Walter Montalvo @WalterMontalvo@qoto.org
Today I presented "Weight Agnostic Neural Networks" https://weightagnostic.github.io/ in our reading group. There are many things I like about this paper! The main idea is to optimize neural architectures so that they perform well for a range of weights.
1/?
A thing I plotted while trying to get all my rotations to go in the right directions.
For those interested, it is tracked points on a fruit fly, all aligned by the thorax -- the big v shape at the origin. The head of the fly is the ^ that is mostly pointing up. Most of the other lines show the flies' legs, which go all over.
@WalterMontalvo Yes, we track a bunch of points from videos of the flies using deep learning. This is some data from a dataset we developed last year: https://www.aicrowd.com/challenges/multi-agent-behavior-challenge-2022/problems/mabe-2022-fruit-fly-groups
Here's a link to the tracking software we're developing, if you're interested:
http://kristinbranson.github.io/APT/
@kristinmbranson I am looking through the APT post and wow thats a lot of work done. It seems that you can train APT to recognize and track different parts of an anatomy. Could it be used to track the movement of cells or bacteria?
@kristinmbranson So you are tracking the tips of the legs, is that right? I am wondering how you are doing the tracking.
