UMAP: Uniform Manifold Approximation and Projection
By Leland McInnes, John Healy, Nathaniel Saul, Lukas Großberger. 2018
7,959 citations recorded by the source corpus.
Uniform Manifold Approximation and Projection (UMAP) is a dimension reduction technique that can be used for visualisation similarly to t-SNE, but also for general non-linear dimension reduction. UMAP has a rigorous mathematical foundation, but is simple to use, with a scikit-learn compatible API. UMAP is among the fastest manifold learning implementations available – significantly faster than most t-SNE implementations.
Subfields: Computational Mechanics, Computer Graphics and Computer-Aided Design, Mechanical Engineering, Representation & Self-Supervised Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computational Mechanics
- Marta Ruiz Costa-jussà
- Aram Galstyan
- Ali Farhadi
- Mohammad Rostami
- Qing Nie
- Aviv Regev
- Carlos Escolano
- Andreas Krause
In Computer Graphics and Computer-Aided Design
- Marta Ruiz Costa-jussà
- Aram Galstyan
- Ali Farhadi
- Mohammad Rostami
- Qing Nie
- Carlos Escolano
- Andreas Krause
- Min Sun
In Mechanical Engineering
- Marta Ruiz Costa-jussà
- Aram Galstyan
- Mohammad Rostami
- Qing Nie
- Carlos Escolano
- Andreas Krause
- José A. R. Fonollosa
- Richard S. Zemel
In Representation & Self-Supervised Learning
- Herman Kamper
- James R. Glass
- Myle Ott
- Kyunghyun Cho
- Mohammad Rostami
- Mengye Ren
- Thibault Févry
- Jason Phang
Citations in the corpus
- Evaluating the Underlying Gender Bias in Contextualized Word Embeddings (2019, methodology): “The clusters (for one particular experiment out of the 10 of them) are shown in Figure 2 after applying UMAP (McInnes et al., 2018; McInnes et al., 2018) to the contextualized embeddings.”
- openTSNE: a modular Python library for t-SNE dimensionality reduction and embedding (2019, methodology): “Popular approaches for dimensionality reduction include principal component analysis, multidimensional scaling, and uniform manifold approximation and projections (1).”
- Unsupervised Model Adaptation for Continual Semantic Segmentation (2020, methodology): “To demonstrate that our solution implements what we anticipated, we have used UMAP (McInnes et al. 2018) visualization tool to reduce the dimension of the data representations in the embedding space to two for 2D visualization.”
- Unsupervised label noise modeling and loss correction (2019, methodology): “UMAP (McInnes et al., 2018) embeddings for training (top) with 80% of label noise and validation (bottom) on CIFAR-10 with (a)(d) cross-entropy loss from Eq.”
- Mix-review: Alleviate Forgetting in the Pretrain-Finetune Framework for Neural Language Generation Models (2019, positive): “We use the default hyperparameter setting of the python implementation of UMAP.”
- Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression (2019, methodology): “For 2D visualization, the PC embeddings were passed into UMAP [36, 37] with default parameters.”
- TextSETTR: Label-Free Text Style Extraction and Tunable Targeted Restyling (2021, methodology): “Figure 4 plots a 2D UMAP dimensionality reduction (McInnes et al., 2018) of the vectors, and shows clear separation along both sentiment and product category.6 6We sub-sample to 3,000 points after dimensionality reduction for clearer visualization.”
- Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer Screening (2019, methodology): “In Figure 4, we visualize two sets of activations for each exam by embedding them into a two-dimensional space using UMAP (McInnes et al., 2018): concatenated activations from the last layer of each of the four image-specific columns (left), and concatenated activations from the first fully connecte”