Suggestions
Bruno Nuño
Microsoft Planetary Computer and "AI for Earth" programs Director
Bruno Sanchez-Andrade Nuño is an accomplished professional with extensive experience in the fields of AI, space technology, remote sensing, and data science. Here are some key details about his career and background:
Professional Experience
Bruno served as the Planetary Computer Program Director at Microsoft from February 2020 to January 2023.1 In this role, he:
- Directed the "AI for Earth" and Planetary Computer programs
- Managed a $50 million+ initiative with over 800 projects
- Focused on monitoring, modeling, and managing Earth's natural systems
- Led strategy, team hiring, external relationships, and operations for Microsoft's Planetary Computer
Prior to Microsoft, Bruno held several other notable positions:
- Strategy Consultant at ImpactScience.dev (2017-present)
- Senior Consultant at the World Bank Group (2017-2020)
- Innovation Labs / Data Scientist at the World Bank Group (2015-2017)
- Chief Scientist at Mapbox (2013-2015)
- Vice President of Social Impact at Satellogic (2017-2019)
Education and Expertise
Bruno has a strong academic background:
- PhD in Astrophysics1
- Postdoctoral work in rocket science1
- Executive education modules from Harvard Kennedy School, Stanford University, and University of Oxford1
His areas of expertise include:
- AI and machine learning
- Remote sensing and satellite technology
- Data science and big data analytics
- Sustainable development and environmental monitoring
Impact and Approach
Bruno describes himself as an "Impact Scientist" and has focused on applying scientific knowledge to create positive social and environmental change.1 He has worked on diverse projects, including:
- Satellite nightlights for development at the World Bank
- Global transportation network assessment at Mapbox
- Fostering entrepreneurship skills in refugee camps
- Advising on competitiveness for the UAE government
His LinkedIn username is indeed "nasonurb", which appears to be his name spelled backwards.1
Highlights
35 trillion pixels to make Clay, 100% open:
Trained on 70 million satellite and aerial image chips of size 224x224px from all around the world, Clay v1 has a global perspective, statistically sampled and spanning from 2018 to 2023. Its diverse training set includes imagery from Landsat 8 and 9 (11 million chips), Sentinel 2 L2A (18 million), Sentinel 1 RTC (16 million), NAIP in the USA (20 million), and high-resolution LINZ data from New Zealand (3 million). Resolutions range from 30 meters to 30 centimeters. All fully open data.
This vast training set, covering roughly 13 million square kilometers (2.6% of the Earth's surface), was processed over 4000 GPU hours on Nvidia H100 onboard p5.48xlarge servers, and clocked 10 epochs. This means that Clay v1 has been exposed in training to approximately 35 trillion pixels, each with 3 to 10 spectral bands. The result is a strongly trained model with 201 million parameters and a compact 768MB footprint, capable of extracting deep insights from Earth observation data.
The Clay model code is available on Github, where it already has 15 direct contributors, 268 stars, 111 clones, 30 forks, and 4,750 Views.
To our knowledge the largest geospatial model by training, instruments and coverage. More info: https://lnkd.in/d9VKRDq4
.... and that's just the starting line, we also have a no-code app, and a growing list of real-world partners and applications, for whom we in the end did this whole bit: AI for Earth.
#AIforEarth #machinelearning
I'll be teaching part of the PhD course on #AIforEarth, so excited!
As we roll out Clay it's become clear to me that this is NOT about creating a finest AI to solve all our problems, nor is it about dismissing AI as just stochastic parroting hallucinations. Technology, or data, doesn't change the world—people do. People with the right data and tech.
This magic happens when you have one foot grounded in each world: the world of Data and Transformer code; and the world of budgets and mud on your toes. Unfortunately, there aren't many people with that combo of skills and experience, but it's exactly that combination that delivers the actual promise.
It won't be Clay v1 or GPT-5 that solves #AIforEarth. It will be a ragtag group of pioneers willing to try new paths, tackle messy data, explain themselves a million times, and risk their careers when they could choose not to. Why? Because the upside is getting the same results much faster, better, cheaper, and getting whole new benefits we cannot yet imagine.
I humbly consider myself part of that oddball group. I've fought the code in my Astrophysics PhD and rocket science, building the Planetary Computer, through fast.ai ML courses and long nights; and I also got my share of mud building a Data Science team in remote Bhutan, teaching OSM in Panama or South Africa, mapping refugee camps in Kakuma and cadasters in Kosovo with drones, improving traffic jams in Cebu with Grab, or using AI for loan risk assessments for a Bank for women in Mexico. I live for this stuff, even wrote a book about it ("Impact Science").
All this to say what, and why, I'll be teaching at this PhD course: SSL4EO: Self-Supervised Learning for Earth Observation, from July 01-05, 2024, in Copenhagen, Denmark. This unique course is designed not just to teach the latest SSL techniques but to integrate them seamlessly into real-world Earth observation applications.
Why Attend? 🔍 Dive deep into self-supervised learning (SSL) methods 💡 Address the unique challenges and opportunities of Earth observation data 🤝 Collaborate and network with fellow PhD students and experts in the field 🛠️ Gain practical, hands-on experience with real-world projects
I'll be speaking alongside Randall Balestriero (META), Marc Rußwurm (Wageningen), Konstantin Klemmer (MSFT), Jan Dirk Wegner (UZH/ETH) and Xiaoxiang Zhu, (TUM). All orchestrated by Stefan Oehmcke, Nico Lang, Ankit K., Christian Igel, Rasmus Fensholt. (U. Copenhagen).
If you're passionate about leveraging SSL for Earth observation, this is the course for you. Register now and join us in Copenhagen for an enriching experience.
📅 Dates: July 01 - 05, 2024 📍 Location: Festsalen, Øster Voldgade 10, 1350 Copenhagen, Denmark 🎓 Earn 2.5 ECTS credit points Learn more and register here: https://lnkd.in/d7ACNciG #SelfSupervisedLearning #EarthObservation #OpenData #Sustainability #Geospatial #Innovation #Education
