
I got the MNIST images rendering in the active learning prototype. White here are the images selected for labeling for the next round.

I got the MNIST images rendering in the active learning prototype. White here are the images selected for labeling for the next round.

Rasmus Andersson's favorite technical papers.
From: https://www.dropbox.com/sh/is0sy5350lr4v9j/AADQlhVSQcRw6vCNKQgGWelqa

First pass of active learning rounds on MNIST in the prototype. I need to think through what the transitions should be more, but this looks promising.

Oooo I bet there are some interesting things to be done with context.isPointinPath().
From: https://beta.observablehq.com/@mbostock/pixelated-world

Mushy: neural network generated isometric tiles by Everest Pipkin.

I'm making a notebook to visualize and figure out what frame and matting options I should choose.

I changed how I'm handling transparency thresholding so the blobs are a lot less muddy now.

I didn't think this all the way through and ran UMAP on each of these groups separately instead of at the same time.

I added T-SNE and UMAP with min_dist=0.8 algorithm options to the UMAP explorer. Three.js & tween.js animating the transition of 70,000 points no problem.

I put up a demo of the UMAP on MNIST interactive visualization I've been working on.

You can see how UMAP clusters are structured by things like the orientation of the number if you travel along their axes.