3 ms·
By and large yes: T-SNE does it's best to achieve that - it's kind of tricky to take an N-dimensional space and flattening it to 2 dimensions. You inevitably ge
by jacobn 6y ago
By and large yes: T-SNE does it's best to achieve that - it's kind of tricky to take an N-dimensional space and flattening it to 2 dimensions. You inevitably get some tearing and some scatter.
So they're not always next to each other, but oftentimes.
- sharemywin 6y agoThese two articles looked pretty interesting: https://towardsdatascience.com/compressionvae-a-powerful-and-versatile-alternative-to-t-sne-and-umap-5c50898b8696 https://towardsdatascience.com/compressionvae-a-powerful-and... https://medium.com/pinterest-engineering/detecting-image-similarity-using-spark-lsh-and-tensorflow-618636afc939 https://medium.com/pinterest-engineering/detecting-image-sim...