Placeholder Image

Subtitles section Play video

  • so welcome to talk on the future of Java.

  • Script is Aye, aye.

  • My name is quick was broken.

  • Uh, my name is asked him.

  • Hussein, you can find me on Twitter as jaw ache.

  • Not Joe Ash.

  • A jaw ache.

  • Um, I blogged about angular and draw the strips on my website code craft dot TV.

  • I'm also a cloud advocate, Microsoft.

  • So if you don't know who we are, we are three open source advocates working on the azure cloud platform.

  • If you go to our get a link there, you'll find a final.

  • That's basically there's mean that's up there.

  • Before we focus on all different technologies on areas.

  • My focus is Java script and on weather.

  • Actually, I've gone into leadership mouse.

  • My focus is people.

  • But you're gonna find out more about us going to that website on just Ah, yeah, Chase people up.

  • I also work with a charity called Pile Internship, and what they do is they help find Palestinians internships in high tech Israeli startups.

  • I last year I I meant toward about 10 of them really hard work mentoring 10 people by men to 10 of them.

  • And this year I'm very, very proud to say that 17 off the interns all being mentally by one cloud advocate for Microsoft.

  • So if any of you work with or for at, ah, startup or or high tech company in Israel and you have a space on opening for him to give me a shout, I'll connect you with the program director.

  • I'll get the ball rolling.

  • Anybody here?

  • Uh, anybody here know about you, Demi?

  • Anybody hear my students on you, dummy?

  • Well, hi.

  • Thank you.

  • Thank you.

  • Pays the mortgage.

  • Um on You did me like this I get over a year ago now the added in automatic subtitling.

  • So they correctly sites up.

  • Spell my name correctly.

  • My name is awesome on DA This talk should be very asked him.

  • I also co organized a meet up group in London.

  • Aye, aye.

  • Java script at London's I co organized with my friend Ellen.

  • 1/2 off.

  • We started it early last year.

  • It's being running very successful.

  • Think about 1718 100 members now.

  • We run some pretty regular sessions all the way through last year.

  • And then what happened was at the end of these sessions, people would come up to us and they kind of give us a link to some.

  • Hey, look at this Java script.

  • Aye aye, Power Jobs kept up after occasion.

  • Isn't it cool?

  • We thought, You know, it's fantastic.

  • It's really cool.

  • Let's put them all together in a website.

  • Everybody else confined them.

  • That's what we did is we made a website called a Js rocks, and to go to a A.

  • I just rocks.

  • What you'll find is a whole collection off a I powered JavaScript applications.

  • Now that Java script and then the Web, which means you can play around them but also means you can usually view source and see how it's working.

  • But the very least, we'll come up with or have a link to the source code or an article or a video explaining how it was made of meat.

  • The idea is to inspire you with an application, and then, if you like it, you can dig into and learn a little bit more and more.

  • That's that's enough.

  • In my journey into a machine, learning and jobs got bean bean, That process, I recommend.

  • Do not yourself on this talk today what we're gonna do is we're gonna go through three off the applications, which you can find on a IGs, rock them and explained how they work on.

  • Then hopefully through that whole process you're gonna learn throughout the course of today for the course of this lecture talk, you know, learn a little bit more in a little bit more, little bit more about what's possible and achievable with a I machine learning in jobs.

  • Get right now.

  • I know it's been a couple talks already in this conference on these topics.

  • Um, I suppose my talk for a bit more higher level give you a bit more of a broader viewpoint of what's possible.

  • Let's start off with the first application.

  • It's called the Emoji Fire.

  • It's not trademarked.

  • It's just the merger fire.

  • So it made sense.

  • Um, this is actually my application.

  • I brought this application.

  • It's gonna change everything.

  • Yeah.

  • So what's it do given image?

  • It will detect any faces in the image.

  • It will then detect the emotion in the face.

  • It will.

  • Then any guesses it will notify it.

  • That's right.

  • Well detailed.

  • Grab the appropriate emoji stick on the face and post it right back at you.

  • Yeah.

  • Elon Musk has got nothing on me.

  • Yeah, that.

  • Yeah.

  • I am open for investment here.

  • Singapore started hospital.

  • You could use them a different way, so you can use that as a slack applications.

  • In fact, if you go to the modified dot com and click there after slack, but it'll add it to your own slack work spaces.

  • Right now, this that works.

  • So you basically find them a major?

  • This is my son.

  • Yeah.

  • Slash Modify.

  • Put the pasty image in.

  • Yeah, I see appropriate emerge.

  • It can handle multiple people.

  • So you got that one in together that they're, Well, I I are answers An age old question that has been plaguing people for Diego.

  • That's the Mona Lisa.

  • Smile.

  • Bigger.

  • So the motive fire does quite a lot is a whole lot of things that it's doing by saying the most impressive thing that it's doing is it's How the hell is it calculating emotion, Emotion, right?

  • It's kind of kind of complex things.

  • A lot of us can't really tell other people's emotion, but how's the computer figuring that out, right?

  • I've got good news for you is really easy is incredibly easy, setting up two steps, just two steps.

  • First, that detecting the facial features.

  • We know how to do this.

  • Now we've examined enough faces.

  • We know the different points that every single space has.

  • And so basically there's different algorithms you can get.

  • Some of them are very arithmetic and some of them a little bit more.

  • Ah, machine learning Basic essentials made.

  • There's lots of different albums you can use was given image will extract you those spatial features And every single face has won best features Has 70 step one Step two is really, really easy.

  • You just use a neural networks that said just two steps extract the features and then just train and used.

  • And your network lets me for on Germans explain you just just for this guy in the front.

  • I'll explain in your network just for just for him over there and nobody else knows what they are.

  • Now a neural network is has a basis in biology.

  • Okay, this is a new run.

  • You should have at least one of these.

  • Um maybe some of you.

  • I have a few less.

  • Then you did yesterday, but it has some dendrites going in a body, a body on backs on going out.

  • If enough electricity flows in through the dendrite, the body fires.

  • It's electricity out through an accident.

  • That's it.

  • So it does.

  • If you stick enough of these together, you get a brain.

  • I can think, right.

  • That's it that's in your network.

  • That's awful.

  • That's a brain, I should say.

  • Um, if I wanted to code this up in Java scripts, you probably kind of try and give it a go in a long time.

  • So I've kind of studied the computer science basics right now, but I still remember some of it.

  • Quick note.

  • Ah, create some edges.

  • Ah, graph theory.

  • No, I'm talking about EJ is going in and EJ going back out on.

  • Then what?

  • Your features?

  • One of the things you're you're pumping into it would you would you was the inputs.

  • And what's the output?

  • Let's say this neural network is very one.

  • You're on your network.

  • You're gonna pump in the day of the year 23 and maybe the temperature on that day.

  • That's the imports.

  • Then for each of the edges, you basically create random number.

  • You initialize the random number.

  • Math.

  • But round maybe that got initialized minus 0.5 Mary.

  • That's got 2.1.

  • You multiply them together, you add them up when you pump him food on activation function.

  • Okay, whatever that function pumps out, that's electricity going back out.

  • That's how you would simulate that in code, I said, just to see a little bit more about how that might work.

  • Boom.

  • Just this.

  • This is it.

  • Input features and multiply them by the weights, some of them together.

  • Whatever that number is, you passed into the activation function.

  • Whatever that pumps out, you pump out.

  • That's just a neuron.

  • What are you doing when I when I teach, if some of you coming to my workshop let it run at some point, the workshop will just be like it's just multiplication.

  • Yes, just multiplication.

  • Like a large part of it, right?

  • That's what That's what it's all about, right?

  • I'm basically there's loads of activation functions out there.

  • This is a very, very simple one.

  • Anything below zero.

  • Is there anything above zero is one.

  • There's loads of all the ones you can use.

  • Um, basically, you just grab a bunch of those neurons together.

  • You got a network?

  • It was a simple feet forward.

  • Neural network, huh?

  • It's raining.

  • Um, that's what you do.

  • You you connect them altogether.

  • But remember, you initialize all those edges with random numbers, right?

  • Well, then, for this canyon that we need a training set.

  • Let's say we've got a whole bunch of images, but we've already gathered.

  • We've already figured out where all the facial features are in this image.

  • Okay, this data sets out there that they're available with all this stuff, right?

  • Feature.

  • Really Figure out what the facial features are off this image.

  • I'm actually, we also know the emotion in it as well.

  • We pump those features in maybe the features of kind of the X Y coordinates of each pupil and other things.

  • We humped those in.

  • We multiply all across.

  • That gives us a number at the end, right?

  • We we though what does three mean?

  • We've decided, actually that in this neural network, zero maybe is unhappy and 10 is happy, right?

  • Someone has a human being, is looked at image and they said that So we know it's wrong.

  • Of course it's wrong.

  • You initialized a bunch of random numbers and multiply them together.

  • It's not gonna It's not finish right to start it off its course.

  • It's wrong, right?

  • But what we need to do, isn't it just those numbers based off of how wrong it is that we we know?

  • Because we know this data that it's off by five.

  • This is the magic.

  • Use an algorithm, use a technique called back propagation.

  • Okay, given how wrong the network is, how do you tweak those numbers?

  • The next time you run it, it's more right.

  • Okay, that's back propagation.

  • And that's what you do.

  • That's when it tweaks it.

  • Perhaps you keep him running it with your date and you keep on running with data until he gets less wrong.

  • Less wrong restaurant, less wrong until eventually what happens is given a image, and it correctly predicts that is eight.

  • Okay, that process is training a neural network.

  • All this during this tweaking those weights.

  • So the next time you go through, it becomes better and better and better.

  • Okay, That's how you could do it.

  • There are actually well known data sets out there that you can get the freely available, which have ah whole bunch of images 30,000, 30,100 Army loads of images.

  • Each image has bean.

  • Human being is gone through and determine what emotion is.

  • And you can build and training your network in order to detect emotion all so you could do okay.

  • Well, what's that?

  • What's been happening over the last two years?

  • One of the lots of things happen in the space of the last year.

  • What's really been happening in the last two years?

  • Isa commoditization off the service is especially the very, very commonly used models love different companies, Microsoft included, have been making them available via AP eyes.

  • And that's what I used.

  • That's what I use for this service.

  • You should call the face AP how we can get for Maja.

  • Um, I was very, very, very easy to use.

  • You basically just host on image, too, that you are l.

  • And it gives you some data back at the backhand.

  • It's using something exactly like I've showed you.

  • It's been trained in a much, much bigger data set over much, much, much longer.

  • Time to get a whole bunch of information for each image in the face image in the face in the image look, it gives you a face rectangle, the face, actual boots with emotion.

  • A whole lot more is all of the strips, all out.

  • Anger, contempt, disgust, happiness.

  • You just sadness and surprise between zero and worn.

  • Do you see your bed 300 or bad?

  • Yes, yeah, you cannot be hundreds unhappy.

  • I'm really sorry about that.

  • It's actually impossible.

  • I've tried.

  • I can't be 99 percent happy.

  • That's basically I used in the motor fire, right?

  • That's how I built it, that I detected emotion inside There.

  • There's the summarize.

  • No one that works.

  • It's incredibly powerful.

  • They're incredibly powerful, actually.

  • Conceptually, the really simple to understand, I explained, You're really basic one in about five five minutes.

  • Something like that, right?

  • Basic of other coming lesson learned here might be when I do teach workshops to get people asking questions afterwards, taught them how to do training for a neural network from scratch.

  • Then they asked me, I want to build this my my 18 AP eyes available that you do that right?

  • So first off, also especially these days, check to see there's an A P.

  • I already gives you the answer that you're looking.

  • Um, next demo.

  • Okay, since the mobile.

  • That and I'm fine.

  • So I run workshops of machine learning with Jarvis comes in one of the workshops in London, a guy called Turner.

  • Then build this demo on the center back to quite proud moment.

  • When you go workshop on your students, build something, sends it back to him.

  • Um, this is what does Basically, if you do, yeah, it's doing ah, search for unspool ash for puppies.

  • Getting that image back there.

  • That percentage in the bottom left are trying to figure out what's inside the image because he here is detectable.

  • Here is like a fountain.

  • Puppies.

  • 10% a terrier.

  • Not bad.

  • You search unspool ash, for I'm fine against in this image.

  • In it says 78% T shirt.

  • Not bad.

  • Okay, not bad.

  • It's an important lesson here.

  • Is that the only a p I request is being made here is to get the image the actual detection of what's inside the image is old and in Java script in the browser.

  • Okay, what we heard I know, I know.

  • We've heard of tensorflow already in this.

  • Ah, in this where these things called conferences In this conference?

  • Yeah.

  • I arrived last night from London thing quite well.

  • I think, um, the attentive close the tent tensorflow then slows a technology which is open just like Google 45 years ago.

  • Call number and allows you to run.

  • Um, since what 10 Sir, is actually a multi dimensional.

  • Right?

  • That's what that's what tense that means the high dimensional.

  • All right, well, I do kind of matrix calculations across a whole bunch of service GP use scaling out massively right building.

  • See?

  • Truly hard core.

  • Um, yeah, that's basically works full about, but star of last year, he created tensely Js Well, like that head, don't we that little dot Js Whenever they add that to things, we get really excited.

  • It's like butter, you know, Plane jacket, potato with butter.

  • It's nice, right?

  • Js was launched.

  • Now it's just a month or two ago icon over the last 10 1st summit reached 1.0.

  • So has actually reached a state of maturity.

  • Um, I know we hope we've gone through this before, but this is all we need to do not to use tensorflow.

  • Jesse's well if you mean you use this one, right?

  • If you're proper jobs, developer may be used.

  • No, no, But the important thing here is that that's all you need.

  • You know where to start playing around with machine running jobs.

  • But is it my first Hortense for dress?

  • I thought it was like node bindings, and you have to install tensorflow not to use it.

  • No, it's actually re potentially rerun from the ground up in Java script.

  • It was like a subset of what's available in the proper tense.

  • Labor has enough to do some pre some pretty cool stuff.

  • Um, and you couldn't help wanting things that you can do.

  • What I just described you really wrong with training model, the whole thing with the neurons and then the way to the back propagation.

  • All that shooting stuff you could do the intensely.

  • Jess, you can't, right?

  • Oh, you can let pre trained models what we like doing pre trained models.

  • If you go to 10 for Jess Website, you actually find now there's like growing a set off pre trained models, very easy preacher and mortars.

  • You could just grab from their website and start using you can also is also no scripts available where you can take a model chained up using tensorflow proper proper that that's the right word.

  • But and then convert those models for use in tensorflow.

  • Jess so you can start using them essentially was just cut to the chase and file available.

  • You can load it up, um, from within the browser.

  • That's kind of all of the used to use chemical Mobile net Ah, which is just a very optimized neural network that can detect one of 1000 things.

  • Only 1000 things.

  • That's why it wasn't very good.

  • 20,000 things that can detect in the world.

  • Um, but his mobile.

  • So it's an optimized size that could be small to use on a mobile.

  • Um, busted on your four lines of code, I say it's crazy.

  • Four lines occurred in Java script to do that right, I'm gonna I'm gonna I'm gonna get this wrong statistically, but I think it was 2008.

  • Was it way still couldn't detect whether an image was a cat or a dog, right?

  • Using whatever service that we could now in jobs, that we could do this crazy.

  • Um, but It's my bonnet, right?

  • So it Zeke, you're not very good.

  • You So it wasn't very, very good.

  • It wasn't that good, right?

  • It's useful for a love of the thing that's useful.

  • Something transfer learning.

  • It's useful for some use cases, but not toe really know what's in an image.

  • Like if your life depended on it, you wouldn't use me up on that, right?

  • You want to use somebody's been trained upon a much larger data set with much more features.

  • These things are kind of gigabytes, terabytes, big and not not big enough to download on computer, but they're available via I'm gonna pretend something that a p I C U said FBI.

  • Yeah, a p I is correct.

  • I mean, we got another one computer vision, Um, and you can you can detect like an insane amount of information.

  • If you give an image.

  • One of the really scary things it does it shocks me, still scares me.

  • Actually, is it given image and they can give you a human readable description of what's in the image as if a human being wrote it right?

  • It's my friend Sarah Jasmine made ah ah, cool demo with their she thought.

  • Wouldn't it be cool if you could use that to describe the old text, the old tack and images?

  • Because we will do that.

  • Right?

  • Roll.

  • Supposed to add old tax for images.

  • Right?

  • Roll doing it.

  • So this is completely unnecessary.

  • I know, but he did want to use that.

  • You could try out.

  • It's a code pen so you could put in an image on the text of top.

  • Um, then we know this TV show.

  • No.

  • Wow, You're in for a treat.

  • Halt and catch fire.

  • It's an amazing TV show.

  • It talks, talks about starts of income in 1985.

  • Talks about computing back in those days.

  • And she's one of the stars in the show is Mackenzie Davies.

  • This is when you give this image.

  • This is the This is the caption it gives.

  • It was Mackenzie, Davies and all standing in front of the building, right?

  • Free Pretty good for service.

  • But you know, she released on Twitter on the Twitter is full of really helpful people.

  • No, over very supportive.

  • Kindly tell you when something doesn't work.

  • All right, So some people said so They said ah Oh, this is the work So star filled Sky Texas for I don't know, Maybe the Asian.

  • I don't know what they were.

  • Characters.

  • That is what it is.

  • It's not accurate.

  • That's true.

  • No, the next one, I think, is ah, could be true.

  • Could be true.

  • No, I mean, you don't know.

  • Let's be dead cats stuffed on a bed.

  • Really good ones.

  • Um, I really like the next one because because you're half right half right on the next one.

  • No.

  • Yes.

  • Good.

  • 50%.

  • That's past Mark.

  • Where I come from, I'm So, uh, does that summarize this for with one takeaway to take from this?

  • Just tensorflow just doesn't have any dependencies gonna play around machine learning you want quickly.

  • A servant in the browser confusing knowed.

  • Well, actually, I take that back.

  • Tensorflow chess in the does have dependencies.

  • Potentially Jess in the browns.

  • You can you stop playing around with it with nothing mobile?

  • There's a very simple way to analyze images and its uses.

  • Something becomes the works out later on is how we start off.

  • He's going to start off with mobile on that, um, but if you really want to some kind of really deep image analysis.

  • You want to use an A P I Oh, you want to spend 100,000 dollars training one up or something like that?

  • Right.

  • Time for the lost one.

  • Who image to image?

  • You find this on the air, Jess rocks Website.

  • Let's just take a look at.

  • Actually, this is running in the browser, Okay?

  • Outline of a cat and it draws the rest of the my guy called zate.

  • You still university?

  • I wasn't doing this stuff.

  • University was the stuffing out of us?

  • Um, yeah.

  • Amazing work, right?

  • How's it work is an allegorical picks.

  • Two picks.

  • It was an example of something called a gam.

  • Ah, a generative, adversarial neural network on what they How they work is this tune.

  • Neural networks there are competing with each other.

  • Okay, first off, you have a generator.

  • Okay, Have another one called a discriminator of the generators.

  • Job is given some imports.

  • Whatever input there is in this case, it's an outlines of images of outlines of cats.

  • Given that import, its job is to generate on image of a cat.

  • That's his job, right?

  • Um and that's what you do.

  • You get in a whole set of import outline images, you get a generator, generate output.

  • It's not gonna do a really good job right now cause it's all math at random, right?

  • It's not.

  • It's all gonna be noise to begin with.

  • But whatever you take all those images and you combine them with pictures of real cats, jump it all together and you pass into discrimination.

  • Discriminating job is to figure out if an image is of a real cat or generated cats.

  • Remember, the discriminator has also bean initialized with math.

  • But random sze it's not going to do a very good job was always that right?

  • Probably might get it wrong the first time around.

  • They might think the rial cat is a fake cat, But then what do you do when it wasn't a neural network is wrong.

  • We train.

  • I'll take it, training you that back propagation thing.

  • We tweak it with chewing it retune until it gets better.

  • Okay, but if it got it right detected right, that means the generators not doing a very good job.

  • Generators wrong retuned the generator, and that's what's going on as a generator gets better.

  • The discriminating is better as a generator gets better, discriminate on the fighting.

  • Worse is each other.

  • What are you doing?

  • Until eventually the generator is doing such a good job generating.

  • Can images the discriminate Akane Figure out what's real and what's fake anymore, right?

  • And then, you know you've done Your training's finished.

  • What we do is we throw away the discriminate.

  • We don't need any more.

  • We need a generator and you can export that.

  • If you're careful, you can export it into Ah, small enough size.

  • You can load up in the browser, and that's what they did.

  • And then you can give your own outlines and it can generate images of cats.

  • Okay, that's a GAM.

  • But you don't just have to use their with cats works or anything, right?

  • If you trained with right data set by this you left is the import on the right is the output.

  • Okay, but this isn't picks.

  • Two picks, this bit of it.

  • This doesn't work in Java script yet that maybe not so far away.

  • How about this example?

  • One input multiple outputs.

  • Nyro silent.

  • Remember?

  • It doesn't.

  • It doesn't just have to take out lines as imports you can give a gan whatever you want as an import.

  • Whatever you want, right.

  • In this case, you can give it what's called a segmented image, which is just an outline.

  • The different colors kind of give you a clue as the three D nature of the images of the depth of the image.

  • And you could do something like this.

  • Okay, She's not dancing.

  • I wish I could get real life one of one of these for me when I'm in a club, because I can't.

  • Well, you don't.

  • Uh, yeah.

  • How about this one?

  • This a segmented street image top left.

  • These two are just different algorithms, and that's the victim.

  • Did well in there again.

  • It's generated completely generated and computer, but you don't have to even give it even image.

  • You don't even have to give an image.

  • You could just give it text as input remembers.

  • Just whatever text you give, it's gonna generate something.

  • So you give it some Texas import and you ask the generator generate an image.

  • That's what you doing.

  • These are two examples there.

  • So the top one.

  • Ah, the import Texas.

  • This flower has long, thin yellow petals on a lot of yellow panthers in the center stage One is after 600 iterations.

  • Stage two that the 1200 iterations you could see.

  • Well, this one's not very good, but the rest of pretty good right?

  • I miss another one using birds.

  • So this bird is white, black and brown in color with a brown beak.

  • Again, stage one is ah, 600.

  • Not the best.

  • Stage two is very good.

  • He is one who is not very good.

  • Maybe No.

  • Um, so that's it.

  • Just a summarize.

  • Yeah.

  • Ganz learn to generate new images they generate.

  • They generate things.

  • That's what's so exciting about them to cover the creative side off machine learning, right?

  • They generate things.

  • Discriminate er's thirst for used, mostly used machine learning.

  • Just the tellers what something is or isn't right.

  • But generators could generate create.

  • That's what's exciting about, um oh, but they take a lot of computer train a lot more than a normal network.

  • But ah, generates could be exported and run in the browser.

  • I think I'm over time.

  • Um, but what if you want to let a little bit more about tense for Jess?

  • I am considering writing a book.

  • So we take a screenshot of this If I get 100 sign ups, I think I'm about 90 right now.

  • Essentially 10.

  • I will start Kickstarter or something Thio receive.

  • See?

  • Fun writing it if you want a little bit more about machine learning, especially going to use the AP eyes, I have ah, module.

  • If you go there, such a toile step by step will show you how to build something that the motive fire like the slack box Gonna learn more kind of simpler ap I root and go that way.

  • Andi.

  • Thank you.

  • Mystery time like you.

so welcome to talk on the future of Java.

Subtitles and vocabulary

Click the word to look it up Click the word to find further inforamtion about it