Madeline Lisaius (@mlisaius) header
Madeline Lisaius (@mlisaius) profile picture

PhD student (remote sensing + food security + justice) @AI4ER_CDT @Cambridge_Uni / Alumna of @Stanford, @NatGeoSociety, @Planet, @RockefellerFd (she/her)

Cambridge, UK
Joined November 2015
Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)Photo by Madeline Lisaius (@mlisaius)
Looking forward to share our EO foundation model work at @esa #livingplanetsymposium #lsp at the end of the month! Happy to chat over a coffee all week during the event to swap ideas, especially on applications of FMs ☕️
39
(1/) Why do Earth observation foundation models disappoint? I had a conversation in Cambridge recently and used this analogy. Imagine that you have a foundation model (FM) that takes input images of foods, and organizes them into a latent space that we imagine as a 2-D box.
1
64
(3/) Now imagine we train a second FM on a more diverse dataset: fruits, vegetables, and bakery. The box stays the same, but now it needs to hold many more types of food. As a result, fruits only get a small part of the map, so distinctions between fruit types become blurrier.
5
50
(8/) Looking forward to share more soon!
30
(7/) In recent months, our work at Cambridge has found that a smart FM approach can beat best practice across multiple tasks while offering computational and ease of use benefits.  And in doing so, we’ve seen this generalization–specialization tension play out firsthand.
19
(6/) While we can increase model size, expand the latent space, or tweak learning objective, the fundamental challenge remains: broad generalization often comes at the cost of local precision. The challenge is to balance these two.
20
(5/) ...like distinguishing different types of crop stress or identifying subtle urban heat islands, the downstream approaches may struggle. The “fruit details” are lost in the crowd.
18
(4/) this is what happens in Earth observation FMs. When trained on diverse types of satellite imagery—across deserts, tropical forest, tundra, urban areas, etc.—the model learns broad features that apply generally. But when you want to do something specialized...
15
Surprised - an interesting read and big change as well. So much respect for folks who worked on exceptional projects at Descartes. philosophygeek.medium.com/me…
Madeline Lisaius retweeted
By far, the most accurate representation of machine learning pipelines in the real world 🍿 😀
204
1,547
419
10,927
Perceptions of neural nets over the decades... A thread. [1/30]
20
167
31
795
My "spark" for caring about landscape change was watching the old growth in WA state be cut down as a child - a deeply emotional and painful experience for me + communities in this region. Of many agonizing things happening right now in the world, this is included.
This quoted post is unavailable.
1
Madeline Lisaius retweeted
Did you know that a UK PhD stipend is so little that it puts recipients below the absolute poverty line? UK PhD stipend (outside London): £16,062 Absolute poverty line (60% of 2011 median income): £16,883 No one does research for the money but surely we can do better than this?
163
770
198
5,424
Madeline Lisaius retweeted
BREAKING: The world's elite have arrived in Davos in their private jets, helicopters and black luxury cars to lecture you about carbon emissions.
152
2,808
649
11,669
"African civil-­society groups are now pleading with the U.N. to act, lest Ethiopia devolve into ethnic cleansing reminiscent of Rwanda." This piece makes it sound like ethnic cleansing hasn't happened yet when we've seen how much of it for how long? time.com/collection/100-most…
37
637
116
657
Madeline Lisaius retweeted
#Copernicus🇪🇺🛰️provides detailed information on #AirQuality on a global scale #DYK that if you check air quality 🌬️in your city using an app on your smartphone 📲, it is probably using forecasts from @CopernicusECMWF? ⬇️Atmospheric #NO2 forecast for 7 May at 11 pm
Madeline Lisaius retweeted
As promised, I am showing off some of the decolonial postcards that my students produced this semester. They are really, really cool!!
39
274
71
1,433
Madeline Lisaius retweeted
While $1.1 billion for a new sustainability school is a monumental investment, if #FossilFuel funding is being prioritized over #EnvironmentalJustice and advocacy, whose lives are going to benefit from the work it enables? nytimes.com/2022/05/04/clima…
3
6
Excited to see more investment into sustainability space and the official news of this long-awaited pivot! news.stanford.edu/?p=43684 via @Stanford