The A.I Megathread (LLM , GPT , Development)
MetNet-3: A state-of-the-art neural weather model available in Google products
blog.research.google
MetNet-3: A state-of-the-art neural weather model available in Google products
WEDNESDAY, NOVEMBER 01, 2023Posted by Samier Merchant, Google Research, and Nal Kalchbrenner, Google DeepMind
Forecasting weather variables such as precipitation, temperature, and wind is key to numerous aspects of society, from daily planning and transportation to energy production. As we continue to see more extreme weather events such as floods, droughts, and heat waves, accurate forecasts can be essential to preparing for and mitigating their effects. The first 24 hours into the future are especially important as they are both highly predictable and actionable, which can help people make informed decisions in a timely manner and stay safe.
Today we present a new weather model called MetNet-3, developed by Google Research and Google DeepMind. Building on the earlier MetNet and MetNet-2 models, MetNet-3 provides high resolution predictions up to 24 hours ahead for a larger set of core variables, including precipitation, surface temperature, wind speed and direction, and dew point. MetNet-3 creates a temporally smooth and highly granular forecast, with lead time intervals of 2 minutes and spatial resolutions of 1 to 4 kilometers. MetNet-3 achieves strong performance compared to traditional methods, outperforming the best single- and multi-member physics-based numerical weather prediction (NWP) models — such as High-Resolution Rapid Refresh (HRRR) and ensemble forecast suite (ENS) — for multiple regions up to 24 hours ahead.
Finally, we’ve integrated MetNet-3’s capabilities across various Google products and technologies where weather is relevant. Currently available in the contiguous United States and parts of Europe with a focus on 12 hour precipitation forecasts, MetNet-3 is helping bring accurate and reliable weather information to people in multiple countries and languages.
|
|
| MetNet-3 precipitation output summarized into actionable forecasts in Google Search on mobile. |
Densification of sparse observations
Many recent machine learning weather models use the atmospheric state generated by traditional methods (e.g., data assimilation from NWPs) as the primary starting point to build forecasts. In contrast, a defining feature of the MetNet models has been to use direct observations of the atmosphere for training and evaluation. The advantage of direct observations is that they often have higher fidelity and resolution. However, direct observations come from a large variety of sensors at different altitudes, including weather stations at the surface level and satellites in orbit, and can be of varying degrees of sparsity. For example, precipitation estimates derived from radar such as NOAA’s Multi-Radar/Multi-Sensor System (MRMS) are relatively dense images, whereas weather stations located on the ground that provide measurements for variables such as temperature and wind are mere points spread over a region.In addition to the data sources used in previous MetNet models, MetNet-3 includes point measurements from weather stations as both inputs and targets with the goal of making a forecast at all locations. To this end, MetNet-3’s key innovation is a technique called densification, which merges the traditional two-step process of data assimilation and simulation found in physics-based models into a single pass through the neural network. The main components of densification are illustrated below. Although the densification technique applies to a specific stream of data individually, the resulting densified forecast benefits from all the other input streams that go into MetNet-3, including topographical, satellite, radar, and NWP analysis features. No NWP forecasts are included in MetNet-3’s default inputs.
|
| A) During training, a fraction of the weather stations are masked out from the input while kept in the target. B) To evaluate generalization to untrained locations, a set of weather stations represented by squares is never used for training and is only used for evaluation. C) Data from these held out weather stations with sparse coverage is included during evaluation to determine prediction quality in these areas. D) The final forecasts use the full set of training weather stations as input and produce fully dense forecasts aided by spatial parameter sharing. |
High resolution in space and time
A central advantage of using direct observations is their high spatial and temporal resolution. For example, weather stations and ground radar stations provide measurements every few minutes at specific points and at 1 km resolutions, respectively; this is in stark contrast with the assimilation state from the state-of-the-art model ENS, which is generated every 6 hours at a resolution of 9 km with hour-by-hour forecasts. To handle such a high resolution, MetNet-3 preserves another of the defining features of this series of models, lead time conditioning. The lead time of the forecast in minutes is directly given as input to the neural network. This allows MetNet-3 to efficiently model the high temporal frequency of the observations for intervals as brief as 2 minutes. Densification combined with lead time conditioning and high resolution direct observations produces a fully dense 24 hour forecast with a temporal resolution of 2 minutes, while learning from just 1,000 points from the One Minute Observation (OMO) network of weather stations spread across the United States.MetNet-3 predicts a marginal multinomial probability distribution for each output variable and each location that provides rich information beyond just the mean. This allows us to compare the probabilistic outputs of MetNet-3 with the outputs of advanced probabilistic ensemble NWP models, including the ensemble forecast ENS from the European Centre for Medium-Range Weather Forecasts and the High Resolution Ensemble Forecast (HREF) from the National Oceanic and Atmospheric Administration of the US. Due to the probabilistic nature of the outputs of both models, we are able to compute scores such as the Continuous Ranked Probability Score (CRPS). The following graphics highlight densification results and illustrate that MetNet’s forecasts are not only of much higher resolution, but are also more accurate when evaluated at the overlapping lead times.
|
| Top: MetNet-3’s forecast of wind speed for each 2 minutes over the future 24 hours with a spatial resolution of 4km. Bottom: ENS’s hourly forecast with a spatial resolution of 18 km. The two distinct regimes in spatial structure are primarily driven by the presence of the Colorado mountain ranges. Darker corresponds to higher wind speed. More samples available here: 1, 2, 3, 4. |
|
| Performance comparison between MetNet-3 and NWP baseline for wind speed based on CRPS (lower is better). In the hyperlocal setting, values of the test weather stations are given as input to the network during evaluation; the results improve further especially in the early lead times. |
In contrast to weather station variables, precipitation estimates are more dense as they come from ground radar. MetNet-3’s modeling of precipitation is similar to that of MetNet-1 and 2, but extends the high resolution precipitation forecasts with a 1km spatial granularity to the same 24 hours of lead time as the other variables, as shown in the animation below. MetNet-3’s performance on precipitation achieves a better CRPS value than ENS’s throughout the 24 hour range.
We developed such a real-time system that is capable of producing a precipitation forecast every few minutes for the entire contiguous United States and for 27 countries in Europe for a lead time of up to 12 hours.
The system's uniqueness stems from its use of near-continuous inference, which allows the model to constantly create full forecasts based on incoming data streams. This mode of inference is different from traditional inference systems, and is necessary due to the distinct characteristics of the incoming data. The model takes in various data sources as input, such as radar, satellite, and numerical weather prediction assimilations. Each of these inputs has a different refresh frequency and spatial and temporal resolution. Some data sources, such as weather observations and radar, have characteristics similar to a continuous stream of data, while others, such as NWP assimilations, are similar to batches of data. The system is able to align all of these data sources spatially and temporally, allowing the model to create an updated understanding of the next 12 hours of precipitation at a very high cadence.
With the above process, the model is able to predict arbitrary discrete probability distributions. We developed novel techniques to transform this dense output space into user-friendly information that enables rich experiences throughout Google products and technologies.
The state-of-the-art accuracy, high temporal and spatial resolution, and probabilistic nature of MetNet-3 makes it possible to create unique hyperlocal weather insights. For the contiguous United States and Europe, MetNet-3 is operational and produces real-time 12 hour precipitation forecasts that are now served across Google products and technologies where weather is relevant, such as Search. The rich output from the model is synthesized into actionable information and instantly served to millions of users.
For example, a user who searches for weather information for a precise location from their mobile device will receive highly localized precipitation forecast data, including timeline graphs with granular minute breakdowns depending on the product.
|
| Case study for Thu Jan 17 2019 00:00 UTC showing the probability of instantaneous precipitation rate being above 1 mm/h on CONUS. Darker corresponds to a higher probability value. The maps also show the prediction threshold when optimized towards Critical Success Index CSI (dark blue contours). This specific case study shows the formation of a new large precipitation pattern in the central US; it is not just forecasting of existing patterns. Top: ENS’s hourly forecast. Center: Ground truth, source NOAA’s MRMS. Bottom: Probability map as predicted by MetNet-3. Native resolution available here. |
|
| Performance comparison between MetNet-3 and NWP baseline for instantaneous precipitation rate on CRPS (lower is better). |
Delivering realtime ML forecasts
Training and evaluating a weather forecasting model like MetNet-3 on historical data is only a part of the process of delivering ML-powered forecasts to users. There are many considerations when developing a real-time ML system for weather forecasting, such as ingesting real-time input data from multiple distinct sources, running inference, implementing real-time validation of outputs, building insights from the rich output of the model that lead to an intuitive user experience, and serving the results at Google scale — all on a continuous cycle, refreshed every few minutes.We developed such a real-time system that is capable of producing a precipitation forecast every few minutes for the entire contiguous United States and for 27 countries in Europe for a lead time of up to 12 hours.
|
| Illustration of the process of generating precipitation forecasts using MetNet-3. |
With the above process, the model is able to predict arbitrary discrete probability distributions. We developed novel techniques to transform this dense output space into user-friendly information that enables rich experiences throughout Google products and technologies.
Weather features in Google products
People around the world rely on Google every day to provide helpful, timely, and accurate information about the weather. This information is used for a variety of purposes, such as planning outdoor activities, packing for trips, and staying safe during severe weather events.The state-of-the-art accuracy, high temporal and spatial resolution, and probabilistic nature of MetNet-3 makes it possible to create unique hyperlocal weather insights. For the contiguous United States and Europe, MetNet-3 is operational and produces real-time 12 hour precipitation forecasts that are now served across Google products and technologies where weather is relevant, such as Search. The rich output from the model is synthesized into actionable information and instantly served to millions of users.
For example, a user who searches for weather information for a precise location from their mobile device will receive highly localized precipitation forecast data, including timeline graphs with granular minute breakdowns depending on the product.
|
| MetNet-3 precipitation output in weather on the Google app on Android (left) and mobile web Search (right). |
Conclusion
MetNet-3 is a new deep learning model for weather forecasting that outperforms state-of-the-art physics-based models for 24-hour forecasts of a core set of weather variables. It has the potential to create new possibilities for weather forecasting and to improve the safety and efficiency of many activities, such as transportation, agriculture, and energy production. MetNet-3 is operational and its forecasts are served across several Google products where weather is relevant.Acknowledgements
Many people were involved in the development of this effort. We would like to especially thank those from Google DeepMind (Di Li, Jeremiah Harmsen, Lasse Espeholt, Marcin Andrychowicz, Zack Ontiveros), Google Research (Aaron Bell, Akib Uddin, Alex Merose, Carla Bromberg, Fred Zyda, Isalo Montacute, Jared Sisk, Jason Hickey, Luke Barrington, Mark Young, Maya Tohidi, Natalie Williams, Pramod Gupta, Shreya Agrawal, Thomas Turnbull, Tom Small, Tyler Russell), and Google Search (Agustin Pesciallo, Bill Myers, Danny Cheresnick, Jonathan Karsh, Lior Cohen, Maca Piombi, Maia Diamant, Max Kamenetsky, Maya Ekron, Mor Schlesinger, Neta Gefen-Doron, Nofar Peled Levi, Ofer Lehr, Or Hillel, Rotem Wertman, Tamar Shevach,Vinay Ruelius Shah, Yechie Labai).No the actual canva online app. Looks like there is a chatgpt I don't know about for it! Good looking out!you using this?
ChatGPT - Canva
Effortlessly design anything: presentations, logos, social media posts and more.chat.openai.com
Marketers Can Upgrade Or Subscribe To ChatGPT Plus For GPTs
OpenAI reopened ChatGPT Plus subscriptions. Here are some reasons why marketers and SEOs will want to upgrade.www.searchenginejournal.com
I’ve resigned from my role leading the Audio team at Stability AI, because I don’t agree with the company’s opinion that training generative AI models on copyrighted works is ‘fair use’.
First off, I want to say that there are lots of people at Stability who are deeply thoughtful about these issues. I’m proud that we were able to launch a state-of-the-art AI music generation product trained on licensed training data, sharing the revenue from the model with rights-holders. I’m grateful to my many colleagues who worked on this with me and who supported our team, and particularly to Emad for giving us the opportunity to build and ship it. I’m thankful for my time at Stability, and in many ways I think they take a more nuanced view on this topic than some of their competitors.
But, despite this, I wasn’t able to change the prevailing opinion on fair use at the company.
This was made clear when the US Copyright Office recently invited public comments on generative AI and copyright, and Stability was one of many AI companies to respond. Stability’s 23-page submission included this on its opening page:
“We believe that Al development is an acceptable, transformative, and socially-beneficial use of existing content that is protected by fair use”.
For those unfamiliar with ‘fair use’, this claims that training an AI model on copyrighted works doesn’t infringe the copyright in those works, so it can be done without permission, and without payment. This is a position that is fairly standard across many of the large generative AI companies, and other big tech companies building these models — it’s far from a view that is unique to Stability. But it’s a position I disagree with.
I disagree because one of the factors affecting whether the act of copying is fair use, according to Congress, is “the effect of the use upon the potential market for or value of the copyrighted work”. Today’s generative AI models can clearly be used to create works that compete with the copyrighted works they are trained on. So I don’t see how using copyrighted works to train generative AI models of this nature can be considered fair use.
But setting aside the fair use argument for a moment — since ‘fair use’ wasn’t designed with generative AI in mind — training generative AI models in this way is, to me, wrong. Companies worth billions of dollars are, without permission, training generative AI models on creators’ works, which are then being used to create new content that in many cases can compete with the original works. I don’t see how this can be acceptable in a society that has set up the economics of the creative arts such that creators rely on copyright.
To be clear, I’m a supporter of generative AI. It will have many benefits — that’s why I’ve worked on it for 13 years. But I can only support generative AI that doesn’t exploit creators by training models — which may replace them — on their work without permission.
I’m sure I’m not the only person inside these generative AI companies who doesn’t think the claim of ‘fair use’ is fair to creators. I hope others will speak up, either internally or in public, so that companies realise that exploiting creators can’t be the long-term solution in generative AI
First off, I want to say that there are lots of people at Stability who are deeply thoughtful about these issues. I’m proud that we were able to launch a state-of-the-art AI music generation product trained on licensed training data, sharing the revenue from the model with rights-holders. I’m grateful to my many colleagues who worked on this with me and who supported our team, and particularly to Emad for giving us the opportunity to build and ship it. I’m thankful for my time at Stability, and in many ways I think they take a more nuanced view on this topic than some of their competitors.
But, despite this, I wasn’t able to change the prevailing opinion on fair use at the company.
This was made clear when the US Copyright Office recently invited public comments on generative AI and copyright, and Stability was one of many AI companies to respond. Stability’s 23-page submission included this on its opening page:
“We believe that Al development is an acceptable, transformative, and socially-beneficial use of existing content that is protected by fair use”.
For those unfamiliar with ‘fair use’, this claims that training an AI model on copyrighted works doesn’t infringe the copyright in those works, so it can be done without permission, and without payment. This is a position that is fairly standard across many of the large generative AI companies, and other big tech companies building these models — it’s far from a view that is unique to Stability. But it’s a position I disagree with.
I disagree because one of the factors affecting whether the act of copying is fair use, according to Congress, is “the effect of the use upon the potential market for or value of the copyrighted work”. Today’s generative AI models can clearly be used to create works that compete with the copyrighted works they are trained on. So I don’t see how using copyrighted works to train generative AI models of this nature can be considered fair use.
But setting aside the fair use argument for a moment — since ‘fair use’ wasn’t designed with generative AI in mind — training generative AI models in this way is, to me, wrong. Companies worth billions of dollars are, without permission, training generative AI models on creators’ works, which are then being used to create new content that in many cases can compete with the original works. I don’t see how this can be acceptable in a society that has set up the economics of the creative arts such that creators rely on copyright.
To be clear, I’m a supporter of generative AI. It will have many benefits — that’s why I’ve worked on it for 13 years. But I can only support generative AI that doesn’t exploit creators by training models — which may replace them — on their work without permission.
I’m sure I’m not the only person inside these generative AI companies who doesn’t think the claim of ‘fair use’ is fair to creators. I hope others will speak up, either internally or in public, so that companies realise that exploiting creators can’t be the long-term solution in generative AI
Last edited:
.jpg)
DeepMind Wants to Use AI to Solve the Climate Crisis
WIRED spoke with DeepMind’s climate lead about techno-utopianism, ways AI can help fight climate change, and what’s currently standing in the way.
DeepMind Wants to Use AI to Solve the Climate Crisis
WIRED spoke with DeepMind’s climate lead about techno-utopianism, ways AI can help fight climate change, and what’s currently standing in the way..jpg "aerial of flood in Australia")
Aerial view flooded tree-lined Bookpurnong Road, the main Loxton to Berri connector road on River Murray in South Australia.PHOTOGRAPH: GETTY IMAGES
It’s a perennial question at WIRED: Tech got us into this mess, can it get us out? That’s particularly true when it comes to climate change. As the weather becomes more extreme and unpredictable, there are hopes that artificial intelligence—that other existential threat—might be part of the solution.
DeepMind, the Google-owned artificial intelligence lab, has been using its AI expertise to tackle the climate change problem in three different ways, as Sims Witherspoon, DeepMind’s climate action lead, explained in an interview ahead of her talk at WIRED Impact in London on November 21. This conversation has been edited for clarity and length.
WIRED: How can AI help us tackle climate change?
Sims Witherspoon: There are lots of ways we can slice the answer. AI can help us in mitigation. It can help us in adaptation. It can help us with addressing loss and damage. It can help us in biodiversity and ecology and much more. But I think one of the ways that makes it more tangible for most people is to talk about it through the lens of AI’s strengths.
I think of it in three parts: First and foremost, AI can help us understand climate change and the problems that we face related to climate change through better models for prediction and monitoring. One example is our work on precipitation nowcasting—so, forecasting rain a few hours ahead—and our models were voted more useful and more accurate than other methods by Met Office forecasters, which is great.
But it’s also just the start because you can then build to predict much more complex phenomena. So AI can be a really significant tool in helping us understand climate change as a problem.
What’s the second thing?
The second bucket that I like to think about is the fact that AI can help us optimize current systems and existing infrastructure. It’s not enough to start building new green technology for a more sustainable tomorrow, life needs to go on—we already have many systems that we rely on today, and we can’t just burn them all down and start from scratch. We need to be able to optimize those existing systems and infrastructure, and AI is one of the tools that we can use to do this.
A nice example of this is the work we did in data centers, where we were able to improve energy efficiency and achieve a 30 percent energy saving.
And then the third thing is new technology?
Yes, the third bucket is the way that most people think about AI, when they think about the Hollywood version or what you read about in sci-fi novels and things, which is accelerating breakthrough science.
I really like the example of nuclear fusion and plasma control—we published a Nature paper where we used neural nets to train a reinforcement learning model to learn how to control plasma shapes in a real-world tokamak [a nuclear fusion reactor]. And that’s really important because actually understanding plasma physics and being able to control those shapes and configurations is an incredibly important building block to ultimately achieving a nearly inexhaustible supply of carbon-free energy.
You can’t really talk about AI and climate change without reference to the carbon footprint of AI itself, and the massive amounts of energy being consumed by data centers, which is something people are becoming a lot more aware of. How do you think about that problem? When will AI get to the point where it’s saved more carbon than it’s used to be trained?
I would love to see that analysis; I don’t know if anyone’s done it. Many of the language models and generative AI success stories we’ve seen over recent years, it’s true, they are energy-intensive, and this is a problem that we’ve documented. We believe it’s really important to see and understand how much energy these models use and be open about that, and then we also have a number of efforts to reduce the compute needed for these models. So we think about it in a few ways—not as globally as, “Is the carbon we’ve burned worth the solutions?” but more about, “How do you deploy solutions that are as carbon-efficient as possible?”
What are the roadblocks that are going to stop AI being used to fight climate change?
The first one is access to data. There are significant gaps in climate-critical data across all sectors, whether it’s electricity or transportation or buildings and cities. There’s a group that we work with that publishes a “climate critical data set wishlist,” and I think having those datasets and getting people comfortable—where it’s safe and responsible to do so—with opening up climate-critical data sets is incredibly important.
The other part that I put almost on par with data is working with domain experts. At Google DeepMind, we are focused on AI research and AI product development—we’re not plasma physicists, we’re not electrical engineers. And so when we’re trying to figure out problems that we want to solve, we really need to be working with those experts who can teach us about the problems that they experienced and the things that are blocking them. That does two things. One, it ensures that we fully understand what we’re building an AI solution for. And the second thing is it ensures that whatever we’re building is going to get used. We don’t just want to make this cool piece of tech and then hope that somebody uses it.
Are there any safety considerations? People might be nervous about seeing the words “nuclear fusion” and “artificial intelligence” in the same sentence …
In my area specifically, one of the ways we deal with that is again going back to working with domain experts—making sure we understand the systems really well, and what they need to keep the system safe. It’s those experts that teach us about that, and then we build solutions that are within those guardrails.
Within climate and sustainability, we also do a lot of impact analysis: what we expect our potential impact to be and then all the downstream effects of that.
You’ve said you’re a techno-optimist, so what’s the techno-optimist view of a future where AI is fully brought to bear on climate change?
The techno-optimist’s view is that—provided we’re able to wield it effectively—we’re able to use a transformative tool like AI to solve sector-specific and non-sector-specific problems more quickly, and at a scale we wouldn’t be able to without AI. One of the things I’m most excited about is the versatility and scalability of the tool. And given the amount of problems we need to solve related to climate change, what we need is a highly versatile and highly scalable tool.
Join Sims Witherspoon and our world-class speaker line-up at WIRED Impact, November 21, at Magazine, London, as we examine the challenges and opportunities for organizations to innovate to tackle humankind’s most pressing challenge. Get tickets now: events.wired.co.uk/impact
dondatta24
Rookie
For the beginners, this might be a helpful guide to start with AI:
/cdn.vox-cdn.com/uploads/chorus_asset/file/25088828/unnamed__1_.jpg)
Google is embedding inaudible watermarks right into its AI generated music
SynthID comes to audio.
www.theverge.com
Google is embedding inaudible watermarks right into its AI generated music
/SynthID will be used to watermark audio from DeepMind’s Lyria model, so it’s possible to work out if Google’s AI tech has been used in the creation of a track.
By Jon Porter, a reporter with five years of experience covering consumer tech releases, EU tech policy, online platforms, and mechanical keyboards.Nov 16, 2023, 6:38 AM EST|7 Comments / 7 New
Audio created using Google DeepMind’s AI Lyria model, such as tracks made with YouTube’s new audio generation features, will be watermarked with SynthID to let people identify their AI-generated origins after the fact. In a blog post, DeepMind said the watermark shouldn’t be detectable by the human ear and “doesn’t compromise the listening experience,” and added that it should still be detectable even if an audio track is compressed, sped up or down, or has extra noise added.
Watermarking tools like SynthID are seen as an important safeguard against some of the harms of generative AI. President Joe Biden’s executive order on artificial intelligence, for example, calls for a new set of government-led standards for watermarking AI-generated content. It’s a promising area, but current technologies are far from a silver bullet to defend against fakes.
According to DeepMind, SynthID’s audio implementation works by “converting the audio wave into a two-dimensional visualization that shows how the spectrum of frequencies in a sound evolves over time.” It claims the approach is “unlike anything that exists today.”
The news that Google is embedding the watermarking feature into AI-generated audio comes just a few short months after the company released SynthID in beta for images created by Imagen on Google Cloud’s Vertex AI. The watermark is resistant to editing like cropping or resizing, although DeepMind cautioned that it’s not foolproof against “extreme image manipulations.”
