“Ripley’s Believe or Not” features some of the weirdest and most bizarre facts from around the world. How about creating our own Ripley’s Hall of Fame for Data Science applications?
Think about it – what’s the first thought that comes to your mind when you think about Data Science? It’s usually all about techniques, algorithms, programming, among other things. Here’s what I’m proposing – how about we explore the road less traveled?
Yes – I’m talking about embracing the weird and fun part of data science!
In this weird and wacky world, our data science work reflects surprising connections, such as – if you are buying diapers then you are most likely to buy beer. Or, people who go to bars are a higher credit risk! Oh, I cannot miss out on this one – Smart people prefer curly fries. Liking “Curly Fries” on Facebook is a decent predictor of high intelligence!
So in this article, let’s take this roller coaster ride into the fun-world of data science and its weird and hilarious applications. And before we begin, here is a word of caution – this article might tickle your funny bone!
A new Harry Potter piece is hitting the book stores soon:
No, that’s not another J.K. Rowling book. It’s a Natural Language Processing (NLP)-generated Harry Potter chapter! Check out this weird line generated entirely by the model:
“Harry tore his eyes from his head and threw them into the forest. Voldemort raised his eyebrows at Harry, who could not see anything at the moment.”
This was generated by Botnik Studio, an open community of writers, artists, and producers who generate strange content using Data Science.
They collaboratively fed the text from all the seven volumes of this mesmerizing wizarding saga into a predictive algorithm. In just a fraction of minutes, the algorithm detected the pattern of words paired throughout the series. Interfaced in the form of predictive keyboards, the algorithm started making suggestions based on the pattern.
Two different predictive keyboards were used for this task — one for narration and the other for dialogues. And the sentences framed turned out to be pretty hilarious and outright crazy! Here is a relief for all those who lamented the death of Albus Dumbledore in Harry Potter And The Half-Blood Prince- the NLP model has resurrected him. 🙂
In August last year, a fan of the Game Of Thrones series had “written” five AI-generated chapters using recurring neural network (RNN). While the fan had managed to capture George RR Martin’s style of writing, the results were again in the realm of absurd.
What about getting one of your own stories written by a data science model?
So here it is – just try and build your own story by feeding articles about your favorite topic/series into an algorithmic text emulator, give it some input about your choice of words, and voila: you will have the lead paragraph for your story.
See, data science is entertaining us in the most bizarre way. Hopefully, it doesn’t make Harry the villain of this series going forward!
I love the creativity involved in this application.
This is what happens when we mix the world’s most advanced face-generating AI with an easy-to-use website – it gives you faces of people who don’t exist at all!
This website is created by silicon valley software engineer Philip Wang. Each time you visit the site, it generates a unique face – a photo of a human dreamed up by the computer code.
It’s a simple website with no explanation, no code, no FAQ, and no branding. It just reflects a face that has never existed before (at that moment). Keep refreshing and it will keep generating fresh faces to infinity “from a 512-dimensional vector”!
What’s the computer vision concept powering this model, you ask? It’s Generative Adversarial Networks (GANs)!
“Recently, a talented group of researchers at Nvidia released the a state of the art generative adversarial network, StyleGAN, so I decided to dig into my own pockets and raise some public awareness for this technology.” – Philip Wang
On a good note, Wang did not create thispersondoesnotexist.com to enable abuse of any kind. He created it because he says he’s “very worried” about our future, and he wanted to create awareness of how far technology has advanced so quickly.
But you must be wondering how GANs actually work. Here’s a brief explanation:
Broadly, the GANs training phase has two main subparts that work sequentially:
- Phase 1: Train discriminator and freeze generator (freezing means setting training as false. The network does only forward pass and no backpropagation is applied):
- Phase 2: Train generator and freeze discriminator:
If you want to dig deep into it then here are the steps to train your own GAN model. Pull up your socks and get ready to generate something as cool as this:
Phew! GANs are quite exhaustive and fun!
GAN was introduced in 2014 by Ian Goodfellow but it wasn’t until 2017 that researchers were able to create high-quality, 1024 x 1024 images detailed in the now-famous ProGAN paper.
I have been refreshing the site constantly – it’s super fun to see all sorts of new people!
3. Look who’s making us Dazzle with Ultra Stylish Clothes – AI Of course!
I never thought I would witness something like this – a machine learning model designing clothes much, much better than designers! The reaction of people, when asked to compare an AI designed dress against a human-designed dress was epic! Check it out:
Bizzare and amazing, this touches both cords for me as a fashion enthusiast. Imagine its applications – it will solve all the worries of deciding what to wear in the morning! A data science model will read my mind as I wake up and get my clothes ready according to my mood.
Talking about the fashion show, this data science use case has been created by Intelistyle – an AI styling service company. Typically, computer vision techniques that focus on Pattern recognition are used to find visually similar clothes but their approach was a little different.
Their main focus was on building a system that can recognize ‘style’. It is a lot more nuanced than finding visually similar clothes. So, they used deep learning to ‘extract the essence of style’. Of course, this step was executed after they successfully crawled the web for fashion photography examining thousands of outfits put together by influencers and stylists, designers and retailers.
Intelistyle even tested it against real stylists and fashion influencers at London Fashion Week. As Forbes reported, 70% of respondents unwittingly chose the looks created by their model.
Image Credits – Intelistyle
Here’s another fun data science application in fashion I spotted on Twitter. The AI, created by artist Robbie Barrat, has generated an entire collection based on Balenciaga’s previous styles. The fashion week came early on Twitter, courtesy of a clever neural network.
• Neat asymmetrical multi-component coat (one arm hidden… probably unintentionally though)
• Three-component coat with matching piece of cloth to carry around (posted this one before)
• Patterned sweater (shirt ?) with what looks like a two-tone turtleneck underneath (?) pic.twitter.com/jHsgIt4OZo — Robbie Barrat (@DrBeef_)
You can also check out this paper, Generating High-Resolution Fashion Model Images Wearing Custom Outfits, for more information on creating AI-generated Fashion Trends.
4. Who’s killing the Academy Awards Game? Let’s have Data Science predicted Nominations Please!
Who is your pick for the best actor and best actress at this year Oscar’s?
Quite often we can guess the eventual winner by combining our intuition with the power of the media rumor mill. But there’s a major issue there – our picks can be heavily biased based on our own preferences and watching history.
Where there is a guessing game – there is data science!. You might think it’s weird but yes, AI had an impressive 94 percent score in predicting last year’s Oscars, for the third year in a row, you will be thrilled to know that AI outperformed industry experts.
The Hollywood Reporter and Microsoft teamed up to predict who would take home an Oscar at the 2019 Academy Awards through an online prediction poll called Awards Predictor (powered by Microsoft AI).
Unanimous A.I., a company that uses “Swarm A.I.” technology to create artificial intelligence products, comprises of a hive-mind of dozens of movie enthusiasts.
This TED Talk by Unanimous A.I. founder Louis Rosenberg, Ph.D., goes into further detail about this fascinating application of technology and the human brain.
These are the Swarm AI predictions for the 2019 Oscars:
Swarm AI combines real-time human input with artificial intelligence algorithms, optimizing a group’s combined wisdom and intuition into a unified output. In other words, Unanimous builds artificial “hive minds” that amplify the intelligence of human populations to create an artificial super-expert that can outperform traditional experts.
How can you build this exciting predictive machine learning model?
Well, it’s simple – combine OptiML, the optimization process on BigML that automatically finds the best-supervised model (and runs some top-performing models including deepnets, ensembles, logistic regression, and decision trees) with Fusions, that combines multiple supervised models for improved performance and make a batch prediction – and boom – your model will be ready!
5. Let’s Have a Good Laugh with AI-Powered ‘LOL BOT’
Now we have a robot who can make us laugh – literally! By far I found this to be the most unusual application of data science because I would never expect a bot to make me laugh, that too on the topic of my choice.
Meet – LOL BOT. The rib-tickling bot made its debut at the Melbourne International Comedy Festival and has been savoring fandom ever since then. It’s the world’s first bot that is capable of generating its own jokes, detecting real-time human reactions, and reacting accordingly.
Creative technologist Steven Nicholson said his team had developed LOL-Bot to use stochastic modelling to pull data and learn from thousands of hours of live comedy shows.
Stochastic modeling enables LOL BOT to pull data from the enormous library of jokes from all the comedians around the world. It uses deep learning algorithms to tease out the meaning from those shows and generate its own jokes. Here is a glimpse:
The idea of using data science to have some fun is really fascinating. My aim in this article was to open up a new portal for all of you who think of data science as a technical and all-too-serious field. Let’s have a little fun on the way!
Some of these weird use cases might come across as light-hearted fun but remember – big advancements in any field are made by people who have played around with what they were given and churned something no one could think of.
How lucky we are, without any heavy investments, that we get to freely experiment. And I firmly believe that by using this power, our champion data science community will take the world by storm.
What other weird and hilarious uses of data science can you think of? No idea is a bad idea! Let me know in the comments section below.
Are you new to the world of data science? I highly recommend checking out our comprehensive and popular beginner to advanced course: