Deep unsupervised feature selection by discarding nuisance and correlated features

Shaham, Uri; Lindenbaum, Ofir; Svirsky, Jonathan; Kluger, Yuval

doi:10.1016/j.neunet.2022.04.002

Cited by 12 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Clustering is inherently unstable, especially when dealing with many classes or highdimensional datasets. Several authors have proposed using feature selection (Solorio-Fernández et al, 2020;Shaham et al, 2022;Lindenbaum et al, 2021) to improve clustering capabilities by removing nuisance features in tabular data. We are interested in stabilizing clustering performance on diverse high-dimensional image data.…”

Section: Multiple Clustering Headsmentioning

confidence: 99%

Domain-Generalizable Multiple-Domain Clustering

Rozner¹,

Battash²,

Wolf³

et al. 2023

Preprint

View full text Add to dashboard Cite

Accurately clustering high-dimensional measurements is vital for adequately analyzing scientific data. Deep learning machinery has remarkably improved clustering capabilities in recent years due to its ability to extract meaningful representations. In this work, we are given unlabeled samples from multiple source domains, and we aim to learn a shared classifier that assigns the examples to various clusters. Evaluation is done by using the classifier for predicting cluster assignments in a previously unseen domain. This setting generalizes the problem of unsupervised domain generalization to the case in which no supervised learning samples are given (completely unsupervised). Towards this goal, we present an end-toend model and evaluate its capabilities on several multi-domain image datasets. Specifically, we demonstrate that our model is more accurate than schemes that require fine-tuning using samples from the target domain or some level of supervision.

show abstract

Section: Multiple Clustering Headsmentioning

confidence: 99%

Domain-Generalizable Multiple-Domain Clustering

Rozner¹,

Battash²,

Wolf³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…These 16 features were clustered by spectral clustering into 26 clusters 35 . In order to add new data to the spectral net clusters, we utilized Spectral Neural Net to learn the graph Laplacian transform for the dataset 36 . Newly generated helix backbones can now be clustered and compared to the original reference clusters.…”

Section: Generator Evaluationmentioning

confidence: 99%

A Generative Model for Creating Path Delineated Helical Proteins

Woodall

Kibler

Wicky

et al. 2023

Preprint

View full text Add to dashboard Cite

Engineered proteins with precisely defined shapes can scaffold functional protein domains in 3D space to fine-tune their functions, such as the regulation of cellular signaling by ligand positioning or the design of self-assembling protein materials with specific forms. Methods for simply and efficiently generating the protein backbones to initiate these design processes remain limited. In this work, we develop a lightweight neural network to guide helix fragment assembly along a guideline using a GAN architecture and show that this approach can rapidly generate viable samples while being computationally inexpensive. Key to our approach is the transformation of the input structural data used for training into a parametric representation of helices to reduce the generator network size, which in turn facilitates rapid backpropagation to find specific helical arrangements during generation. This approach provides a method to quickly generate helical protein scaffolds.

show abstract

Deep Feature selection

Ros,

Riad

2023

Unsupervised and Semi-Supervised Learning

View full text Add to dashboard Cite

Deep unsupervised feature selection by discarding nuisance and correlated features

Cited by 12 publications

References 11 publications

Domain-Generalizable Multiple-Domain Clustering

Domain-Generalizable Multiple-Domain Clustering

A Generative Model for Creating Path Delineated Helical Proteins

Deep Feature selection

Contact Info

Product

Resources

About