We are currently in the middle of an artificial intelligence revolution, with huge advancements in machine and deep learning happening everyday. While much of the technology might go over the average person’s head, there’s no denying that some of the resulting applications are very cool.
We’ve compiled a list of 10 applications which we think are the most fascinating, state-of-the-art examples of what artificial intelligence can do. So get ready to have some fun with these!
1. Samsung uses deep learning to create video from still portraits
Researchers at the Samsung AI Centre have used a special type of neural network called a Generative Adversarial Network (GAN) to map faces (even those in paintings) to moving videos. Given a recording of a person speaking or moving their head, the original image is mapped onto the movements in the recording using facial landmarks. This project distinguishes itself from other research in this area as it is able to achieve shockingly realistic video with only a single image as an input.
2. AI can photoshop a face given a simple text instruction
Researchers from the Hebrew University of Jerusalem, the University of Tel-Aviv, and Adobe Research have developed a tool that can apply photoshop-like transformations to images when given only a few words of instructive text. You can check out this video to see it live in action or, if you’re using Windows, try it out yourself using a graphical interface.
3. Machine learning model reconstructs movies using brain activity
This video and accompanying paper from the University of California show the usage of regression models to decode people’s brain signals whilst watching movies. The video image below right is completely reconstructed from brain activity alone, and says a lot about what we might be able to learn about the inner workings of the brain through machine learning in the future.
4. This person does not exist
Thispersondoesnotexist.com uses a type of generative neural network, trained on millions of examples of human faces, to create brand new, completely fake, images of people that have never existed. The site, created by Uber software engineer Philip Wang, is built using a machine learning model called StyleGan2 developed by Nvidia’s research team. Refresh the page to see a new, never before seen image of someone that doesn’t exist!
5. AI composes pieces in the styles of famous musicians
Out of OpenAI comes MuseNet, a deep neural network trained to create completely novel compositions in the styles of everyone from Lady Gaga to Chopin. If you head over to the demo site, you can play around with combining styles of different artists to create brand new music.
6. 1993 demo of the world’s first Text Recognition Convolutional Neural Network
It’s hard to put into perspective how far machine learning, particularly deep learning, has come in the last three decades. The demonstration that can be seen in this video shows an early AI model reading characters on a piece of paper to a computer screen. While this may seem simplistic given the technology we have around today, the context surrounding the time period and the technology available makes it just as mind-blowing as any of the other entries on this list.
7. AI automatically reads and summarises academic papers
PaperAI is a tool developed to ease the tedious literature review process. Instead of trawling through hundreds of academic papers for key details you can read automatic summaries created by the tool. If you’re only interested in the basics of what a paper proposes and the conclusions that it draws, you can tell PaperAI and it will retrieve that information automatically.
8. Machine learning art project detects when politicians are on their phones during parliament
This art piece by Dries Depoorter uses machine learning and computer vision to tackle themes of surveillance and privacy legislation, by focusing awareness on those in power. The Flemish government streams every parliamentary meeting to their youtube channel, and so Depoorter came up with the idea of using a machine learning model to automatically scan the video feed and find politicians using their phones during parliament.
9. Neural network generates vivid artwork that responds to music
This project, dubbed ‘Lucid Sonic Dreams’, is another on this list to use Generative Adversarial Networks (GANs) to create stunning artwork. This neural network synchronises music to create generative art to stunning effect. Check out this video to see it in action.
10. AI paints photographs in the styles of famous artworks
The below recreation of Stanford University campus, in the style of Van Gogh’s Starry Night, is an example of the type of artwork created by the AI model developed by researchers at the University of Tübingen in Germany. Given a painting, this model is able to paint any photograph in the style of the art piece. For more examples, or to download and use it yourself, check out this GitHub page.
Machine learning and Melt Digital
If these examples have sparked an interest in the world of AI and you want to know how it could be used to help your business, get in touch and have a chat to our resident data science experts.
Machine learning can be really useful in digital marketing, and we’d love to show you how.