Freddie Kalaitzis

It took me a while, but I'm very excited about this one–that I've joined @AspiaSpace Watch this space.

The amount of backlash I get when I refuse to do free work (e.g. a keynote presentation, a "quick chat") is really shocking to me. I'm sorry, but why would I provide free labor to for-profit entities? (And don't say "exposure")

Very happy to say that I'll be kicking off Season 3 of this very promising podcast on the space industry, along with @OSA_projects James Parr (@FDL_AI/@trillium_tech) and Maral Bayaraa (@SatAppsCatapult). 2nd of May. Stay tuned.

@anndvision @Columbia @blei_lab Congrats Andrew. Great fit!

thrilled to share that I will be joining David Blei and Donald Green at @Columbia in July for a postdoc exploring scalable machine learning for field experimentation in the social sciences! looking forward to working with everyone at @blei_lab and learning from the NYC community

I gave a brief interview on @JACKfmOxford on our recent @NatureAstronomy paper w @SETIInstitute. You can listen to it here (starts at the 11:43 mark)

Catch this lot on The Scoop with @josummerbell tomorrow from 9am ... AI and the the search for life on Mars with @alkalait, @HelenAndDouglas's pop-up at a new mega charity shop in Reading and @drjanegregory on why people get angry about sounds like loud chewing!

👽 | A new study involving Oxford researchers has demonstrated that AI and machine learning methods can support the search for extraterrestrial life by identifying hidden patterns within geographical data. ⬇ ox.ac.uk/news/2023-03-2…

📢#xcube v1.0 released 🎉! Create, serve, view, analyse, and process geospatial data cubes. Builds on @xarray_dev, @dask_dev, compatible with @pangeo-data. Data stores (@CMEMS_EU, @esaclimate, @sentinel_hub, etc.), cube server (WMTS, STAC, S3), viewer. ➡️github.com/dcs4cop/xcube

A new study by an international team of researchers including @compscioxford's Freddie Kalaitzis @alkalait shows artificial intelligence could help locate life on Mars. It's been published @NatureAstronomy. Read more here: ox.ac.uk/news/2023-03-2… @UniofOxford @mplsoxford

Our recent @NatureAstronomy paper featured on @UniofOxford news ox.ac.uk/news/2023-03-2…

Using AI to Search for Life on Mars and Other Planets | Psychology Today Read psychologytoday.com/us/blog/the-fu… .@SETIInstitute and .@NASA use #AI #MachineLearning learning to search for #biosignatures beyond #Earth #innovation .@PsychToday #Mars #Venus #space #SpaceNews #microbes #SpaceExploration #astrobiology #future

Sr Research Fellow Freddie Kalaitzis @alkalait at @CompSciOxford, co-authored with a team at @SETIInstitute a study in @NatureAstronomy on a AI based framework to accelerate the search for life on Mars. Freddie is the key AI contributor in this study. nature.com/articles/s4155…

Kimberley Warren-Rhodes & colleagues present an adaptable framework that couples statistical ecology with deep learning to recognize and predict biosignature patterns. Such a framework might be applied to (e.g.) drone flight imagery of planetary surfaces. nature.com/articles/s4155…

I'm very proud that my collaboration w @NASAAmes & the @SETIInstitute is finally published in @NatureAstronomy We show a framework to search for life on another planet in a data-driven fashion, starting from orbit & all the way through to a rover's view nature.com/articles/s4155…

This work shows the first framework for searching for life on another planet in a systematic data-driven fashion based on good-ol' statistics and A.I. starting from an orbital vantage point and all the way through to a rover's camera.

@shasta721 It was a pleasure working with your team @shasta721

Variants a) Selection can be based on the max, or the soft-max of dot products. The latter is soft-attention. b) When A=B, this is self-attention. 3/n

5. We can match keys via the dot product with a "key estimate"—a QUERY vector 6. We can inform selections on objects of a set A based on the QUERY of an object from set B 2/n

Attention in 6 sentences 1. Each of N objects is encoded by a VALUE vector 2. Representing an object has different requirements to addressing it 3. To decouple the two tasks each object is addressed by a separate KEY vector 4. KEY matching is easier than selection 1/n

if the Earth's friekin core can take a break, then my brother you deserve one too!