Proving you must crawl before you walk or run, AI Warehouse tasked a group of five AI agents with completing a 100-meter dash. Each was trained using Deep Reinforcement Learning to reward them for correct behaviors and punish them for incorrect ones. Eventually, they get their boxy butts off the line and wobble their way through the course.
Awesome Machine Learning
After a failed attempt to do a backflip in real life, animator Nikita Diakur thought it would be safer to have a digital stand-in perform the maneuver. With the help of engineer Maximilian Schneider and machine learning tech based on the paper Deep Mimic, he attempted to nudge his avatar in the right direction to achieve his gymnastic goal.
Machine learning technology has come a long way in recent years and continues to get better. In this video from AI Warehouse, they show how an artificial intelligence named Albert goes from flopping along on the floor to walking as it learns to coordinate its joints via positive and negative reinforcement.
Spotify’s Basic Pitch is free-to-use, Open Source software that uses machine learning tech to convert tunes you hum, sing, or play on a single instrument into a MIDI file that can be used to control digital synthesizers. You can download and run the software locally, or try it out right now on their demo website.
Tools like DALL·E 2 have proven it’s possible for AI tech to create art based on text. Neural Synesthesia fed text descriptions of the history of the Earth and the evolution of its species into StableDiffusion, which it used as a guide to creating the video Voyage Through Time. The music is Order from Chaos by Max Cooper.
Art meets science in Vadim Epstein’s trippy digital compositions. Ghosts is an experimental short film that explores complexity emergence from AI feedback loops. It makes use of the StarGAN2 neural network to morph between machine-generated images. Sound by Alexander Kopeikin and Fractal Heads.
Machine learning technology can produce some fascinating results when asked to create art. TikTok user Recursive Identity used AI tech to create this trippy-as-hell generative art using the work of artist Edward Hopper as its data source. As we fly deeper and deeper into the painting, the images seem to turn in on themselves.
(Gore) Forget everything you know about the Teletubbies. Wizardhead turned these kid-friendly characters into nightmare fuel. Using AI tech, the Wiz remodeled Dipsy, Tinky-Winky, Laa-Laa, and Po into the kind of gruesome monsters you’d blast playing DOOM. They also worked their magic on You Can’t Do That on Television.
Researchers at MIT CSAIL have been working on improvements to its Mini Cheetah quadruped robot, among them a model-free learning that helps it figure out how to traverse varied terrain through trial and error. Along the way, the little guy managed to set a new speed record for its platform – 3.9 m/s (~8.7 mph).
This amazing art generator uses AI tech to create original images based on a phrase. Simply enter a description of your desired artwork, choose a style, and click “create.” In less than a minute, it produces unique works of art that you can download, share, or print. There’s an NFT minting feature in the works.
We previously featured Noah Kalina’s photographic memoir, composed of a selfie a day since January 2000. His previous videos had a staccato look to them, but with AI tech and the help of data scientist Michael Notter, this new video smoothly transitions his aging process over the course of 20+ years.
Brickit is an ingenious smartphone app that uses machine learning tech to analyze a photo of random LEGO bricks, count and identify them, then provide ideas of things you can build with your available inventory. it’s available for iOS now, with an Android version on the later in 2021.
Machine learning technology continues to get more and more impressive – especially when it comes to working with images. A group of researchers from China are showing off DeepFaceDrawing, an amazing piece of software which can synthesize photorealistic human faces using nothing more than a rough pencil sketch.
Max and Dave Fleischer’s 1940s Superman cartoons are beloved. But the quality of the public domain videos online isn’t very good. Jose Argumedo ran the episode “The Bulleteers” through Waifu2x, a neural network upscaling application that recovered an incredible amount of detail, especially compared to the source video.
Machine learning tech is getting pretty good at things like object recognition, but can it write a decent song? Funk Turkey fed artificial intelligence software a bunch of AC/DC lyrics and asked it to create a new track. This is what it came up with. We’re pretty sure this AI is actually a 10-year-old boy.
Ever wonder what a Rick Astley song written by an AI might sound like? YouTuber Lil’Alien demonstrates what neural network called Jukebox came up with when asked to create more of Never Gonna Give You Up based on the data fed into its electronic brain. Play with more weird AI music on the Jukebox Sample Explorer.
With people wearing all kinds of improvised protective gear these days, an outing to the grocery store can be a strange and surreal experience. But Pouff’s trippy grocery shopping video was created back in 2015, using neural network technology which attempted to identify animal faces in places where they didn’t actually exist.
After wowing us with upscaled vintage train footage, Denis Shiryaev takes us on a tour of New York City, circa 1911. Using neural network tech, he injected extra frames to smooth out the choppy movements from this century-old slice of life film. He also boosted its resolution to 4k. We don’t care for the colorization though.
(PG-13: Language) While true artificial intelligence is a fascinating concept, most machine learning tech still uses some kind of algorithmic decision making. Ordinary Things provides a layperson’s explanation of how these systems work, and how our reliance on algorithms could make us stupider, and take our jobs in the process.
We’ve seen machines that can sort LEGO bricks before, but they’re generally limited to just a few specific shapes or colors. Daniel West’s machine is much smarter, using AI algorithms to identify and sort nearly 3,000 different LEGO shapes and colors. We think it’ll need more than 18 sorting bins to be really useful though.
Comedian Keaton Patti claims to have fed an AI system 1000 hours of footage from Batman movies (we didn’t know there were that many), and then let its tech produce a new script based on what it learned. Nerd Odyssey posted the very silly result of his efforts, with animation by C4DNerd.
Scientists from Samsung’s Moscow-based AI Research Center recently showed off “Few-Shot Adversarial Learning” tech, which can generate talking head videos from just a handful of still images, and a source video of another head. It’s so impressive, they even made the Mona Lisa talk.
Experimental band Hardcore Anal Hydrogen created a trippy and vibrant music video for their thrash metal track Jean-Pierre, created with the help of artificial intelligence tools like Deep Dream, Neutral Style Transfer, and DeepFlow. Read more about the project here.