From helping transform the way we can predict wind power output to improving the performance of Document AI and giving wider access to #AlphaFold, we’re proud to partner with @GoogleCloud to bring our AI research into the real world.
Here’s how. 🧵
1️⃣ Industries aiming to use AI to read documents need lots of training data, which can be hard to find.
We helped develop machine learning models that need 50% less data to parse utility bills and purchase orders for @GoogleCloud Document AI users.
2️⃣ Wind farms are crucial to building a carbon-free future - but the weather makes it hard to predict.
With @GoogleCloud, we helped produce a Custom AI tool to better predict how much wind power could be generated - trained on forecasts and a customer’s historical turbine data.
3️⃣ After releasing #AlphaFold to the world, which accurately predicts the 3D structure of proteins, we made it available on @GoogleCloud’s Vertex AI.
Scientists can now run their prediction workflow effectively by tracking experiments - plus more. ➡️ dpmd.ai/dm-googlecloud
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Working together with @YouTube’s product and engineering teams, we've helped optimise the decision-making processes that increase safety, decrease latency, and enhance the viewer, creator, and advertiser experience for all.
Our team developed a label quality model that helps label videos with greater precision according to @YouTube’s ad friendly guidelines, improving how videos are identified and classified. 2/
We applied #MuZero to improve the VP9 codec, a coding format that helps compress & transmit video over the internet. It was then applied to some of @YouTube’s live traffic - resulting in an avg. 4% bitrate reduction, helping reduce internet traffic, data usage & loading times. 3/
KFAC-JAX is a library for second-order optimisation of neural networks, and for computing scalable curvature approximations (such as the one used in K-FAC): dpmd.ai/kfac-jax
Congratulations to @demishassabis, John Jumper, and David Baker who will be awarded the Wiley Foundation 20th annual Wiley Prize in Biomedical Sciences on April 1: dpmd.ai/Wiley-Prize 🧵 1/7
Demis and John accept the award on behalf of the @DeepMind team who worked on #AlphaFold, which was recognised as a solution to the “protein folding problem” at CASP14 in Nov 2020: dpmd.ai/casp14_blog 2/7
From the start, we committed to giving broad access to our work and, in July 2021, we published our methods in @Nature along with the open source code.
From packing an umbrella to preparing for extreme conditions, predicting short term weather patterns is crucial for daily life.
New research with the @metoffice and SOTA model advances the science of Precipitation Nowcasting - the prediction of rain: dpmd.ai/nowcasting 1/4
Today’s weather systems provide planet-scale predictions several days ahead, but often struggle to generate high-resolution predictions for short lead times. Nowcasting fills this performance gap, with predictions on rainfall within the next 1-2 hours. 2/4
Compared to widely used-nowcasting methods, meteorologists from @metoffice rated this method as their 1st choice 89% of the time.
There's more to do but our researchers hope this will act as a base for future work & promote greater integration of ML & environmental science. 3/4
In the first lecture of the series, Research Scientist Hado introduces the course and explores the fascinating connection between reinforcement learning and artificial intelligence: dpmd.ai/RLseries1
To tackle all the challenges we meet while solving intelligence, we need tools that are as adaptable as possible. Announcing the paper & code for Perceiver IO, an architecture that handles a wide range of data and tasks, all while scaling gracefully: dpmd.ai/perceiver-IO 1/4
Perceiver IO has the benefits of the Perceiver - domain assumptions✖️, large data✓ - but can produce a huge variety of outputs. W/ one general architecture:
*SOTA on Sintel optical flow
*Beats BERT *without using tokens*
*Multimodal or simultaneous multi-task training
2/4
Perceiver IO can simplify how engineers build systems. For example, Transformers often group language inputs into smaller chunks (“tokenization”) before processing. But Perceiver IO does well without tokenization by learning how to process the raw inputs for itself. 3/4