Self-organizing map
A self-organizing map (SOM) is an unsupervised learning algorithm that can be used for dimensionality reduction.
Details
A simple start: training a SOM to project R3 inputs down to R2. The inputs represent (R,G,B) color tuples.
A little more advanced: training a SOM to project R4 inputs down to R3. The inputs represent (R,G,B,radius) tuples.
Experimenting with visualizing techniques using vectors.
Projecting high-dimensional inputs (icons) down to R2. Here, the trained SOM grid is being visualized along with the training samples superimposed on top.
Projecting high-dimensional inputs (icons) down to R2. I came up with a stacking visualization technique to handle the cases where multiple input vectors project down to the same output.
Experimenting with visualizing the input vectors as displaced, colorized meshes.
Projecting high-dimensional inputs (album cover images) down to R3.
Projecting (R,G,B,radius) tuples down to (x,y). I am using a similar stacking technique here to visualize the results.
Projecting (left eye radius,right eye radius,mouth radius) tuples down to (x,y).
