Rabbits, Faces & Hyperspaces - Computerphile

[@ToddHowardWithAGun]

Robert Miles is one of my favorite computerphile people. All of his talks are interesting.

[@kyberuserid]

All? Were you born yesterday?

[@john_hunter_]

Now I want to see the program that can make the caricatures.

[@AlexSimpsonsPage]

I wrote one many (many!) years ago as a child in BASIC that ran on the 8-bit ZX Spectrum, just with line art. I was inspired when I read about typical measurements of the human face at the time. It was quite interesting to work with and if you measured your face it would caricature it, in as much as you can with very basic line art! e.g. enlarge or shrink the nose, ears, eye size, spacing between them. There's a paper on a more modern technique, morphing photos in to caricatures here: graphics.cs.cmu.edu/courses/15-463/2004_fall/www/handins/brh/final/

[@uusijani]

Now I want to see those caricatures!

[@lednerg]

There is a program called FaceGen that does exactly what he's talking about here. You first create a face based off of a photograph or from scratch. Then there are a series of slider controls to adjust all sorts of things such as how 'male' or 'female' you want the face to look, as well as making exaggerated caricatures. There is a free version on the site. www.facegen.com/

[@kagitsune]

lednerg, Monster Factory. 😈😂

[@ddjanji]

this gotta be my favorite series so far on computerphile, please keep them coming. I love the topic and the guy is just brilliant at explaining it (:

[@Heschoscho]

"We can do it with Maths" should be the computerfile slogan

[@thenorup]

If things are worth doing, they are worth doing with maths!

[@lindhe]

Or "We can do maths on it".

[@SendyTheEndless]

Warning: maths in progress

[@aednil]

thenorup really? what about intercourse? :-P

[@FreeScience]

aednil Cross product?

[@mushkamusic]

Robert seems to be a wonderfully clear thinker. Great vid thanks.

[@anaraug]

I loved the part where the characters talked about this in Anathem. Explaining orbital coordinates as a type of configuration space. And then later, describing the entire universe as a giant configuration space that any possible universe is a path through.

[@dmhouse1024]

Just to emphasise an interesting property of these algorithms which wasn't explicitly touched upon in the video: At no point do we have to say "male faces are like this". We don't explicitly describe that male faces have strong jawlines, or large noses, or whatever. We simply come up with a reasonably long list of ways in which faces can differ, and then we throw a bunch of real-life data at it -- "training" the algorithm if you like. Once we've done the training, we've established where the average male/female faces are, and we're ready to do the real challenges, like classifying a new ("out of sample") face, or doing the extrapolation, or whatever. Also, the more real life data you throw at it, the better an average you will get, since it'll be less likely that all of the faces in your training data had wide set eyes, or something else idiosyncratic. This fits the training metaphor very nicely -- the more you train your algorithm (= the more real life data you prime it with), the better it is! Even nicer than that, you don't even have to worry about measuring too *many* things. Let's say that you think the radius of the pupil might be significant in establishing gender, but that you're actually wrong. That means that average male pupil radius is about the same as the average female pupil radius. So if you travel along the line from the average male face to the average female face, you won't vary much in the "pupil radius" dimension. Hence, if you pick any face F (point in the face space) and try to find the point P on this line which is closest to F, F's score in the "pupil radius" doesn't matter very much -- i.e. your algorithm is correct ignoring pupil radius when classifying gender.

[@longleaf0]

Really enjoy Robert Miles' way of presenting concepts, hope we see lots more of him :)

[@TravisPattersonPhD]

First off, great video! Clustering appears to utilize statistical analysis, such as correlation, mean, standard deviation and multivariate Gaussian distribution. Based on the plotted data, applying a correlation analysis will identify the relationships of the data based on the parameters. Second, it appears that by finding the mean on each side of the 'correlation line', each side can be broken down as its own data sets. By doing so, this allows for both means to have a calculated standard deviation. Typically, in statistics, if a data point is within 2 standard deviations of the data set, then is consider within the normal distribution. Anything outside of two standard deviations would be seen as extreme. Thus, by taking the basics of applied statistical analysis and performing multiple functions within the data, this type of algorithm could be used to catch movement within data set as whole, deviations among subsets and additional random variables that indicate a cluster around a mean. By adding the 3rd dimension, the multivariate Gaussian distribution, referes to the k-value. By taking the 2 dimensional analysis and identify all data correlations on the third plane, representing the 'density' per se, allows for the multidimensional analysis of plotting a single data set, identifying the possible correlations of each variable and proposing the density weight on the other dimension. So by inferring that the initial data set is correlated based on random variables, the K-value will inherently add the additional dimension and offer a deviation based on all three planes. This is my simplified hypothesis of the application of statistical analysis to reach to the 3-sigma eclipse. Beyond this, many other statistical algorithms have been developed and applied in this same field or manner. Bayesian statistics provides the predictive power of a population within the set of parameters. Non-parametric methods will still provide the intervals and future probabilities, but there is a chance the a value may fall out of the 'range'. Although additional models can account for such outliers, this may also provide a small peak into why clustering consumer variables (observed) to predictive variables (recommendations of products or movies), that roughly higher than 50% may apply to the consumer. Again, I know this is an extremely simplified hypothesis of the application of statistical analysis based on the video, but the visual depiction and explanation in the video were terrific in simplifying the process without having dive into the varying complexities behind the statistical algorithms involved to make consumer predictions and adding additional dimensions. There is a 'probability' my stated inference or hypothesis is way off, since I have not applied statistics in this manner of application. Based on my experience and the statistical formulas running through my mind while watching the video, I was hoping to provide a simplified notion of my perception. So, I wanted to reach out and see if this is in the ballpark or if there are non-statistical functions that are applied which reduce the complexities, limits and other issues in statistics. Thanks! Love the new channel!

[@vishva8kumara]

We did this with 'words' of natural language as a research in SLIIT. One outcome was a software that can research the internet and write essays. Another was a search engine algorithm. Also, we could use that to improve a translation (English to Sinhalese) engine.

[@aglees2b]

One of the best of the last few months I reckon

[@Nilguiri]

3:58 Come on, Brady/Sean! Three dimensional paper, please...

[@mdalaminmahamud3775]

I would like to know more about the guy "Robert Miles". He is an excellent presenter. Moreover thanks to ComputerPhile for these excellent video

[@TheUnchainedMind]

I like this guy. Moar.

[@jg-reis]

Best Computerphile video I’ve seen so far (OK, it’s a small series, I’m a recent convert).

[@Faxter313]

Coooool stuff! Watching videos like these make me always feel like I haven't learned anything until now. Amazing. :D

[@gonigeena]

this guy is the ultimate explainer

[@Birkirrey]

This video takes on a whole new form after studying data mining. One of my favorites

[@LordMarcus]

Male/female variances work fairly well for phenotypes of many creatures, like birds. Your example works fine.

[@stealth9799]

Idk why you're assuming amazon uses any good system at all. Christmas two years ago I bought princess slippers for my niece and I'm still getting recommended products for little girls

[@FishKungfu]

Very cool! Robert Miles does a great job in these videos.

[@TiagoSeiler]

At the end of that line that goes off in the "more male" face is a picture of Chuck Norris.

[@Maric18]

+Tiago Seiler after a while they stop looking human

[@irok1]

@@pixelchai correct

[@Giantcrabz]

Or Henry Caville

[@naustradmas]

This is the most interesting video i have ever seen on youtube.

[@史昊珩-q5j]

Can we get playlists for the regulars? It'd be great if there was a Robert Miles playlist.

[@RobertMilesAI]

I have my own channel, which has a playlist

[@Kobriks1]

This guy is totally awesome more vids with him please!

[@karlkastor]

Awesome video! Very interesting.

[@quantummenace7203]

Are there photos of these "anime characters" ? Please share the grotesqueness !

[@aboveallholidaylighting9305]

Thanks for sharing. Neat to finally get some insight into how a computer thinks

[@thebookofjoy]

Robert Miles. Kicking ass in Computer Science in his free time when he's not making hit dreamhouse tracks

[@Soldier842]

This was really interesting&fascinating!

[@Pawn87]

I like this guy, you should get him on Computerphile more!

[@simoncarlile5190]

About extending the line to create more masculine or feminine faces than actually exist: how far can you take the line? Does it extend to infinity? If it does, then what do the faces tend to look like as you go further along (do they tend towards some limit or do they keep changing drastically forever)? If the line doesn't extend to infinity, what two faces are at the opposite ends of it?

[@michaelstueben2880]

Important mathematical note: Be careful with using the word "vector". Why? A vector's scalars must all be of the same units (or all of no units), or else the magnitude of the vector cannot exist. We cannot add square heights to square weights, because you can't add numbers with different units.

[@kagitsune]

I'm tracing back the rabbit hole from the "AI stop button problem" video and I've landed here.

[@Felixkeeg]

6:14 I love how the camera zooms in as soon as he makes a joke^^

[@BluephoenixX]

That small red rabbit looks very familiar, I have one sat on my desk from Electromagnetic Field.

[@wickedtiger1553]

At 6:15 I started wondering. Rabbits didn't look like the same in the past, so as human faces changed a lot (including the increasing life expectancy as well) so it might happen that the average the machine counted is going to be out dated in the future. Can an intelligent machine monitor slight changes and trends in such a sense?

[@OJungell]

Fantastic video!

[@jdgrahamo]

Fascinating about the caricatures.

[@welsh_cake]

If you ever feel alone just remember that Computer Science disagrees; you just haven't spotted the other members in your cluster yet. :)

[@asphaltpilgrim]

Loving the pencil drop to finish, "... but now we can do it *with maths*!" XD

[@tnghunter]

It would be interesting to see a video explaining web crawlers and search engines!

[@TheBandScanner]

This was a good discoarse about the topic. But consider what the computer does everytime a person orders or looks at something recommended by a chance friend. For the next three months you get suggestion ads that just don't apply.

[@nicolaserriquenz9905]

The video that plays at the very end (third cuadrant) its an extension of this video? Or is there another video of him explaining how to math a cartoon character's face?

[@lordecircojeca2039]

6:45 is this how Gigachad was born?

[@nand3kudasai]

Nice, there are engines for training a model like that with python. Also i think Google has a freemium tool for that too. I wish this guy talks more about AI.

[@Cygnus0lor]

Give him a mic to drop at that ending...

[@Zerepzerreitug]

amazing!

[@ElectronSpark]

So, the software can make male or female exaggerated caricatures? Why would you not show an example then? ???

[@ferencgazdag1406]

Probably copyrighted. The entire algorithm is copyrighted.

[@egokick]

Great vid!

[@mart3323]

How are the dimensions normalized? as in, how do you pick the units For example, on a space of mass vs gender.., how large of a mass difference is equivalent in distance to being of the opposite gender? Or mass vs color..., If person A is 50 shades bluer and 10 kg heavier than me and person B is 25 shades bluer and 15kg heavier then which is closer? Is there some math to automate that as well or is that done manually?

[@BicyclesMayUseFullLane]

I lost it at 'anime characters'.

[@petros_adamopoulos]

A "phase space", oh I see what you did there.

[@matheusdardenne]

A face phase space :P... It's sad how few people will get the joke, tho... D:

[@lightsidemaster]

So what's a hyperspace now? Simply a higher-dimensional space with many different attributes? I expected something way more complicated honestly.

[@MattiasBuelens]

Nope, that's literally all there is to it: an n-dimensional space where n > 3.

[@lightsidemaster]

***** Interesting... thanks for clearing that up.

[@EcceJack]

***** sci-fi and hyperspace having been mentioned: technically what is meant by "travelling through hyperspace" in that sense is travelling along some 4th spatial dimension in such a way as to make the journey shorter than it would be by connecting the same two points in 3-D (think a hole through the Earth vs. a flight from London to Sydney, for example) [Edit: so that goes nicely with what Mattias said before me]

[@lightsidemaster]

EcceQuomodoMoritur00 Thanks for the comment. I know more or less how wormholes, hyperspace, warp drive etc. work. Or are supposed to work. I'm more into physics than mathematics so I just didn't know anything about the mathematical hyperspace.

[@schok51]

Well, yeah. A Cube is tridimensional. A hypercube is 4-dimensions or more.

[@UstedTubo187]

I was hoping you guys were going to talk about the sci-fi version of hyperspace. :(

[@robl39]

is saying "maths" (with the s) a British thing?

[@Computerphile]

Yes, in fact, Numberphile did a whole film on it: kzbin.info/www/bejne/iZO9dHh5q9SUiqM >Sean

[@brandonlewis2599]

It's really fun to tease them about it ;)

[@DannysGalaxyTab]

+Brandon Lewis My American maths teacher had a nightmare teaching at our British school haha

[@circuit10]

Yes I think (I'm from the UK)

[@henrypostulart]

Yes

[@TheiLame]

Can someone tell me if 4:31 IS A RANDOM NUMBER 150, or what is up with 150?? So it doesnt work starting from 151??

[@tarcal87]

***** Are you kidding? Yes a random number

[@himselfe]

Watching this in 2019 is so very different to watching it in 2014.

[@bobwebster835]

6:15 facebook crossed with myspace

[@ataraxia7439]

Is there somewhere you can find the computer generated faces?

[@matheusdardenne]

It's a face phase space =P

[@hasch5756]

He has tamed the beast of Caerbannog! Subscribe to the all-powerful!

[@blazebluebass]

That is a simple way to quantify things.

[@LimeGreenTeknii]

So could you create a caricature of what somebody would buy on Amazon?

[@jcasaubon]

Fascinating thanks!

[@JohnBogle286]

TRIGGERED

[@KetanSingh]

Wish he did a MOOC on Artificial Intelligence.

[@Jookeerr]

I wonder how does such a multidimensional spaces/graphs algorithms look like :D

[@NachiketVartak]

Erm, a 'space' is just the collective name of all possible values a given quantity may take.

[@lindhe]

7:06 "eyes"...

[@justinmcdaniel7419]

In all honesty, if given enough data, the computer would know who and what you are better than anyone and anything.

[@Azur1al]

so if i had amazon and ad company history data i would in principle be able to find everyone in my vicinity who shares my interests......

[@daanverbrugghe8982]

just listen to the speed at wich he talks when he is not drawing and when you seed the paper and he is drawing, there is a diferents so HE TALKS FASTER WHEN HE IS DRAWING of they just sped the video a litlle bit up

[@danwilson5630]

When you use different shapes as plots, doesnt that make it a 3D space?

[@RynRobitske]

Facespace - A space for faces.

[@96Bogg]

He watches anime, more of a reason to

[@NavnikBHSilver]

Always fun to see, that the moment we translate 'abstract' concepts into strong, mathematical values, and then translate it back, we can make a computer calculate almost anything, regardless of how mathematical the problem really is. its no surprise to me when considering this, that they make a TV show, wherein people marry to a by computer selected person.

[@coffeefish]

Like it.

[@coffeefish]

***** I meant that I like it.

[@TheAwesomeTool]

Any good books about AI theory?

[@RVM451]

Is that how Bayesian filters work? .....RVM45

[@thelastinuit]

"I'm a person not an animal" FOR GRAYSKULL!

[@TaberIV]

I want a purple bunny!

[@johnnybravo1041]

2:50 I identify as a point in space

[@deesabird6799]

Lol

[@deesabird6799]

What about faces spaces

[@crumplezone1]

`Spaceface close to mine love` Bowie `Moonage Daydream, Ziggy Stardust and the spiders from Mars

[@azmanabdula]

well we view 2D space from a 3D perspective... but drawing 2D from its own perspective is too easy... (_____) < your world (im getting sidetracked i paused it at the start : S) PS is it wrong to say we see in a somewhat 2.5D style? we dont visually see something in 3D as you cannot see whats behind something...

[@XxJ0nas98xX]

Well, kind of, but not really, as what you perceive as 3D is just your brain comparing two 2D images to get depth information. If you look at a random point with your eyes, what you'll see is 3 characteristics of it, the light-level, the color and the depth. So we have 2D vision with information about where what we see in 2D is placed in 3D. If we had real 3D vision we would be able to see what's behind the object we're looking at.

[@azmanabdula]

XxJ0nas98xX Exactly!

[@cmankows]

why male and female? it's a straight forward, easy example. that's why you chose it. 2 year old kids understand that difference. nature understands the difference. why wouldn't a career academic understand it?

[@azmanabdula]

Space is absolute, the Distance isnt.... "from recollection i cannot remember where i heard that"

[@tallestGirafffe]

This guy is hilarious!

[@lyrimetacurl0]

Facespace and Mybook

[@CharlesVanNoland]

It's a "Face Space"!

[@pjousma]

Almost scary how human interpretation gets digitized.

[@_Atzin]

This guy reminds me of Kyle kulinsky from secular talk

[@UntouchedWagons]

"They end up looking like anime characters" Uguu~~~~

[@y__h]

she's way waay beyond the anime cluster, far far away ugu~

[@maxharrison9918]

Anyone able to link photos of the most male male or most female female?

[@bookslug2919]

Ok, now explain why I got a message from Amazon saying "We see you've bought this book on robotics, so we think you might like this book on chicken breeding" .... .... unless there is a secret cluster of chicken breeding roboticists.

[@tho207]

interesting! data is power.

[@dharma6662013]

Right. Right? Right?!

[@jervey123]

YEA!! DO THAT SHIT WITH MATH!!

[@windowslogo3577]

3:56 LOL :D