EVA AI Seminar: Spring Semester 2025 🤖🌸

In my last post, I introduced my semester topic - Harnessing Foundation Models for Medical Image Segmentation ⚕️. The title is a bit of a mouthful. And the planned work is a definite handful! At the moment, I'm scrambling around with a homemade U-Net implementation, trying for not-terrible results after training on a few small datasets. Eventually, I'll graduate to the evaluation of actual foundation models - like Segment Anything (SAM), SAM2, and fine-tuned variants, i.e., MedSAM - which boast parameter sets and pre-training samples several orders of magnitude larger than the pile of weights I've haphazardly thrown together. When I say evaluation, that implies a strong understanding of each model. I'll need to get a grasp on everything! From architecture & training procedure to the problems their design sought to solve, how well they did this, and what research avenues might need exploring to perform even better 🔎. If you & I both took a wild guess on a decen...

Intro & Motivation Hi there! 🫡 I'm Thamena. I'm taking the AI Seminar with you this semester and one of my most defining characteristics is how awful I am at introducing myself. There are many tasks at which AI challenges or even exceeds human performance, so this seems a prime opportunity to check whether stilted personal introductions count among them, right?!   Altogether, pretty good! Sometimes I'm slightly embarrassed to admit I had to be introduced to AI's potential by playing with mainstream toys like ChatGPT^ alongside majority of the general population during the ongoing boom. I was largely (and blissfully) ignorant prior. That means I only have about a year's worth of study experience in the field from beginning my Master's in early 2024; a timeline that also gives away my similarly limited scholarly experience!  That background (or lack thereof) is why I selected this course. I'd like a comprehensive idea of how the black box of academia wor...