03 Feb 2016
3 extremely cool uses for machine learning
Machine learning plays a big role in our lives today, and this role is growing. Fast.
When it comes to machine learning, and its potential to change the way the world works, the sky is the limit.
We are currently living in the beginning phases of a machine learning / data mining revolution which will, sooner or later, change EVERYTHING.
Exactly which areas of our lives will be effected is a subject limited only by our imaginations.
One of the most interesting realms into which machine learning has been introduced is the creative field, commonly referred to as “the arts”. Machines have slowly begun, with close human supervision, to move into creative endeavours once thought to be the exclusive domain of the wonderful human mind.
We prefer to think that creating something wholly new, something beautiful, like a painting, a song, a sculpture, or a poem, which is capable of impacting us on an emotional level, is something that a machine could never do. Machines don’t have a soul, don’t have a heart. How could they ever “create”?
But in reality, the machines might be closer to achieving just that than we might like to think.
Let’s look at a few examples, and keep in mind that this field of study is still in its very infancy.
1) Script writing:
Comic Artist Andy Herd fed the scripts for every episode of 90’s mega-hit sitcom “Friends” into a recurrent neural network and asked it to generate new scenes based on the patterns it could mine from the old ones.
Image Credit: Warner Bros
Because there were 236 episodes of Friends made over the show’s 10 years on the air, the sample size was good. The show’s small cast and limited plot lines helped as well, since it revolved almost exclusively around six people talking to each other, in their apartments.
Image credit: Warner Bros
Well, here are two examples:
(Scene: Chandler and Joey’s, Joey is waiting for the door, Ross is looking for Joey and Chandler’s stuck on the couch and screaming)
Chandler: Well, I proposed to my shoe…
Joey: (laughs) This is his father… (cave children) Hey Pheebs?
Monica: I hate men! I hate men!
Ross: What are you gonna do?
(All the dinner enters)
Monica: Happy Gandolf
(Pause for a beautiful women walk into his apartment and she keeps glaring at Joey’s jacket)
Monica: Okay, I’m going to Minsk
Rachel: Yeah, sure.
Granted, this is pretty much gibberish, but all the elements are there. It may seem quite random, but it isn’t. How long do you reckon it would’ve taken a thousand monkeys at a thousand typewriters to get that far?
Image credit: 20th Century Fox
Whatever the case, the machine got just a bit closer in its next attempt:
(Scene: Monica and Chandler’s)
Joey: (entering, curling voice) Rach, enough, I’m so sorry
Rachel: What kind of people do these thing?
Rachel: (to Joey) I gave you my wife in soap opera. (pause) Uh! (snaps off her jacket) I have a baby!
Monica: Oh my God!
Phoebe: Maybe you need a pizza.
Chandler: (laughs) You got him baby Geller? (Rachel shouts something)
I guess the Screen Writer’s Guild can breathe a sigh of relief. For the moment, it appears that their jobs are safe.
The same guy (Heard) also began producing a comic strip, generated completely by computer. There isn’t really any machine learning in it, since the results are almost entirely random, but it is still worth a look.
Its called “Pandyland”, and it is awesome
see more here…
But we digress…
2) Rap lyrics
The variables for writing a sitcom script are pretty extreme, but what about something with a bit more structure, like, say, rap lyrics?
Eric Malmi at the University of Aalto in Finland wondered the same thing. Malmi has no relation to our CEO Parry Malm, but Parry did write a very interesting interpretation of the technology here.
He and a few friends used a database of over 10,000 songs from more than 100 rap artists to build an algorithm that analyzes and generates rap lyrics.
All they had to do was give the machine a starting point, like they did for this rhyme, by asking it to write a 16-line rap on the topic of “love”:
For a chance at romance I would love to enhance
But everything I love has turned to a tedious task
One day we gonna have to leave our love in the past
I love my fans but no one ever puts a grasp
I love you momma I love my momma – I love you momma
And I would love to have a thing like you on my team you take care
I love it when it’s sunny Sonny girl you could be my Cher
I’m in a love affair I can’t share it ain’t fair
Haha I’m just playin’ ladies you know I love you.
I know my love is true and I know you love me too
Girl I’m down for whatever cause my love is true
This one goes to my man old dirty one love we be swigging brew
My brother I love you Be encouraged man And just know
When you done let me know cause my love make you be like WHOA
If I can’t do it for the love then do it I won’t
All I know is I love you too much to walk away though
If we compare that to some songs that have been mega hits (looking at you, Daft Punk’s “Around the world”), which looks more like it was generated by a machine to you?
Still not impressed? Check this out.
Image credit: PBS
Researchers at the University of Tubingen in Germany developed a deep artificial neural network which mimics the way a brain finds patterns in objects. They taught their learning machine to record how different famous artists used colors, shapes, lines, and brush strokes, so that it could view the world in the same way they did.
In the image below, the system was asked to recreate an image of a row of houses overlooking the Neckar River in Tubingen, in a style matching one of each artist’s most famous paintings.
Image credit: University of Tubingen
The program’s attempts to match the artists’ styles is pretty impressive.
However, the level of supervision was still quite high, since the machine produced several versions of the image, varying how much the image should be changed to mimic the style of the artist represented as a percentage, and the scientists monitoring the experiment had to be the ones to choose the prime percentage of change to the image.
This is because the program is not yet capable of identifying which image looks the best; that takes subjective reasoning.
But how long until an algorithm is developed which can mimic subjective reasoning?
Realistically, it’s just a matter of time.