UMAP does not reproduce high-dimensional similarities due to negative sampling.
Sebastian DamrichFred A. HamprechtPublished in: CoRR (2021)
Keyphrases
- high dimensional
- parameter space
- low dimensional
- positive and negative
- similarity search
- data points
- similarity measure
- high dimensionality
- random sampling
- variable selection
- multi modal
- high dimensional problems
- dimensionality reduction
- nearest neighbor
- dimension reduction
- small sample size
- sampling strategy
- information systems
- metric space
- multi dimensional
- monte carlo
- sample size
- high dimensional data
- microarray data
- training set
- data analysis
- pairwise
- machine learning
- sampling algorithm
- neural network