asterisk labs

Introducing 𝐄𝐚𝐫𝐭𝐡 𝐂𝐨𝐦𝐩𝐫𝐞𝐬𝐬! asterisk labs and National Oceanography Centre are joining forces to tackle the bottlenecks created by the size and complexity of petabyte-scale environmental data, backed by £7 million from the Advanced Research + Invention Agency (ARIA). We’ve never had access to more open data, but there’s so much more that we as a community can do if moving, storing and working with environmental data on a global scale is made feasible for all. 🏭 We want to enable 𝐩𝐞𝐭𝐚𝐛𝐲𝐭𝐞-𝐬𝐜𝐚𝐥𝐞 𝐜𝐨𝐦𝐩𝐫𝐞𝐬𝐬𝐢𝐨𝐧 and analysis using a mixture of recent advancements in the space of AI models, Earth embeddings, and data compression 🫙 Lossy compression and AI are sometimes met with distrust, which is why everything will be designed with 𝐭𝐫𝐚𝐜𝐞𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐚𝐧𝐝 𝐭𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲, allowing users to understand and select appropriate levels of compression 🔓 To make this system trustworthy, 𝐚𝐛𝐬𝐨𝐥𝐮𝐭𝐞𝐥𝐲 𝐞𝐯𝐞𝐫𝐲𝐭𝐡𝐢𝐧𝐠 𝐰𝐞 𝐝𝐨 𝐰𝐢𝐥𝐥 𝐛𝐞 𝐨𝐩𝐞𝐧 𝐬𝐨𝐮𝐫𝐜𝐞, and collectively managed by our community, as public infrastructure 🛰 We will 𝐜𝐨-𝐝𝐞𝐬𝐢𝐠𝐧 𝐰𝐢𝐭𝐡 𝐤𝐞𝐲 𝐬𝐭𝐚𝐤𝐞𝐡𝐨𝐥𝐝𝐞𝐫𝐬 and data producers – including European Centre for Medium-Range Weather Forecasts - ECMWF, EUMETSAT and NERC: Natural Environment Research Council 🤙 We are actively 𝐬𝐞𝐞𝐤𝐢𝐧𝐠 𝐜𝐨𝐥𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐢𝐨𝐧𝐬 and input from everyone in the community. Please reach out, and look out for workshops and events where you can get involved. 🌱 To deliver Earth Compress, Asterisk will be looking for new members, so keep your eyes peeled for job postings in the coming weeks!

CNG Forum 2026 Workshop Want to build AI-ready geospatial datasets – but not sure where to start?  At #CNG2026, Cesar Aybar and Mikolaj Czerkawski from from asterisk labs are leading a hands-on crash course: BYOT (Build Your Own TOM) – AI Dataset Building Crash Course. You'll learn how Major TOM datasets are built using missions like Sentinel-1 and Sentinel-2, distributed in TACO for efficient cloud-native access and GPU throughput, and – in the second half – actually build your own. Join us in Snowbird, Utah for the workshop day (Oct 6) and the main gathering (Oct 7-9)

When DeepSeek dropped, one of the more uncomfortable conversations it sparked was about distillation — how much of a closed model can you recover just by watching its outputs? Google released the Alpha Earth's embeddings. Not the model. The implicit assumption: embeddings are a safe distance from the underlying intelligence that produced them. BetaEarth is a direct stress test of that assumption. Can you train a capable Earth observation model using AEF's public embeddings as the sole supervision signal — no weights, no architecture, no training data? A really interesting work from Mikolaj Czerkawski and team at asterisk labs. Read more at: https://lnkd.in/g7yKtTyA TL;DR: • A light weight student trained purely on AEF's public embeddings recovers 97%+ of its downstream task accuracy. Even a single RGB snapshot gets you surprisingly far. • If a lightweight student can replicate most of the practical utility with no access to your weights, the argument that embedding-only releases preserve your model's moat gets much harder to sustain

📦 Introducing cozip — open a ZIP like a table! ZIP files are everywhere: simple, portable, and supported by almost every OS, language, and data platform. But they were not designed for modern cloud-native data workflows. When a large archive lives on S3, GCS, Hugging Face, or any HTTP server, reaching the file you need can still be slow, awkward, or wasteful. cozip makes ZIP archives cloud-native without breaking ZIP compatibility. The idea is simple: replace the traditional ZIP Central Directory with a Parquet index. That means the archive remains a valid ZIP file, but its contents can also be explored with modern analytical tools. For Earth Observation and ML datasets, this means large archives of tiles, patches, chips, labels, masks, or GeoTIFFs can become easier to browse, filter, and stream from the cloud. The core library is written in C, with bindings for Python, R, and Julia, plus a DuckDB community extension (thanks to Roy Yali Samaniego for showing me how to connect it! 🦆) and a pure JavaScript reader. One early example: the Major TOM VIIRS Nighttime Light dataset is already published as cozip archives on Hugging Face. https://lnkd.in/ekGFPjpc We are releasing the format, tools, docs, and playground — and we would love to hear what people build with it. 🌐 Project page, spec, and docs → https://lnkd.in/eXpEGptX 🎮 Try it in the browser → https://lnkd.in/eC_9cQ7C ⭐ GitHub → https://lnkd.in/e9Fc-DWQ 🦆 DuckDB reader → https://lnkd.in/eBBcJDMP

𝐂𝐀𝐋𝐋 𝐅𝐎𝐑 𝐏𝐀𝐏𝐄𝐑𝐒: 𝐓𝐞𝐫𝐫𝐚𝐁𝐲𝐭𝐞𝐬 𝐈𝐈 𝐢𝐬 𝐚 𝐠𝐨! Save the date of 18 June 2026 - you can submit your papers for both archival and non-archival tracks at the second edition of TerraBytes. This year it will be hosted at European Conference on Computer Vision in Malmö, Sweden. 🔗 𝐂𝐚𝐥𝐥 𝐟𝐨𝐫 𝐏𝐚𝐩𝐞𝐫𝐬: https://lnkd.in/dy8yVFRP Organised together with Valerio Marsocci (European Space Agency - ESA), Mikolaj Czerkawski (asterisk labs), Heng Fang (KTH Royal Institute of Technology), Ritu Yadav (KTH Royal Institute of Technology), Nicolas Audebert (IGN - French Mapping Agency (Institut Géographique National)), Andrea Nascetti (KTH Royal Institute of Technology), Javiera Castillo Navarro (Conservatoire national des arts et métiers - Cnam), Valentin Barriere (Universidad de Chile), Dr Jacqueline Campbell (asterisk labs)

For those of you who, like me, spend a lot of time looking at clouds through a scientific lens, here's a chance to see them from a completely different perspective! asterisk labs and the Cloud Appreciation Society are co-hosting an online conversation between Gavin Pretor-Pinney (founder of the Cloud Appreciation Society) and Prof. Trevor Harley (Psychology, University of Dundee), author of Head in the Clouds. They'll explore how weather and skies shape our minds, moods, and behaviour. Register here to join tomorrow's discussion at 18:00 CEST: https://lnkd.in/dquxm6kw

Join us this Friday for a fascinating chat about how the weather affects our moods!

𝐋𝐄𝐓 𝐓𝐇𝐄𝐑𝐄 𝐁𝐄 𝐍𝐈𝐆𝐇𝐓 🌃 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 𝐂𝐨𝐫𝐞 𝐍𝐢𝐠𝐡𝐭𝐥𝐢𝐠𝐡𝐭𝐬 expansion is now out 🥳 📂 Instant access on Hugging Face: https://lnkd.in/dqpETaCS 🖥️ Web data browser: https://lnkd.in/dBfJE4Pu ☁️ Bonus: we use our custom open-source format developed at asterisk labs - cloud-optimised zip! https://lnkd.in/dhvPGtjQ --- You can learn more about this dataset and what we use it for today at #EGU2026: 𝐓𝐢𝐦𝐞, 𝐒𝐩𝐚𝐜𝐞, 𝐚𝐧𝐝 𝐍𝐢𝐠𝐡𝐭𝐥𝐢𝐠𝐡𝐭𝐬: 𝐆𝐥𝐨𝐛𝐚𝐥 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 𝐄𝐚𝐫𝐭𝐡 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 Mikolaj Czerkawski, Marcin Kluczek, Jędrzej S. Bojanowski 📅 Wednesday, 14:05–14:15,  Room -2.33 🔗 https://lnkd.in/dRinmAkM --- 💡 This pre-release only contains median annual tiles sampled over the years in Major TOM Core. A full dataset is being prepared by Marcin Kluczek so make sure to stay tuned and follow him if you're interested in full data coverage. Joint community effort between asterisk labs and CloudFerro S.A. w/ Marcin Kluczek & Jędrzej S. Bojanowski

Thank you for joining us and Development Seed tonight! 🥰

Getting ready for the Development Seed and asterisk labs EO Community Social at #EGU26! 🛰️ 🌍