Profile avatar
variint.bsky.social
Mechanistic interpretability of modular convnets applied to 👁️ at U Miami and 🛰️🐝 at Boku Vienna | she/her 🦒💕 variint.github.io
105 posts 1,901 followers 758 following
Regular Contributor
Active Commenter

Day 32: I promise, I am an expert in deep learning. #100DaysOfCode

Day 31: Searching for some benchmarking datasets and metrics! #100DaysOfCode

Day 30: A lot of room for improvement (the graph is not making too much sense from a medical point of view), but I made Named Entity Recognition with PUBTATOR3 work! #100DaysOfCode

Day 29: A lot of javascript these past days!! Also, learned more about parallax scrolling. #100DaysOfCode

Day 28: Creating demo ipynbs for my repo! #100DaysOfCode

Day 27: Quiz is (pretty much) done and online! Link may change at some point though ... nodes-and-edges.github.io/hatched/eyeq...

Had a Photoshop-Illustrator session and created some eye graphics for my anatomy javascript quiz!

Day 26: Working on a javascript quiz app to study for an eye-anatomy exam I have to redo cause I failed. 🙃 #100DaysOfCode

Day 25: Definition of experiments for my first PhD publication are in progress! Focusing on remote sensing image analysis for bee colony loss. 🐝 #100DaysOfCode

Day 24: Had a session with my lovely senv-team on weather forecast, climate cycle prediction and growing degree days! #100DaysOfCode

Day 23: Yesterday, I was working a bit more on baseline models! Also, progress with getting data! #100DaysOfCode

Day 22: More name entity recognition (NER) coding and additionally baseline brainstorming for the sparse vision model! #100DaysOfCode

Day 21: After getting quite reasonable results with BERN2, I am working with the pubtator3-api today to get some more/different json files with annotated diseases. It's a work in progress. #100DaysOfCode

Day 20: Used different Named-Entity Recognition (NER) and Relation Extraction (RE) codes and this is the current result!! We are getting somewhere! #100DaysOfCode

Day 19: Today I have been playing around with the code provided by a tutorial on knowledge graphs generated with NLP: "Building a Knowledge Base from Texts" - room for improvement, but good for a start! #100DaysOfCode www.nlplanet.org/course-pract...

What a day, eye anatomy exam, microbio presentation, biochemistry exam, now studying for pathology until my Spanish class starts in an hour! Doing a great job today with staying away from home that also hosts a screaming and swearing flatmate. Best thing - I am finally moving out on Sunday 🙃😊

Calling my mom due to antibiotic side effects - her on the phone: the only reason why I wake up early for work is cause I get to see the sunrise. Also, there will be snow tomorrow. Best distraction @gabiett.bsky.social ❄️😂

Day 18: The baseline is just not working. I will take a nap now. #100DaysOfCode

Day 17: Saturday-update - Working on a baseline, favourite candidate at the moment is SqueezeNet! #100DaysOfCode

Day 16: Thanksgiving as a non-American away from family made me bored enough to deal with tax, fixing a bank account that has been inactive since 2017 and I did code for a prof here - plotting chromosomes from bw files (and it works)! Anyways, hope everyone enjoyed the holidays!! 🐤#100DaysOfCode

Day 15: A lot of javascript today! I have been working on a quiz function for an in-progress blog. #100DaysOfCode

Day 14: Yesterday, I was working on the dataloader - I added some lines of code to save the dataset size. Also made some progress with reviewing possible evaluation techniques for the modular network! #100DaysofCode

@ramanshsharma.bsky.social - now that you are here, bluesky feels complete 😉🦋

Day 13: Started training a model and the loss finally seems to do what it's supposed to do!! #100DaysOfCode

Day 12: Optimiser problems are solved, my |CI-CC| loss term is changing in each step now!! Really happy about that 😀 #100DaysOfCode

Day 11: Gave up on the optimiser - I am doing it the "proper" way now without all the workaround-callbacks that I didn't manage to get to work. Hence - everytime I prune the model, I just create a new trainer with a new optimiser. Seems easier and cleaner after all. #PytorchLightning #100DaysOfCode

Day 10: A day or two ago I realised that my replaced nodes are not updated. Aka something is wrong with the optimiser. No idea how to resolve this (yet), but keep going, keep growing!! #100DaysOfCode

me: omg I feel so stupid, every result I ever created was wrong @mateosuffern.bsky.social [excited]: oh, so the loss plots will work now?? me: don't remind me, they are the one thing that were correct hence wrong for another reason 🙃