When using t-SNE, one is trying to represent hyperdimensional data in two dimensions. Ideally, the geometry or distances between data points are preserved in the new, two-dimensional space. With t-SNE, what does this algorithm try to preserve? How is this ...