We learnt about Dimensionality Reduction techniques on W1D5 (July 17) and found PCA often misses important features in the data as it is a linear algorithm! We looked at t-SNE, a non-linear approach which can extract these nonlinear features but we can’t use it for clustering analysis! What a coincidence that on the same day Iain Couzin at MPI AB put out a non-linear method like t-SNE called VAE-SNE that automatically figures out clusters in the data!
Read more here:
8 Likes
VAE stands for variational autoencoder, which is pretty much the last topic of NMA (W3D5). You can peek ahead at the content if you’d like, though I don’t think it’s fully finished yet.
1 Like