As I discussed in my previous blog post, the paper “Early Convolutions Help Transformers See Better” replaces the patchify operation in the original vision transformer paper with convolutions. I implemented this paper here.
[Read More]
A work friend I used to have weekly virtual meetings with to discuss machine learning papers messaged me a couple weeks ago saying “There’s another idea we had that turned out to be a good one,” along with a link to a paper that had been uploaded to arXiv the...
[Read More]
Here are some of the lessons I’ve learned so far in my career as a machine learning engineer and from doing machine learning projects. This is advice I’d try to impress upon my younger self if I could send a message back in time.
[Read More]