Using the paper Goffinet et al., 2021 and birdsong data, I explain how Variational Autoencoders and UMAP works. I also explore why UMAP is good at preserving the global structure of high-dimensional data.
Slides available at:
Original paper:
Jack GoffinetSamuel BrudnerRichard MooneyJohn Pearson (2021) Low-dimensional learned feature spaces quantify individual and group differences in vocal repertoires eLife 10:e67855.
Other useful resources:
Coenen, A., & Pearce, A. (n.d.). Understanding UMAP.
Kobak, D. (Director). (2021, April 13). Manifold learning and t-SNE. https://www.youtube.com/watch?v=MnRsk...
Oskolkov, N. (2021, March 10). How Exactly UMAP Works. Medium. https://towardsdatascience.com/how-ex...
Sainburg, T., Thielk, M., & Gentner, T. Q. (2020). Finding, visualizing, and quantifying latent structure across diverse animal vocal repertoires
Andreas Geiger: Variational Autoencoders YouTube Series
Other references see in slides
Comments