PulseAugur
EN
LIVE 16:25:33

AI Emulates Climate Model 25x Faster, Wins Best Paper at ICML 2024 · 2 sources tracked

Researchers from UC San Diego and the Allen Institute for Artificial Intelligence have developed a generative AI capable of emulating the NOAA FV3GFS climate model approximately 25 times faster. This AI can simulate 10 years of climate data in under three hours, a task that previously took nearly 78 hours on a supercomputer. However, the AI introduces biases that are about 50% higher than the original model, particularly in upper atmospheric layers and polar regions, and currently only supports a single scenario without user-defined runs. This work was recognized with the Best Paper award at ICML 2024. AI

IMPACT This AI's speed could accelerate climate change research and prediction, though its biases need further mitigation.

RANK_REASON AI research paper recognized with an award at a major conference.

Read on Mastodon — fosstodon.org →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

AI Emulates Climate Model 25x Faster, Wins Best Paper at ICML 2024 · 2 sources tracked

COVERAGE [2]

  1. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    UC San Diego + the Allen Institute for AI built a generative AI that emulates NOAA's FV3GFS climate model ~25x faster. 10-year sim: 2h 56m on a GPU cluster vs 7

    UC San Diego + the Allen Institute for AI built a generative AI that emulates NOAA's FV3GFS climate model ~25x faster. 10-year sim: 2h 56m on a GPU cluster vs 78h on a supercomputer. 100 years of climate in ~25 hours. Catch: biases are ~50% higher than FV3GFS, worse up high and a…

  2. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    I found an Interesting new paper on the environmental impact of training language models from the Allen Institute for AI. They're the only organization I've run

    I found an Interesting new paper on the environmental impact of training language models from the Allen Institute for AI. They're the only organization I've run across so far that seems to have any motivation to control the environmental costs of LLMs or in creating fully open so…