Skip to content

wangyi-fudan/npca

Repository files navigation

Neural Principal Component Analysis

Anuran_Calls

Bike_Sharing

mnist test

The Model

It is based on the following neural network:

data => 2 PCs => 32 => 32 => data

The two 32-node layers act as nonlinear "decoder/intereptor" of the 2 PCs. But the 2 PCs are just linear projection of raw data. Thus the PCs have similar meanning with conventional PC, but due to the nonlinear intereptation, it can hold rich structural information of the raw data.

We can simply replace PCA with npca!

About

Neural Principal Component Analysis

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors