What's New In ML #1
Generative AI, robot dog, protein folding, privacy preserving ML dataset, AI-designed circuits, startup funding
Hello readers,
I’m launching a new weekly segment today called What’s New In ML.
The goal is to help you stay updated on what’s happening in the world of ML, AI, and Data. We’ll talk about real-world applications, cutting edge research, startups, and everything else that’s moving this field forward.
If you have a question, submit it here and I’ll get back to you with my thoughts. Subscribe to this newsletter to receive it in your inbox every week:
APPLICATIONS
OpenAI opens up access to its generative AI system to a million people
DALL-E is an AI system developed by OpenAI that creates realistic images from a text description. OpenAI has announced that they'll provide access to DALL-E to over a million people in the coming weeks. You need to be on the waitlist to get an invite. Users can create images using text prompt.
Users will get full usage rights to commercialize the images they create. Users will get 50 free credits during their first month of use and 15 free credits for every month after that. Each credit can be used for one prompt and each prompt generates four images. You can buy additional credits in 115-generation increments (460 images) for $15. Read more.
Midjourney opens up access to its generative AI system
Midjourney is an independent research lab that has built a generative AI system. They've opened up access to their generative AI art tool Midjourney. It allows anyone to sign up for an account and generate images using text input. You need a Discord account to sign up and see it in action. I gave it a shot and the images are pretty cool.
You need to get a hang of what type of text prompt can generate good images. It would be interesting to see how it stacks up against DALL-E. Beta users will receive 25 free images. After that, you'll either have to pay $10/mo for 200 images or $30 per month for unlimited use. Read more.
Robot dog learns to walk in an hour
Researchers at the Max Planck Institute for Intelligent Systems in Stuttgart, Germany have built a robot dog that learnt to walk in an hour. They conducted a research study to find out how animals learn to walk and learn from stumbling. This helped them figure out the mechanics of it. Read more.
New Technology Gives AI Human-Like Eyes
Researchers at the University of Central Florida have built a device for AI that replicates the retina of the eye. Why does it matter? It would results in AI systems that can identify what they see right away e.g. automated descriptions of photos that are captured live. This could be useful for robots and self-driving vehicles. Read more.
Microsoft announces Project AirSim
It's a platform to create, train, and validate autonomous agents through simulation. It's useful for building and testing drones in a 3D simulated environment. Read more.
Nvidia opens up access to its Canvas app to help people create beautiful paintings
The Nvidia Canvas app lets people paint by material rather than color. It uses AI to turn brushstrokes into realistic images. The app displays the photographic result as people paint, so they don’t need to wait to see the form of their vision. They can see it right away. This app is now available as a free beta. It brings the real-time painting tool GauGAN to anyone with an Nvidia RTX GPU. Read more.
Making deep learning models smaller and faster
For context, DeepSpeed is an open source deep learning optimization library for PyTorch. DeepSpeed and Microsoft announce DeepSpeed Compression. It's a composable library that helps make deep learning models smaller and inference speed faster. Read more.
BIO / PHARMA / HEALTHCARE
DeepMind has predicted the structure of almost every protein known to science
DeepMind released and open sourced AlphaFold almost a year ago. It's an AI system to predict the 3D structure of a protein from its 1D amino acid sequence. They created the AlphaFold Protein Structure Database to freely share this scientific knowledge with the world. Today they're releasing predicted structures for nearly all catalogued proteins known to science. This will expand the AlphaFold database to over 200 million structures. This has been done in partnership with EMBL’s European Bioinformatics Institute. Read more.
AI model finds life saving drugs 1000x faster
Our universe possibly contains an infinite number of molecules. But only a fraction of these molecules have potential drug-like characteristics. We can use them to develop useful drug treatments. But even this fraction is a huge number. There are potentially 10^60 molecules that have drug-like characteristics. That's 1 followed by 60 zeros. This slows down the drug development process. To address this, MIT researchers have developed a geometric deep-learning model called EquiBind. It is 1,200 times faster than one of the fastest existing computational molecular docking models. Read more.
RESEARCH
Open source platform enables research on privacy-preserving machine learning
Researchers at University of Michigan have open sourced the largest benchmarking dataset for ML designed with data privacy in mind. This approach is called federated learning. Instead of transferring private data to central servers, it trains ML models on end-user devices such as smartphones and laptops. Read more.
Nvidia shows that AI can learn to design circuits from scratch that are faster and smaller
In this paper, they demonstrate that AI can learn to design circuits from scratch. And these AI-designed circuits are smaller and faster than those designed by state-of-the-art circuit design tools. Their latest Hopper GPU architecture has approx 13,000 instances of these AI-designed circuits. Read more.
STARTUP FUNDING
Diabeloop raises €70M Series C led by LBO France for its AI-powered insulin delivery tech
Theator raises $39.5M Series A led by Insight to analyze surgery videos and help identify moments when the surgery might have gone wrong
Datch raises $10M Series A led by Blackhorn for its voice AI technology aimed at the industrial workforce. Through the use of natural language processing, they help workers interact with their tool through voice and touch.
Zesty.ai raises $33M Series B led by Centana Growth Partners for its AI-powered climate risk platform. They provide data about wildfires, storms and other severe weather events.
Diagnostic Robotics raises $45M Series B led by StageOne Ventures. They help their customers carry out care management programs and leverage AI to predict patient risk, which improves patient outcomes and care paths.
Neon raises $30 Million Series A-1 led by GGV for its postgres-as-a-service offering
PriceLabs raises $30M led by Summit Partners for its dynamic pricing and revenue management solutions aimed at the short-term rental industry
HiddenLayer raises $6M led by Ten Eleven Ventures to protect AI models from attacks
FeatureByte raises $5.7M Seed led by Glasswing Ventures and Tola Capital. Its SaaS solution is specifically built for data scientists to simplify the creation, serving, managing, and monitoring of ML features.
You raises $25M led by Radical Ventures for its search engine that uses AI to understand search queries, rank the results, and parse the queries into different languages.
If you liked this post, consider sharing it with your friends: