Introducing Machine Learning Workloads

0:00[VIDEO LOGO]

0:06Welcome to the content on the AI 900 certification exam.

0:10I'm really excited to be sitting down and recording this content

0:13because my passion, my sweet spot

0:16has always been leaning towards the cutting edge of technology.

0:21Whether we like it or not or whether we realize it or not,

0:24machine learning and AI is now relevant and prevalent

0:29in nearly every single tool that we're using behind the scenes.

0:32Even the devices that we interact with every single day

0:36are now being designed and built specifically for AI workloads.

0:41Hey, Siri.

0:44What is AI.

0:48See.

0:50This is AI.

0:51This is machine learning in action.

0:53You've just witnessed it right there.

0:55The actual phone device that I'm working with right there

0:58in the spot have the ability to process the words that I

1:02was actually using, the language, the speech,

1:04the tones, the dialect.

1:06It was able to detect that.

1:07I was asking a question and then specific details

1:10about that question, and then respond to it

1:13and reply with it.

1:14How they got Siri to do that with so much accuracy,

1:19that's really where machine learning comes in.

1:22So in this set of videos and throughout the rest

1:25of this course, we're primarily focused in

1:27on how Azure, the cloud, makes working with AI tools

1:32and machine learning and building out

1:34all of our processes to run in the cloud responsibly,

1:39safely, reliably, and cost efficiently.

1:43That's what Azure is bringing to the table.

1:46And that's why they have a certification track dedicated

1:49to just their AI tools.

1:50So right now, we're focusing on what can we even build with AI?

1:55Where is AI being used?

1:57And then we'll wrap that up with where does Azure come in?

2:01Then throughout the rest of this set of videos,

2:03we're actually going to focus in on what the individual tools

2:06are that Azure brings to the table

2:08and how they actually work with building your AI workloads.

2:11Let's take a look.

0:06Nowadays, most of the professionals or thought

0:09leaders in the AI or machine learning space,

0:12they've all come up with different terms

0:14for what they want to use to describe

0:16their workloads, their algorithms, or what it is

0:19in general that they're doing.

0:20So what is the difference between AI, and machine

0:23learning, and data mining, and knowledge mining,

0:26and deep learning, all of these different terms, cool buzzwords

0:30that you see get tossed around?

0:32Really, at the end of the day, it all

0:34boils down to three things.

0:37AI, which is the all encompassing

0:40artificial intelligence and then underneath that

0:43is what makes AI work?

0:46Basically, what is it trying to solve?

0:48What is it trying to do?

0:50And that's machine learning or data mining.

0:53These are two different ways of training

0:56AI to think intelligently.

0:58So let's explore this just a little bit more

1:01at the concept level before we start

1:03talking about what the different workloads are that use AI.

1:06Let's go.

1:08So I've got VS Code brought up here

1:10on the screen on my computer.

1:12And yeah.

1:12You're going to see me work a lot more with VS

1:15Code in an upcoming set of videos

1:17where we do a little bit of a machine learning crash course.

1:20But the first thing I wanted to do before we dive into this,

1:23is of course talk about what you can

1:25do with AI machine learning, data mining, and even more.

1:30But that's kind of the point is we hear all of these buzzwords

1:33get tossed around AI.

1:35Machine learning, machine mining,

1:39or knowledge mining, or data mining, or deep learning.

1:44And it's like, what is all of this?

1:46Are they that different from each other?

1:49Sort of.

1:50They all have their own little flavors.

1:52They all have their own little styles or dialects.

1:55But if you were to look at it as a tier,

1:57you would say all of it, all of these words,

2:01they all belong to the family of artificial intelligence.

2:05The idea with artificial intelligence when it was first

2:08surmised by IBM was could we teach a machine, a machine,

2:15to play checkers and eventually learn from all of the games

2:22that it plays.

2:24And it improves every single time

2:27it learns about moves to the point

2:30to where could the machine ever learn enough about checkers

2:33to beat the person who created the algorithm

2:38to train the machine.

2:41Ultimately, the machine its all purpose here

2:44is to learn like a human learns from experience

2:49and then derive predictions, make predictions

2:54about what should happen next.

2:56So if we're thinking in the game of checkers.

2:59If I'm hopping my guys around, and they

3:02make it to the other side, and you get to king them

3:05or whatever.

3:05You know what I'm talking about when we're playing checkers,

3:07right?

3:07Eventually, the machine is going to be

3:10learning from its own moves, as well as my moves.

3:14It's going to be collecting data on every single move

3:16that I make until it gets so good at understanding

3:21the game as well as my behaviors,

3:23that it can accurately predict what I'm

3:25going to do next and more or less front run me and basically

3:30beat me at my own game.

3:32So the idea here is we can take machines and artificially teach

3:38them how to take and ingest data and learn from that data

3:42in order to make guesses on what would happen next.

3:46So artificial intelligence is using different learning

3:50techniques to ultimately make the predictions

3:54in the applications that we use every single day.

3:57So that's where the rest of these things come in.

4:00Now, if I were to really break it down,

4:02you could pretty much safely say the rest of all

4:07of these buzzwords really boils down to machine learning,

4:10and split it in half, and data mining.

4:14So this kind of begs the question, OK, well,

4:16what are the differences of machine learning and data

4:18mining?

4:19Ultimately, what we're trying to do here

4:21is we're trying to find ways for the machine to learn.

4:26So that way, we're talking about if it's playing checkers again,

4:30and it's AI that we're talking about here,

4:36when it collects all of this data about my moves

4:38and its move, how is it going to figure out

4:40when something was good or when something was bad?

4:44And this is what machine learning and data mining

4:46really bring to the table.

4:48With machine learning, we typically

4:51start off with inputs and outputs.

4:55So for instance, if I'm going back to my checkers analogy,

5:00my input would be the moves that I'm making,

5:04and the output would be, were they good moves or not?

5:07Like what was the result of those moves?

5:10Did I end up jumping over one of their little guys,

5:13or did I get a king, or whatever the case.

5:15Or was it a bad?

5:16Did I make a mistake and set myself up for failure?

5:20So with machine learning, we have inputs,

5:23and we have outputs, and we usually

5:25already know or have a little bit

5:28of an idea of the correlation between these.

5:32If something came in and we know the output,

5:34we can determine if that was good or if that was bad.

5:37So we can train the machine using

5:40statistics and all sorts of other algorithms

5:42that we'll talk about.

5:44We can train the machine based on all

5:47of these inputs and these outputs.

5:49The more and more examples of this

5:51that we give you, the better or more accurately you'll

5:54be able to predict it.

5:56With data mining, it's a little bit different.

5:59In machine learning, we know the correlation

6:02between inputs and outputs.

6:04But what if we didn't?

6:06What if we had inputs and outputs

6:08and it was really at the end of the day

6:10just a humongous amount of data?

6:14And we don't really know because it's just so much data

6:17that we've collected, and it's maybe unstructured,

6:20we don't really know if this data came in,

6:22and that data came out.

6:23We don't really know how it got from this point to that point.

6:27So with data mining, we're training our machines

6:30to comb through the data and find relationships or paths

6:37to various outputs.

6:40And as we get more and more inputs,

6:42we get more and more results, we're

6:44able to find more and more correlations between the inputs

6:48and the outputs.

6:50So as we start to learn the correlations between the inputs

6:54and the outputs, then we can judge if this is going well.

6:59We can judge if this is a good correlation

7:02or a bad correlation and use that in machine learning.

7:06See?

7:07It kind of goes full circle there.

7:08So this is just really setting up a very short video

7:12to talk about what are all of the pieces that make up

7:15this puzzle?

7:16The idea is we're training machines

7:18to learn from data based on inputs and outputs

7:22and make judgments or predictions based

7:24on those inputs and outputs.

7:25And sometimes, we don't know the correlation

7:28between our inputs and outputs.

7:30And that's why we lean on data mining.

7:34And once we've got a good correlation, a good sample

7:37set of inputs and outputs and their correlation,

7:39we can train the machine to learn more accurately how

7:43to predict what comes next.

7:45So now in the next few videos, we'll

7:47take a look at real world use cases

7:49of where AI is being used.

7:51I hope this has been informative for you,

7:53and I'd like to thank you for viewing.

0:00[AUDIO LOGO]

0:06Language processing is the one that you just

0:08witnessed me do in a couple of videos

0:10ago when I was actually introducing this

0:12and I was talking to my phone, asking it a question.

0:15The idea here is that we have a machine, like an application

0:19that's just running, and we want it

0:21to be able to understand questions,

0:24maybe coming from a customer who's

0:26chatting to a bot for help, or maybe coming

0:29from my voice to my phone device, or my watch,

0:31or whatever the case is.

0:33The core concept of it is that it understands questions.

0:37It understands the language.

0:39It understands the tone.

0:41And it understands the task at hand.

0:43Is it a question?

0:44Is it a command that's asking here?

0:47So let's talk more about what language processors

0:50bring to the table and where you interact with these daily.

0:52In particular, we got a fun one coming up,

0:54talking about ChatGPT.

0:56Let's go.

0:57So I don't want to say something like the most popular form

1:01of AI that's being used, but I will say something

1:04like the most mainstream form of AI that's being used

1:07is language processors.

1:10Now, what are language processors?

1:12Language processors can listen or read languages.

1:19That's kind of the idea at the end of the day.

1:21And from these language processors, we can learn a lot.

1:26It's not just translation or interpretation

1:30that we're talking about.

1:32We can actually look at language or listen to language

1:36and detect things like the tone.

1:38Can we tell the sentiment of what is being said?

1:42We see this a lot on Twitter.

1:45Let's jump on Twitter for a second

1:47here and see what's going on on Twitter today.

1:50See, I see some pretty cool things.

1:52I see someone's auctioning some artwork off.

1:55I see someone is telling you to go outside, move your body,

1:58and enjoy nature.

2:00We see someone's advertising a video game here.

2:03It's kind of what's going on.

2:05And then we've got some jokes from like Barstool, or whatever

2:07the case is.

2:08I don't know.

2:08The point is if I were to put this into Azure's--

2:14Azure has a tool here.

2:14I'm spoiling it for you--

2:16hang on, I hit a button--

2:17Azure's Cognitive reasoning or learning tools.

2:22I can tell Azure, parse this tweet.

2:25And tell me if this person is happy, sad, angry.

2:31We can ask them, what type of mood is coming from this tweet?

2:35Now, take it a step further.

2:37What if I wanted to say, hey, I'm

2:38running a campaign that tracks hashtag #DataKnox right here.

2:43And everybody's going to say something

2:46about hashtag #DataKnox.

2:48And I want to figure out if all of these tweets

2:51that are hashtagging #DataKnox, I

2:53want to figure out if they're happy or mad or sad or euphoric

2:59or whatever.

2:59I want to figure out what the sentiment is

3:01for this particular trend, for this particular thing that's

3:06happening in the world.

3:08And that's where, again, Azure tools like Cognitive reasoning

3:12and learning come into play.

3:14And this is actually where you see like these trends

3:18and what's happening right now and understanding what

3:20the sentiment is around them.

3:22All of this behind the scenes is AI and machine learning.

3:26In real time, effectively, Twitter

3:29is analyzing the amount of mentions for Mac mini,

3:33and hashtag #MATIC, and whatever Antonio Brown is into now.

3:37It's determining what is happening in its own feed,

3:42in its own data.

3:44And then it even determines, why is it trending?

3:47What is the sentiment around the things that are trending?

3:50And who should I be promoting this to?

3:53Based on predictions, based on things

3:56like-- hey, this guy Knox, he really

3:58likes these types of tweets.

4:01So we see that this type of tweet is trending right now.

4:04And we know that Knox likes those types of tweets.

4:06So we predict that this should be in his feed.

4:10All of this comes from processing language and just

4:13understanding how end users interact with that language.

4:17But in recent history, language processors

4:20have become so, so much more thanks to a tool

4:25called ChatGPT.

4:27Honestly, I say, thanks to a tool called ChatGPT.

4:29Tools like this have existed for a long time.

4:32But ChatGPT has really gotten it to a place

4:35where it's really special.

4:36Oh, no, it looks like it's down right now.

4:38Hang on a second.

4:39OK, I got in.

4:40It took me a little bit, but I ended up

4:42going to the actual beta version of ChatGPT.

4:45That's chat.openai.com/chat.

4:50That's how I end up here on ChatGPT.

4:53Now, the idea here is that I can effectively tell

4:56ChatGPT to do anything for me.

4:59And it's not just answer my questions like looking up

5:03things in Wikipedia.

5:04No, I can tell ChatGPT to create things for me

5:09and give it really specific details that I want it to use.

5:14So down here in the little chat box right here,

5:18you tell it what you want it to do,

5:20like explain how Microsoft Azure benefits users

5:28who work with artificial--

5:32got to spell "artificial" correctly-- artificial

5:34intelligence, something like that, just on the fly.

5:37And I press Enter here.

5:39It types it in, and here comes my results.

5:42Look at this.

5:43Do you see?

5:44So it's just looking this up.

5:45That's a very basic use case of how this comes in.

5:48Look at this.

5:49I mean, it gives you your answer right there on the fly

5:52because it understood exactly what it

5:54was that I told it to do.

5:56So I'll let it get done generating the response here

5:58because it's going to type out a list of really cool things

6:01that Azure brings to the table for people who work with AI.

6:04And then I'll show you how this goes even further.

6:08What's interesting about this, now that it's done,

6:10I can ask follow-up questions.

6:13Even cooler, I can even challenge AI.

6:16I could say things like, are you sure that's right?

6:19Or I don't think that's right.

6:21And it will actually learn from that and correct its mistakes.

6:26I can follow it up with, tell me more

6:29about GPU-enabled virtual machines in Azure.

6:37So we let it do some thinking, and here it comes back.

6:40And it tells me more about Azure-specific information

6:44of GPU virtual machines.

6:46And it even tells you how.

6:47This is parallel processing-- a high degree

6:50of parallel processing power required

6:52for deep learning and artificial intelligence workloads.

6:55Keep in mind here, it is keeping on topic

6:59for the initial conversation of AI.

7:04So now that it's done answering all of that, watch this.

7:07I'll say, now create a poem about Azure AI offerings.

7:16Watch what it does.

7:17Just watch.

7:20Here it comes.

7:24Do you see?

7:25Do you see what's happening right here on the screen?

7:28I didn't teach it how to do this, but it did.

7:31It learned how to do this.

7:33Look at this.

7:33[LAUGHS] It's kind of awesome.

7:37This is why I think language processing has

7:41taken on a new level.

7:42I mean, as of time of this writing,

7:44ChatGPT has really only been a mainstream tool

7:48for like a month at this point.

7:50So we're very early to ChatGPT as a tool.

7:54But it not only is able to follow up

7:57in a dialogue with follow-up.

7:59I can actually tell it be creative.

8:02I can tell it be creative with things.

8:04I can even tell it create full-blown applications for me.

8:09Now, that's a slightly different form

8:12of the language-processing tools that OpenAI brings to the table

8:15and offers.

8:16But it does actually have a code generator.

8:21You can tell it things like, using this programming

8:24language, using Python if you want, generate

8:27an application for me that tracks

8:30stars in the sky at night.

8:31I don't know.

8:32But you could tell it to do that,

8:33and it will literally spit out an application for you

8:36in real time.

8:37So it's kind of crazy.

8:39But where else do we actually see creativity and AI combined?

8:44Let's do AI art.

8:46AI Art Generator, look at this right here, hotpot.ai.

8:50So I can tell it to generate art on the fly.

8:53What should AI draw?

8:55How about a computer hacker in knight's armor?

9:01How about that?

9:02And then we'll scroll down and tell it Create.

9:05Boop, there we go.

9:06So let it do a little bit of thinking.

9:09And in roughly 10 to 15 seconds, we

9:12should get some form of artwork that generates a computer

9:16hacker in knight's armor.

9:18So let it spit and spin for just a second.

9:20And then once it's finally generated the artwork,

9:22I'll circle back to it.

9:24And there it is.

9:26After about five more seconds, this is my computer hacker.

9:30It's actually really cool, right?

9:32I'm going to save that.

9:33It's actually really cool artwork.

9:35So anyways, you could tell, using AI art generators,

9:39it can process the language of what I expect to see.

9:42And because it's also processed images,

9:45things that we're going to talk about in the next video--

9:48because it's processed a boatload of images,

9:51it knows, based on these language inputs, what accurate

9:55outputs should look like.

9:57I would say it got this to a pretty good degree of accuracy.

10:02I mean, the armor here, this looks more like cloth

10:04down here.

10:05So I would kind of argue this isn't like a full knight's

10:08suit of armor, but it's got a cool looking helmet thing.

10:11And maybe this is like some shoulder pads or neck pads.

10:15I don't know.

10:15Maybe it's like just an armored hoodie.

10:18I don't know.

10:18But the point is, to a degree of accuracy,

10:21it was able, within a matter of seconds,

10:24to understand what it was that I was telling it to do and spit

10:27out some results.

10:29So the thing about-- you know what, actually,

10:31I was about to get it worked up.

10:32I was about to jump ahead.

10:34We're saving it for another video.

10:35Just hang in there.

10:37All right, this has been understanding

10:39language processing.

10:40It's all about taking instructions

10:43or listening to language and knowing what the request is

10:46and how to respond to it.

10:48I hope this has been informative for you,

10:49and I'd like to thank you for viewing.

0:00[MUSIC PLAYING]

0:06The next fascinating thing that we are going to talk about

0:09is video object detection.

0:11At the surface level, you may look at this

0:13and go, oh, that's a cool gimmick,

0:14but I don't really see what the real world use case for this

0:17would be.

0:18Then, we'll talk a little bit more

0:20about where this is really being used today,

0:23and some of the business use cases for it.

0:25And this part will blow your mind.

0:28Let's talk a little bit more about it.

0:30So another way that you'll find some pretty fascinating things

0:33in AI is just right here on Twitter

0:35because people are doing really incredible things with language

0:39processing as well as video processing.

0:41So in the previous video, we took a look

0:43at language processing.

0:45I just wanted to circle back to.

0:47This is really cool.

0:48This particular artist is an artist,

0:50and actually a fascinating AI follow in general.

0:53You probably want to give them a follow.

0:55I asked AI to turn countries into women,

0:59and it actually created some pretty fascinating results here

1:02where it took the generic styles and imagery of people

1:07who are from those countries and turn them into pictures of,

1:10I guess, what women would look like from those particular

1:13countries.

1:13It was fascinating to look at what I thought.

1:16But now we focus in on what can I do with videos?

1:20And right here, you're looking at it.

1:22You see this?

1:23Isn't that wild?

1:27Basically, these are significant people throughout history,

1:33whether they're fictional or not,

1:35and we see how AI has the ability

1:38to track exactly what's happening

1:41on these individuals faces.

1:43As I was sitting here drawing, we

1:44see how AI has the ability to track exactly what's happening

1:48on these individuals faces, and alter

1:52the artwork or the video work or the actual images

1:57or videos themselves to mimic exactly

2:01what the original source person was doing.

2:04AI now has the ability to track every single one

2:07of these aspects of an end users face

2:11and understand exactly what is happening on that face, which

2:14I think is just so fascinating.

2:16We even see this in Teams, Microsoft Teams or Zoom.

2:23You've ever seen that blur out the background?

2:26How does it know that was the background?

2:29How does it know where my body is?

2:31It's because it's used enough machine learning

2:35from processing images to know what the outline of a human

2:39would be.

2:39It knows where the eyes are, and it knows,

2:43oh, well, this little portion here must be the nose.

2:46And it knows, oh, this little portion here must be--

2:49I mean, it's the worst drawing of a face ever.

2:51But this must be the person's mouth.

2:53And it can detect where all of these objects

2:57are on the screen, and then draw an outline of it.

3:01And that's why it knows how to track the facial movements

3:05and the body movements and blur out

3:07everything else behind them.

3:08It's because it's looked at enough images

3:11and been told what is the outline that it should expect.

3:15But now, even more fascinating, sets up

3:18this particular conversation, which

3:20is about tracking objects in videos again.

3:23Oh, I want to show you this one.

3:24This is pretty fascinating here.

3:26This is not Tom Cruise, but it looks like Tom Cruise,

3:28isn't it?

3:29See?

3:30He's doing the Wednesday Addams dance.

3:33But Tom Cruise didn't actually do the Wednesday Addams

3:36dance on TikTok.

3:38But somebody made an head transmission,

3:41and that's like 40 years ago Tom Cruise too.

3:45That's Risky Business Tom Cruise.

3:47Or maybe in Mission Impossible 1 Tom Cruise.

3:50OK, I got distracted.

3:51This is the one that I really wanted to show you though.

3:54This is real footage of people, and what we're seeing here

3:58is AI has gotten so good at training what is happening

4:02on the screen, it knows when to outline a person

4:05and track exactly where a person is on the screen.

4:08It also shades people on the screen-- hang on one sec.

4:11Let's go back here for a second.

4:13So what we're seeing here is people doing parkour,

4:15they're jumping all over stuff, and it outlines the whole body

4:18of the person.

4:18But did you notice-- hang on one second.

4:20Let's see if I can get right back here, right there.

4:25Do you see?

4:25It didn't even outline a whole person.

4:27You just got a head.

4:28And it was still able to detect, oh,

4:30and instantly, it was able to detect, oh, this is a person.

4:35There's their face.

4:37We're going to outline the shape of this person

4:39and shade it red.

4:41Cars are going to be orange.

4:44We're certain that's a car.

4:45And it says right there, you can see of the little dialog

4:49box it says, car.

4:50Up here in the pink, it says potted plant

4:54and it's shading that pink.

4:55So let's keep playing here.

4:56Let's actually go full screen with this.

4:58Let's actually record this full screen.

4:59That way, you can see-- isn't this fascinating?

5:02The fact that in real time this video footage can absolutely

5:05track exactly what a person or an object

5:08is doing in this exact moment is-- look at this.

5:10This is really cool.

5:11Look at that.

5:11You see this?

5:12Boats way out there in the water.

5:15It's able to detect these boats in real time.

5:18I couldn't even tell that was a boat.

5:20[LAUGHS]

5:21It's able to detect these boats out there in the water.

5:25Now it's confusing maybe some train cars with boats,

5:28but still, we're still doing these things

5:30with a degree of accuracy in tracking exactly what they're

5:33doing in the footage.

5:35So what's the use case for this other

5:38than it's really, really cool?

5:41How is this actually used?

5:42And this is the part that's going to blow your mind.

5:45The part that blows your mind is now our cameras

5:48have built in chipsets.

5:50These become more than just cameras.

5:51They are really IoT devices.

5:53And they actually have the machine

5:55learning AI object detection built directly

5:59into the camera itself.

6:01My first real experience with this

6:04was with demoing the Cisco Meraki line of cameras.

6:07I actually was able to get my hands on some Meraki cameras,

6:10and you can find training on how to use and implement

6:14Meraki cameras right here on CBT Nuggets.

6:17The fascinating use case for this with Cisco Meraki

6:21was not only can it detect people on the screen,

6:26it can then use that data to determine

6:29how many people do I have?

6:31And are those people close to objects?

6:34Does this create a problem?

6:36If it does, let's do something about it.

6:40The use cases that Cisco Meraki likes to brag about

6:43is let's say we've got a camera that's looking down

6:46on a checkout line at something like Target or Walmart

6:50or whatever.

6:51And it sees, OK, we've got two people.

6:54We've got the boxed in with purple

6:56and maybe they're shaded in with purple.

6:58We've got two people in the checkout line, not a big deal.

7:00But all of a sudden, that checkout line

7:03gets busier and busier and busier and busier.

7:05And now all of a sudden, people are

7:07mad because they have to wait in line.

7:09So what Cisco Meraki can do is it can use this object

7:14detection to detect people, and when too many people get

7:19concentrated in a specific space or maybe they're

7:23too close to this particular checkout line object,

7:26Cisco Meraki can now send an alert to an application

7:31to the manager that says, hey, you

7:33need more people working registers.

7:35It can alert them.

7:37Even more fascinating, we may have

7:39a dangerous piece of equipment.

7:44Something that might be prone to exploding,

7:47if things go particularly wrong.

7:50What I can do, what Cisco Meraki can do

7:53is it can do things like detect people in proximity

7:58to this dangerous piece of equipment.

8:01And if certain conditions or thresholds aren't met well

8:05because now too many people are getting

8:08too close to this equipment and they're not

8:10supposed to be close to this equipment right now,

8:12Cisco Meraki can send alert signals to those people,

8:15or it can even take actions to shut down the equipment.

8:20Now hopefully, when I circle back to this object detection

8:25video that we were just looking at here, now hopefully,

8:29you're seeing this in a different light.

8:31Oh, here's people moving in real time,

8:34and we can detect exactly what they're doing

8:36and maybe even predict where they're going next.

8:38And if danger lies in front of them,

8:41this is what makes Tesla's autopilot work, right here.

8:49This is what tells Tesla cars when they're on autopilot

8:52and they're driving on their own,

8:53if somebody runs in front of the street,

8:55this real time detection of human objects

8:58is what tells the Tesla to stop.

9:02See?

9:03This is really fascinating stuff.

9:05And even if it's not the area of AI

9:08or machine learning that you're interested or working in,

9:10you can now really appreciate what

9:13it means to be detecting these objects in real time,

9:17and making predictions about where they're going to go next

9:20and if danger lies in front of them.

9:23So this is the point of image detection

9:27and object detection and data processing on images.

9:31I hope this has been informative for you,

9:32and I'd like to thank you for viewing.

0:00[AUDIO LOGO]

0:06With machine learning, the goal is very purpose oriented.

0:11We're trying to teach the machine

0:12how to process a specific language

0:15or how to process and detect things in a specific video.

0:18What if we just have a tremendous amount of data

0:22and we don't really know what it is trying to tell us?

0:25At this point, what we're really focused in on

0:28is finding correlations in our data.

0:31What are the paths that our data can take us down?

0:34And that's what data mining is all about.

0:37So in this video, we are focused in

0:39on just elaborating a little more on what data mining brings

0:41to the table.

0:42Let's go.

0:44I'm not a big fan of just reading Google results

0:47from the screen, but this particular definition

0:49that comes from Rutgers is a really great definition.

0:52The process of using computers and automation

0:56to search large sets of data for patterns and trends,

1:00that's really it.

1:01That's where I'm going to leave it, because the idea here--

1:04I want you to think about this-- is some businesses collect

1:08so much data, so much extraordinary amounts of data

1:12from it, that they don't even know what it means.

1:15They don't know what it's telling you.

1:18I want to tell you right now--

1:20your credit card company, maybe it's Visa.

1:22Maybe it's Mastercard.

1:23Whether you like it or not, or whether you know it or not,

1:26Visa knows everything about you.

1:29They know not only the obvious things like where you live.

1:33They also know how much money you make.

1:35They know what kind of spender you are.

1:38They know where you like to go on vacation.

1:41Or they can predict with accuracy

1:43where your next vacation will be.

1:44They know how many kids you have and what gender are those kids,

1:49without you ever telling them.

1:50They look at every single detail about your spending habits.

1:56What specifically did you buy from Old Navy?

1:59What specifically did you buy from Amazon?

2:03Where did you go on this last vacation?

2:05And how much money did you spend on that vacation

2:07internationally?

2:08How much do you like to eat out?

2:11And in particular, what are your favorite genres of food?

2:15Are you a Mexican person or an Italian-style person?

2:18I like it all, personally.

2:19I've never met a piece of food I didn't like.

2:21So the idea here is that businesses collect just

2:25extraordinary amounts of data.

2:28In particular, like I'm saying, Visa, Mastercard, they're

2:32collecting so much data about you

2:35that if they were to map it all out,

2:38it would be thousands of data points about you

2:42that these places, these businesses have collected.

2:44You could extrapolate that towards social media.

2:47Heck, remember all of this language processing

2:50that we've been talking about, LLP, Language Processing, and I

2:53was mentioning things like Siri or Alexa?

2:56Those devices, while they're primarily designed-- hang on,

3:00Alexa is listening to me right now.

3:02Alexa, stop.

3:04See?

3:05Ask me to play any song or--

3:07Those devices are 1,000% listening to you

3:11so that they can collect data to more accurately predict

3:16what it is that you're saying.

3:18I mean, really, their end goal is

3:20to get better at listening and understanding the language

3:23and answering your questions more effectively.

3:26But at the same time, they're also

3:29collecting data points about you and learning things about you

3:32so that when this device hears you talking about, oh, I

3:35need to add this to the shopping list,

3:37well, the next time you log in to amazon.com,

3:40guess what's on the home page recommended for you,

3:43because they collected the data on you.

3:45The kicker here was at some point, all of these things, all

3:50of these businesses, all of these tools,

3:53they were all designed to collect data.

3:56And now that they've got thousands

3:58upon thousands, if not millions, of data points

4:02just about you on their hard drives, now

4:05they're left going, OK, well, what does that mean?

4:08What does this data mean?

4:11How can I take info about a person and map that

4:16to shopping habits and recommend things to them that they

4:20might actually like?

4:22It all started with data mining, where

4:26we have to take an input, which is details, all of this data

4:29that we've just collected, whether it's

4:31voice clips or spending habits or just

4:37like even medical history, things

4:39like that-- all of these businesses and these hospitals

4:42and everything out there is just now data.

4:45You're just ones and zeros on hard drives.

4:48Now what they're doing is they're

4:50taking all of this input data about you.

4:53And then they're looking at what they consider

4:56to be output data or results.

4:57You bought this.

4:59And they're trying to find the correlation between which

5:03inputs drove this output.

5:07How did we get to this point?

5:09And that is what data mining does.

5:12It searches large sets of data for patterns and trends,

5:18turning those findings into business insights

5:21and predictions.

5:23Based on the data that we mined, learning the inputs of you

5:28versus the outputs of your shopping habits,

5:30we now predict--

5:32keyword here-- that you would be interested in buying

5:34this thing next.

5:37See how data mining comes to be?

5:39It's all about we don't really know

5:42how this input led to that output the way

5:46we might with machine learning.

5:49When we think about the language processor or the video object

5:54detection, with the language processor,

5:56we typed in language.

5:58We got an output, and we could then score that output

6:01and let it know if it was good or bad.

6:03With the video object detection, we

6:05were able to give it video footage.

6:07And then we were able to tell it this is a person.

6:11So the more persons that it identified,

6:14it was able to know, OK, well, this is accurate or not.

6:17With data mining, we don't really know this correlation.

6:21And that's why it's trying to find these trends.

6:23And this is particularly useful amongst gigantic data sets that

6:28are particularly unstructured.

6:31Data mining is one of the places,

6:33in my opinion, that Azure really shines because Azure can

6:37be a data engineers playground.

6:40It's all about aggregating data across a bunch

6:43of different sources.

6:44Data can come from CSVs.

6:46It can come from SQL databases.

6:48It can come from document databases,

6:51where they store JSON files.

6:52It can come from Parquet files.

6:54Azure is great at connecting to all of those data sources

6:58and aggregating them into a single data source

7:01that we could then process with data mining.

7:04So now that we're seeing this come full circle

7:07and we understand where data mining is,

7:10let's talk about the next thing that comes up,

7:12one of my favorite things, anomaly detection.

7:14I hope this has been informative for you,

7:16and I'd like to thank you for viewing.

0:00[AUDIO LOGO]

0:06Data mining can sometimes be used in conjunction

0:10with machine learning for a very important reason,

0:13and that's to detect outliers.

0:16Once we have a tremendous amount of data

0:18and we know what is the normal pattern of relationships

0:22for our data, when something really weird stands out,

0:26recall that an anomaly.

0:27This doesn't really fit in with what we know about our data.

0:30So what is happening in our environment that's

0:33particularly weird?

0:34Anomaly detection has now become one

0:37of the most important facets to actually managing

0:40today's enterprise networks.

0:42We see this from Cisco.

0:44We see this from Juniper.

0:45We see this from Palo Alto.

0:47And really, any provider that works with security at all

0:51is now leveraging machine learning

0:53data mining, and AI in general, to help them detect anomalies.

0:58So in this video, we're going to focus in a little more on what

1:01anomaly detection is all about.

1:02Let's go.

1:04So I'm getting logged in right now

1:06to Cisco DNA Center, courtesy of Cisco's D Cloud environment.

1:12This is really a data center cloud

1:14based environment that is designed

1:16for people to demonstrate, or demo,

1:18or just take some of their tools for a test drive.

1:22Now the DNA Center is a full enterprise network management

1:27solution.

1:27Basically, if you have--

1:29think about a large college campus, all of the buildings,

1:32all of the wireless access points, all of the people,

1:36all of the user accounts, all of the different firewall rules

1:40and firewall settings.

1:41There's a lot that goes into managing an entire enterprise

1:46grade network.

1:47And then making sure they're mobile

1:50throughout the entire enterprise and

1:52throughout the entire campus--

1:53these are people walking all over the place.

1:55They're going everywhere, right?

1:57But whenever person a here goes from point A

2:01to point B-- here they are not on point B--

2:03they still need access to all of the resources and the rules

2:07that they use when accessing the network.

2:10So making sure those permissions follow them

2:13across this huge network and this huge campus,

2:15it becomes a bit of a pain.

2:17And that's what the entire point of Cisco DNA Center is.

2:21It's all about provisioning and managing

2:24and monitoring large scale networks,

2:26and making sure the users who connect to that network

2:30have the same permissions and security rules

2:32as they migrate through the entire network.

2:35Now one of the biggest things that DNA Center does

2:38is it actually collects data.

2:40It collects a humongous amount of data

2:43on the users and the network devices.

2:46And it's able to determine, with a certain percentage,

2:48first of all, what's the health of our network devices?

2:50How are things going?

2:52It's able to detect critical issues,

2:54and it's able to even detect security issues that pop up.

3:00It does all of this in one of the four workflows

3:03that it has here.

3:04In particular, once your network is online,

3:07you then monitor it and manage it under the Assurance section.

3:11Now this isn't a Cisco course.

3:12This is just to demonstrate some fascinating things for you.

3:16It presents you a bunch of different dashboards,

3:18like the health of your network, any issues or events that

3:21happened on your network, sensors

3:23that you have in your network, Wi-Fi

3:25that you have on your network, any rogue wireless

3:28issues that you may have, and even power over Ethernet.

3:31And it collects data on all of these things.

3:35Now here comes the crazy cool part.

3:36It uses machine learning under the hood,

3:39as it's collecting a humongous amount of data

3:42from thousands of users or devices

3:45across your entire campus.

3:46And it learns what's normal.

3:49It learns what's normal and establishes

3:52a baseline for your network.

3:54That way, when something weird happens,

3:57it immediately flags it as an anomaly

4:01and raises an issue or an event.

4:05So I want you to consider this for a second.

4:07If I jump into Assurance, and then I

4:09go into Issues and Events, I want

4:11you to think about how your IT ticketing systems might work

4:14currently in your IT infrastructure or your software

4:17development world.

4:18Typically, a person opens a ticket.

4:21It's a complaint or question.

4:23And then somebody has to address it.

4:25But what if I told you DNA Center

4:27has the ability to detect things before a person ever

4:31would, and maybe take remediation

4:35steps automatically, or maybe open up a ticket for you

4:39to investigate immediately.

4:40This is what you're looking at here

4:42when you start looking at the DNA automation and services

4:46here.

4:47And in particular, this works hand in hand

4:50with security appliances and implementations.

4:53So not only are we detecting anomalies just

4:56in the network health in general,

4:58we can also detect security based anomalies.

5:01Oh, you know what?

5:01This signature of this application looks off.

5:04Or we know this to be bad.

5:06So let's raise an issue and shut down

5:08these particular ports before contagion spreads

5:12throughout my entire network.

5:14Anomaly detection is now critical in every data center

5:19and every ISP around the world.

5:21And whether you've realized it or not,

5:24machine learning has now become a part of all of the hardware

5:28that we implement in our infrastructure now these days.

5:31Heck, even the MacBook that I'm using

5:33right now that I'm recording this on it

5:35has the M2 processor.

5:37And they've now dedicated a chunk of that M2 processor

5:41to just AI, neural networks, machine learning workloads.

5:45So that way, whenever I'm using XYZ app,

5:50maybe that app has some machine learning built into it.

5:53Well, it can leverage this processor

5:55to make my experience with XYZ app better.

5:59This is exactly what you're seeing here with something

6:01like Cisco DNA Center, where it's learning what is really

6:05going on in our network.

6:06How many users do we normally have?

6:08What are the workloads of these switches and these firewalls

6:12and these routers that we normally have?

6:14And if something on one of these looks really,

6:16really off, we know something is going on.

6:19Or rather, I should say DNA Center

6:21knows something is going on this particular part of the campus.

6:24So let's immediately create some sort of ticket or remediation

6:29steps to repair and heal something that is happening

6:34in our entire environment.

6:35This is also really, really popular

6:38on security information and event management tools.

6:42So these are the tools that are designed

6:45to collect log data or any type of data source

6:48that we can possibly send to it.

6:50This is specifically built to find anomalies

6:55in security issues.

6:57We're talking about things that seem off

7:00with how our Active Directory is behaving, or are

7:03OAuth requests seem off, or something like that.

7:06Or an application was downloaded that meets a certain signature,

7:09and we don't quite recognize that signature.

7:11But it looks bad.

7:13So SIEMs are all about detecting those.

7:16And these, again, are all about machine learning.

7:19But it gets even better.

7:20With all of this, the big thing that makes it work

7:24is it needs to be accurate.

7:26It needs to, when it makes a prediction,

7:29it needs to be accurate.

7:30And the only way it gets accurate is with more data.

7:35When you typically have something like a SIEM,

7:38while, of course, it's not going to leak any private information

7:41as it's learning of bad signatures or anomalies that

7:45detected attacks, it actually reports this stuff

7:49to a global security network so that we can all

7:52share and collect more data on security events

7:56and more proactively predict and negate

8:00security issues when they arise.

8:03So this is what anomaly detection is all about.

8:06And these are the types of tools where

8:08you see it in action today.

8:09I hope this has been informative for you,

8:11and I'd like to thank you for viewing.

0:00[MUSIC PLAYING]

0:06So this is all great, but where does Azure

0:09come in for all of these different things

0:11that we can do with machine learning and AI?

0:14Do they just host our application and it's up to us

0:17to figure out the whole machine learning and training

0:19the model part, or can they help with that too?

0:22That's what we're focused in on in this entire course.

0:25That's what the AI 900 is for.

0:27It's to teach people, hey, now that you've

0:30got a little bit of an understanding about machine

0:32learning, maybe working with our tools could make

0:36your experience even better.

0:38That's what the 900 is about.

0:40So it's introducing the tools and letting

0:42you know what their purposes are for.

0:44Then as you progress into your certification journey

0:48and you go for a higher level of certification,

0:50that's when they start to teach you, here's how you use it.

0:54Here's how you get started with these tools.

0:56So start off with the AI 900 just learning what the tool is

1:00and what it does, and then in the next certification exams,

1:04you'll learn here's how you actually

1:06implement it in your solution.

1:07Without further ado, let's see what the cloud and Azure

1:11brings to the table.

1:13So where does Azure come in?

1:14Well, Azure comes in and the fact

1:17that we are going to be collecting or working

1:21with data, with data.

1:25Lot of data.

1:26Potentially, tremendous amounts of data.

1:29Terabytes of data.

1:30Petabytes of data.

1:32And in order to process these large amounts of data

1:36in a quick amount of time, we need specialized computing

1:41workloads to process this data.

1:43Whenever I actually create a machine learning model,

1:46if I'm trying to process millions, millions

1:51of rows of data and I need to process them quickly,

1:54I can't use my desktop to do that.

1:56Heck I can't even use the servers

1:58I have downstairs to do that.

2:00I need something truly specialized,

2:02a true workhorse to power this type of workload,

2:06and I don't have hundreds of thousands

2:08of dollars to shell out to get it.

2:10This is where the cloud comes into play.

2:12The cloud is scalable.

2:16This means that we can effectively

2:18get more computing, and more storage resources on demand.

2:24And when we're done with them, we tear them down.

2:27That makes it cost efficient.

2:31In particular, we like to say that instead

2:33of spending a humongous capital expenditure or CapEx shelling

2:38out 100k at once on a server that now has

2:42to last me a few years, instead of using a CapEx expense,

2:46I can now use an OpEx or an operational expense.

2:51I only pay for the resources when I need them.

2:55And I pay for them on sometimes an hourly rate, sometimes

2:59a per minute rate, sometimes it might be per task write.

3:04The point is I only pay for what I use when I use it.

3:08When I'm done, I tear the resource down

3:10and I stop getting billed for it.

3:12So I can achieve significantly more resources.

3:16I can get as many resources effectively as I need.

3:19Not just one server, I could get 100 servers

3:21if that's what I want.

3:23And run them for an hour and then tear them down.

3:26And then I'll get billed for effectively 100 hours,

3:28but that's a lot different than buying one Hawking server that

3:32has to be on all the time, and it's a sunk cost.

3:36That's the idea.

3:37Now, of course, your accountant may say, yeah,

3:39but then we get to depreciate the asset.

3:41Well, that's an entirely different thing

3:43and you'll have to fight with your accountants on that.

3:45But here, with this case, we have a predictable cash flow.

3:49And we have the ability to get the exact amount of resources

3:52that we need when we need them.

3:56Azure is incredible at ingesting and processing and securing

4:07and moving data.

4:12The data is redundant.

4:14There will be at least three copies

4:16of your data in a data center.

4:19And you have the option of spreading those copies out

4:21across multiple data centers all around the world,

4:24if that's what you really want it

4:26to do to make sure the data was secure.

4:28But now we have the ability to do

4:30large scale ingestion of data, as well as processing

4:34and computing of that data, and then

4:37securing and making sure that the data is

4:40backed up and redundant.

4:42And then once we have the machine learning models

4:46and we want to use those models in applications in the same way

4:51Amazon promotes or shows you resources

4:54that it thinks you might want to buy, in the same way,

4:57we can host these apps that use our models in the cloud.

5:02The idea now is that we can get reliable, always on, always

5:08secure, always scalable, cost efficient services as opposed

5:13to building everything or self, being

5:16responsible for the network and the operating

5:19system and security patches and access controls.

5:22We can offload a humongous amount

5:26of maintenance and management to the cloud

5:29so that we can focus on our code or on our solution.

5:35We don't have to focus in on IP address space, or firewall

5:40rules, or the operating system needs to be patched.

5:44And we don't have to focus that much on,

5:47did I buy enough server?

5:48Did I buy too much server?

5:50Is it always on?

5:51How much electricity is it burning?

5:53Who is specialized to work this server?

5:56Do you see how the cloud is all about offloading management

6:01tasks so that you can focus in on your solution

6:04while it always remains available online because it's

6:08replicated and it's redundant?

6:11The cloud gives us the ability to do these things.

6:14But even more so, Azure has a specific AI and machine

6:19learning section.

6:20If you log into Azure, go ahead and create your free Azure

6:23account now.

6:24They do ask for a credit card to be on file,

6:28but you don't get build anything until you actually

6:30spin up some services.

6:31So don't worry about it.

6:32You're not going to incur a cost until you deploy a resource.

6:36So spin up your account now at Azure.

6:38And once you get logged in like so, you can go to all services

6:42right here.

6:42And it breaks it down based on categories.

6:44And right here, we see AI and machine learning.

6:47Oh, look at this.

6:49Do you see?

6:50Computer vision, custom vision, anomaly detector.

6:54Here's those cognitive services.

6:56That goes hand in hand with language understanding.

6:59We even have the ability to create a machine learning

7:02accurate search tool here.

7:06Then we even have our own machine learning studios

7:10where we can tell Azure, hey, we're

7:13going to create some Python notebooks directly in Azure,

7:17and you'll have the ability to connect your Azure storage

7:21services, like databases or database backups.

7:26Oh, look there's even face APIs over there.

7:28I missed that.

7:28That's really cool.

7:29So anyways, this is the idea, is now

7:32Azure has the ability to ingest all of our data

7:36into storage accounts, and then we can actually

7:38do our Azure development right here in Azure

7:42and let it handle the scaling and workloads

7:45needed to process our data and turn it

7:48into a data model, which we can then use in our applications.

7:52Azure is here for that.

7:54The cloud is an amazing way to do it.

7:56And now when you have specialized tools and Studios

8:00and workspaces to actually do your work directly

8:03in the cloud, it becomes even better.

8:06So this page right here and these resources

8:10that you're looking right here on the screen,

8:11this is why you're here.

8:13This is why you came to the AI 900.

8:15You want to learn what these do, and when you would use them.

8:19And how you would use them, how you would implement them?

8:23That's the next certification exam.

8:26So for now, let's start to understand when we might

8:30use these different resources.

8:31What is the solution and what is the application look

8:34like when it's done?

8:35I hope this has been informative for you,

8:37and I'd like to Thank you for viewing.