DIMAL: Deep Isometric Manifold Learning Using Sparse Geodesic Sampling

Pai, Gautam; Talmon, Ronen; Bronstein, Alex; Kimmel, Ron

doi:10.1109/wacv.2019.00092

Cited by 26 publications

(18 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To overcome the introduction of overlaps with sequential chunking, we developed a seeded-chunking method (as shown in Algorithm 2). The seeded-chunking first uses the “farthest point sampling” algorithm [16, 3] to find l seeds for l chunks, such that the seeds are farthest away from each other. That is, starting from a randomly selected seed, the seeds are chosen one at a time, such that each new seed has the largest distance to the set of already selected seeds (i.e.…”

Section: Methodsmentioning

confidence: 99%

Practical selection of representative sets of RNA-seq samples using a hierarchical approach

Tung

Kingsford

2021

Preprint

View full text Add to dashboard Cite

Despite numerous RNA-seq samples available at large databases, most RNA-seq analysis tools are evaluated on a limited number of RNA-seq samples. This drives a need for methods to select a representative subset from all available RNA-seq samples to facilitate comprehensive, unbiased evaluation of bioinformatics tools. In sequence-based approaches for representative set selection (e.g. a k-mer counting approach that selects a subset based on k-mer similarities between RNA-seq samples), because of the huge number of available RNA-seq samples and the large number of k-mers/sequences in each sample, computing the full similarity matrix between all samples using k-mers/sequences for the entire set of RNA-seq samples in a large database (e.g. the SRA) has memory and runtime challenges, making direct representative set selection infeasible with limited computing resources. Therefore, we developed a novel computational method called "hierarchical representative set selection" to handle this challenge. Hierarchical representative set selection is a divide-and-conquer-like algorithm that breaks the representative set selection into sub-selections and hierarchically selects representative samples through multiple levels. We demonstrate that hierarchical representative set selection can achieve performance close to that of direct representative set selection, while largely reducing the runtime and memory requirements of computing the full similarity matrix (up to 8.4X runtime reduction and 4.7X memory reduction for 10000 samples that could be practically run with direct subset selection). We show that hierarchical representative set selection substantially outperforms random sampling on the entire SRA set of RNA-seq samples, making it a practical solution to representative set selection on large databases such as the SRA.

show abstract

Section: Methodsmentioning

confidence: 99%

Practical selection of representative sets of RNA-seq samples using a hierarchical approach

Tung

Kingsford

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…To overcome the introduction of overlaps with sequential chunking, we developed a novel seeded-chunking method (as shown in Algorithm 2 ). The seeded-chunking first uses the ‘farthest point sampling’ algorithm ( Bronstein et al , 2009 ; Pai et al, 2019 ) to find l seeds for l chunks, such that the seeds are farthest away from each other. That is, starting from a randomly selected seed, the seeds are chosen one at a time, such that each new seed has the largest distance to the set of already selected seeds (i.e.…”

Section: Methodsmentioning

confidence: 99%

Practical selection of representative sets of RNA-seq samples using a hierarchical approach

Tung

Kingsford

2021

Bioinformatics

View full text Add to dashboard Cite

Motivation Despite numerous RNA-seq samples available at large databases, most RNA-seq analysis tools are evaluated on a limited number of RNA-seq samples. This drives a need for methods to select a representative subset from all available RNA-seq samples to facilitate comprehensive, unbiased evaluation of bioinformatics tools. In sequence-based approaches for representative set selection (e.g. a k-mer counting approach that selects a subset based on k-mer similarities between RNA-seq samples), because of the large numbers of available RNA-seq samples and of k-mers/sequences in each sample, computing the full similarity matrix using k-mers/sequences for the entire set of RNA-seq samples in a large database (e.g. the SRA) has memory and runtime challenges; this makes direct representative set selection infeasible with limited computing resources. Results We developed a novel computational method called ‘hierarchical representative set selection’ to handle this challenge. Hierarchical representative set selection is a divide-and-conquer-like algorithm that breaks representative set selection into sub-selections and hierarchically selects representative samples through multiple levels. We demonstrate that hierarchical representative set selection can achieve summarization quality close to that of direct representative set selection, while largely reducing runtime and memory requirements of computing the full similarity matrix (up to 8.4× runtime reduction and 5.35× memory reduction for 10 000 and 12 000 samples respectively that could be practically run with direct subset selection). We show that hierarchical representative set selection substantially outperforms random sampling on the entire SRA set of RNA-seq samples, making it a practical solution to representative set selection on large databases like the SRA. Availability and implementation The code is available at https://github.com/Kingsford-Group/hierrepsetselection and https://github.com/Kingsford-Group/jellyfishsim. Supplementary information Supplementary data are available at Bioinformatics online.

show abstract

“…Finally, it is possible to use a nonlinear parametric model for the embedding, f Θ (x) : R m → R k , where Θ is a set of model parameters, whose number is much smaller than n. For example, [27] finds such a parametric mapping by training a neural network to minimize the following loss:…”

Section: Subspace Methodsmentioning

confidence: 99%

“…(Image © taken from [13] with permission of the authors). (d) A three-dimensional embedding of the camera pose manifold (images of the same object taken with different elevation and azimuth) from a small subset of distances using a machine learning-based nonlinear parametric MDS model [27]. (Image courtesy of the authors) rank order of dissimilarities such as correspondence problems between non-isometric shapes.…”

Section: Applicationsmentioning

confidence: 99%