Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Columbia University PhD students Sruthi Sudhakar, Marten Garicano, Mithra Karamchedu, Edward Guo, Terry Tong, and Jun Ren have been selected for fellowships including the 2026 Apple Scholars in AIML and the NSF Graduate …
Google has introduced WeatherNext 3, the next-generation release of its global weather AI model, offering hourly forecasts up to 15 days ahead with global coverage and finer spatial resolution than its predecessor. The m…
Google DeepMind and Google Research announced WeatherNext 3 on September 3, 2026, an AI-powered global weather model that generates forecasts every hour using live geostationary satellite data, with precipitation accurac…
Seventy-five researchers, including four from USC Viterbi School of Engineering, attended the Argonne Training Program on Extreme-Scale Computing (ATPESC), a two-week intensive course covering GPU architectures, parallel…
Google launched WeatherNext 3, its most advanced AI weather model, which generates hourly forecasts at up to 5km resolution and improves precipitation forecast accuracy by up to 50% for forecasts a day or more ahead. The…
Google announced WeatherNext 3, its most advanced global weather AI model, which uses live geostationary satellite data to deliver forecasts up to five times sharper than its predecessor and up to 50% more accurate preci…
Sander Dieleman, a research scientist at Google DeepMind, published a blog post on September 25, 2024, explaining the paradox of diffusion distillation, where reducing sampling steps can improve efficiency without sacrif…
Developer gpjt has uploaded PyTorch-compatible versions of all his JAX-trained GPT-2 models to the Hugging Face Hub, including models from his blog series on writing an LLM from scratch and Chinchilla scaling experiments…
A developer built a pipeline that translates podcast episodes into other languages for about a dollar per episode, using the podcaster's own published transcript rather than speech-to-text to avoid compounding errors. Th…
MAI-Transcribe-2, a new speech recognition model from MAI, claims to be the fastest, most accurate, and cheapest in the world, ranking first on the FLEURS benchmark across 60 languages with an average Word-Error-Rate of …
Learning in imagination, a model-based reinforcement learning approach that trains policies solely on data generated by an action-conditioned world model, addresses sample efficiency by generating data for free, as exemp…
Ligatum released Lustro V1, a deterministic state-flow diffusion engine that derives transformation parameters from evolving state rather than predefined schedules, with scalar speeds exceeding 13GB/s on an i5-11600K. Th…
Berlin-based audio engineer Yuval Miller has introduced Dr. Session, a standalone Mac app that uses a neural audio model to identify and sort audio stems and samples by content rather than filenames, and can build ready-…
Google has released WeatherNext 3, the first global weather model to generate hourly forecasts, drawing directly from raw satellite imagery to deliver higher-resolution and more accurate predictions. The model predicts s…
A new reinforcement learning method called semi-pessimistic pseudo labelling (SPL), developed by researchers including Omar Rivasplata of the University of Manchester, has been accepted for publication in the Journal of …
Google DeepMind and Google Research introduced WeatherNext 3, described as the most advanced and accurate global weather AI model to date, generating hourly forecasts at up to 5-kilometer resolution—roughly five times sh…
Google DeepMind and Google Research released WeatherNext 2 and WeatherNext Cyclones (WN-C), AI weather forecasting models that outperform previous models on 99.9% of forecast variables and deliver forecasts up to 8 times…
Google DeepMind and Google Research introduced WeatherNext 3, a global weather AI model that generates hourly forecasts at up to 5-kilometer resolution, roughly five times sharper than its predecessor WeatherNext 2, and …
Google DeepMind and Google Research released WeatherNext 3, an AI weather forecasting model that outperforms leading AI and traditional models on the Operational WeatherBench benchmark, with 5km resolution, 60% improved …
Google DeepMind released WeatherNext 3 on Thursday, an AI weather model that refreshes forecasts every hour from satellite images, breaking reliance on six-hourly government data cycles and improving temperature, precipi…