she’s much classier than I am. How are you doing, Francesca?>>I’m good, I’m great, and thank you so much for

having me on the AI show.>>Fantastic. Well, look, one thing we’re going to talk

about because we like to keep these very crisp

is machine learning, deep learning, what’s the difference, and we’re hoping you’re

going to help us out.>>Yeah. So this is a

great question because actually all customers I have been working with ask me

the same question. We want to do something

with our data, we want to use our data to

answer our business question, but how can we use machine learning, and what’s the difference

between the machine learning, deep learning, what is AI? So I wrote an article

about these so I can give them an answer every

time they ask me this question. This article really compare

the three different area. So let’s start with the central one, with the core one, which

is machine learning. So I think about the

machine learning as a portfolio of algorithms. You have to think about algorithms

as a statistical functions, a mathematical functions

that use your data to learn something and

give you a result. Now, what is AI? AI is about consuming

the machine learning. So you will build the application on top of

the machine learning model, and you consume these

machine learning models, these AI application over time. So it’s a really intelligent way

of consuming machine learning, that’s the tarot reading your data. Then deep learning, deep learning

is just a type machine learning. It’s a very interesting

type of machine learning because it’s based on what

we call the neural networks. So there are many different networks and since we have different networks, different layers also as well, they are called the deep learning. Thereby, it is just a type. It’s just a topology of

the machine learning. So you don’t have to think about machine learning and deep learning as something that they

are completely different. They are in the same area and AI, of course, is using both of them.>>I see. So machine learning

then is a discipline whereby we extract information

or algorithm from data, I think it’s what you said. Deep learning is a subset of that, that has a special

kinds of algorithm.>>Yes.>>So at what point can I be

like I am using deep learning? How do you know when you’re there?>>So you are there when, of course, it depends a lot on the type

of the data that you have, the type the questions

that you want to answer. If you have, for example, a huge amounts of data sourced

from a hardware perspective, you need, for example, to use GPUs instead of CPUs. I would say, these are all

indicators that tells you that you are entering

the deep learning area.>>That’s fantastic.>>Of course, in terms

of also different APIs that you can use in order to

facilitate your deep learning work. There are so many difference

and you can use all of them on Azure, like TensorFlow, Keras. So it’s really up to you, and on the type of data, and the type of question

that you’re using. Another interesting part is that there are multiple

types of neural networks. So this means that many different print type of deep

learning algorithms, there are convolutional

neural networks.>>Which I like. I’m a

vision guy so I like those.>>Recurrent neural network.>>Which I’m trying to understand, so you’re going to help us

out maybe at one point.>>Of course.>>Yeah.>>Then under recurrent

neural networks, for example, we have the LSTM and GRU.>>GRUs.>>Exactly. They are very good

with the sequential type of data. So for time series

scenarios wherein you want to build the end-to-end time

series forecasting solution, LSTM can be very, very hard work. But again, it depends a lot on the

problem that we want to solve, on the size of your data, on the type of data that

you want, and then, of course, how you’re going to build these end-to-end architecture.>>Awesome. So it

turns out that all of these stuff, she broke down. She didn’t write it

down and just in a day, by screw it down in our docs. So why won’t you tell

us about the doc that you own and how

people can go and read it.>>Yes. So this is a very nice

doc because you can consume it in a very nice and easy way in the sense that we didn’t put too

much information about it. We just selected the most

important information. So the doc starts with the mean difference between

artificial intelligence, machine learning, and deep learning. It explain everything here. Then if you keep on, we created this very nice

stable that’s compares machine learning versus deep learning in terms of number of data points, this is exactly the question

that you were asking, hardware dependencies, the

featurization process. Because, for example,

with machine learning, you need to do what we call

the feature engineering. You need to get your dataset, bigger, richer by doing

the feature engineering. With deep learning,

you don’t need to. So there are some differences also

for these: learning approach, execution time, and then

of course, the output. So this is a nice simple

table that compares the two. Then, if you keep scrolling, you can see that there are different type of explanation

for deep learning. What you can do with deep

learning, for example, there is the named-entity

recognition use case, object detection use case,

image caption generation, machine translation, text analytics, and then there is a short explanation of artificial neural networks, and we already

mentioned some of them. Feedforward neural network,

recurrent neural network, and a convolutional neural network.>>That’s awesome.>>So again, you can see that using a very nice article that

people can enjoy reading. Then at the end, you have also lot of different

resources on how you can.>>How you can do it.>>Exactly, how you can do

deep learning on Azure.>>Awesome. So what we’ll do is, we’ll make sure that this

particular article is in the show notes so you can actually go and read it and

get a sense for it. I actually love it. I’m computer vision person, you obviously know all things: time series, forecasting, and sequence. So I’d love to learn more about that. Thank you so much for

spending some time with us. Thank you so much for watching and learning all about

machine learning, and deep learning, and AI, what it means, and a special

doc with the link below. Again, thanks for watching

this special episode of the AI Show coming to from

Ignite. We’ll see you next time. [MUSIC]

Artificial intelligence(AI): every algorithm and automated system developed for problem solvingMachine learning(ML) : subset of A.I algorithms that are divided into two stages,learningandexecutionOn

learningthe algorithm generates a data model that tries to abstract a characteristic of the dataOn

executionthe data model is used to reproduce the learnt characteristic on new data samplesDeep learning: subset of M.L. algorithms that usesconnectionistapproaches, which are heavily based on linear algebra transformations and gradient descent, they often minimize by small steps a function that evaluates how well the "data model" fits the target characteristic, for each data sample, it's calledDeepbecause they often perform more than one transformation on the input data, the more transformations, the more "Deep" the model is