Learning a Parametric Embedding by Preserving Local Structure

By L. V. D. van der Maaten. 2009

620 citations recorded by the source corpus.

Subfields: 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 Representation & Self-Supervised Learning

Citations in the corpus

Adversarial Autoencoders (2015, methodology): “Different methods such as parametric t-SNE [Maaten, 2009] have been proposed to address this issue.”
Representation Learning: A Review and New Perspectives (2012, background): “Note that this estimator can have high variance and that it becomes less reliable (variance becomes too large) as the model becomes more interesting, with larger weights, more non-linearity, sharper modes and a sharper probability density function (see our previous discussion in Section 9.4).”
Financial Time Series Forecasting with Deep Learning : A Systematic Literature Review: 2005-2019 (2019, methodology): “In the CD algorithm, the Kullback Leibler Divergence (KL-Divergence) algorithm is used to measure the distance between its reconstructed probability distribution and the original probability distribution of the input [71].”
Interpreting Deep Classifier by Visual Distillation of Dark Knowledge (2018, methodology): “Motivated by van der Maaten (2009), we suggest one can train a map f p from P T ( ·| x i ) to y i using a neural network.”
The art of using t-SNE for single-cell transcriptomics (2018, methodology): “One important example is parametric t-SNE, where a neural network is used to create a mapping f ð x Þ from high-dimensional input space to two dimensions and is trained using standard deep learning techniques to yield an optimal t-SNE result 42 .”
Joint Unsupervised Learning of Deep Representations and Image Clusters (2016, methodology): “However, in our method, we employ a weighted triplet loss that directly takes the local structure of embedded data into account; 2) In [38], the authors need to pre-train a stack of RBMs layer-by-layer, and then ﬁne-tune the neural network.”
Deep learning multidimensional projections (2019, background): “Yet, tSNE comes with some downsides: it is very slow for large datasets (thousands of observations or more), due to its quadratic nature; its parameters can be tricky to get right, which can lead to unpredictable results;(7) and it lacks the capability of projecting out-ofsample data, which is usefu”
Interpretable dimensionality reduction of single cell transcriptome data with deep generative models (2017, methodology): “The most similar approach for scvis may be the parametric t-SNE algorithm 51 , which uses a neural network to learn a parametric mapping from the high-dimensional space to a low dimension.”
Deep Adaptive Image Clustering (2017, methodology): “Inspired by the parametric t-SNE [28], Xie et al.”
Parametric UMAP Embeddings for Representation and Semisupervised Learning (2020, background): “We note, however, that the poor performance of Parametric t-SNE may reflect setting the degrees-of-freedom (α) at d − 1, which is only one of three parameterization schemes that van der Maaten (2009) suggests.”
Stochastic k-Neighborhood Selection for Supervised and Unsupervised Learning (2013, background): “This provides SNE with more “effective space” in the low-dimensional space, since it can now place far points arbitrarily far. t-SNE (van der Maaten, 2009) alters the definition of qi(j) even more drastically, to qi(j) ∝ (1 + ‖zi − zj‖2)−1, which tends to work even better in practice due to the…”
Stochastic triplet embedding (2012, methodology): “Following [15], the number of degrees of freedom α of t-STE was set to r − 1.”
Categorical Analysis of Human T Cell Heterogeneity with One-Dimensional Soli-Expression by Nonlinear Stochastic Embedding (2016, background): “In particular, the function that t -SNE minimizes is invariant under rotations of the low-dimensional map, which implies that t -SNE visualizations can be arbitrarily rotated (15).”
Dynamic Manifold Warping for view invariant action recognition (2011, background): “As a straightforward association, constructing A is related to unsupervised metric learning [22].”