#t-sne

Question 1: Single choice

mediumAI-checkedDimensionality Reduction

You generate 2-D data from two Gaussian clusters, one with 10 times the spread of the other, and run t-SNE until the layout is stable. In the plot both clusters look about the same size. What is the best explanation?

t-SNE adapts distances to local density, expanding dense clusters and contracting sparse ones1
The perplexity was set too low, so random local clumping hides the clusters' true relative sizes2
The optimization stopped too early and would reveal the true sizes after more iterations3
Any reduction to fewer dimensions distorts distances, and this is that unavoidable information loss4

Timed test