Kasia Kedzierska
LightNews LightNews
kasia.codes
Kasia Kedzierska
@kasia.codes
560 followers 140 following 9 posts
AI Research Scientist @alleninstitute.org | PhD from @ox.ac.uk | prev: intern @msftresearch.bsky.social New England & Novo Nordisk | she/her | kasia.codes
Posts Replies Media Videos
kasia.codes
✨ Brighten up your terminal! ✨

Who wouldn’t want to be greeted by a pony 🐎 or a cow 🐄 during in their terminal?! 🙈

I wrote a post on adding some fun and color to your console: kasia.codes/posts/pony_i...

#bash #zsh #ponysay #cowsay #TerminalFun
Screenshot of a terminal session showing a greeting message with the current date, time, and a piece of wisdom: 'You can't mend a wristwatch while falling from an airplane.' Below the text, there's a colorful ASCII art depiction of a pony with a rainbow mane and tail.
November 17, 2024 at 7:01 PM
kasia.codes
December's coming, and so is my favorite recent tradition: #AdventOfCode! 🎄👩‍💻 Excited to solve coding puzzles in the spirit of helping elves and saving holidays. Highly recommend! adventofcode.com
A festive holiday scene with a laptop displaying glowing code, surrounded by Christmas decorations like fairy lights, ornaments, a candy cane, and a small tree. Snowfall is visible through a window, creating a cozy atmosphere with subtle coding elements woven into the decor.
November 16, 2024 at 1:30 PM
kasia.codes
How robust are proposed foundation models in single-cell biology?

Here, we assessed two models: Geneformer and scGPT. We focus on evaluating their zero-shot capabilities.

📄 biorxiv.org/content/10.1...
💻 github.com/microsoft/ze...
💪 with Alex Lu, Ava Amini, and Lorin Crawford
This is a summary figure of the evaluation setup where two foundation models, Geneformer and scGPT, together with baselines such as no model (HVG) and scVI were evaluated on 5 datasets in 3 tasks for cell type clustering, batch integration, and pretraining objective. The figure illustrates the expected good and bad cases for each task. In cell type clustering a good outcome is when cell types separate in the embedding space, while bad is when they are intermixed. In batch integration, expected is the separation across cell types, not batches. In the pretraining objective, a positive correlation is expected between input and predicted values.
October 18, 2023 at 3:43 PM
kasia.codes
Yay, I passed my viva today! 🎓💪

I had to take a photo home, as thanks to Oxford's weather, I only dried up after the pub. :) Had great fun celebrating and will take my well-deserved time off from my thesis (before submitting the corrections, of course). #phdone #phdchat
photo of me holding flowers with congratulations balloons behind.
September 12, 2023 at 10:08 PM