Distill multi-dimensional data down to 2 dimensions
Common application is image compression. Black and white photograph image has 3 dimensions: x(width) y(height) and z(color). If you do that in a technique that preserve the variance in that image as well as possible you can still reconstruct the image without a lot of loss. Take data that contains four dimensions of data: X position, Y position, Z position, A position Distill that data down to two dimensions that can be useful for things like image compression and facial recogniotion, because it distills out the information that contributes most to the variance in the data set.