A classification-based approach to semi-supervised clustering with pairwise constraints

Śmieja, Marek; Struski, Łukasz; Figueiredo, Mário A. T.

doi:10.1016/j.neunet.2020.04.017

Cited by 31 publications

(10 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We used a semi-supervised siamese neural network (see Supplementary Figs. 2 and 3) (Śmieja et al, 2020) dealing with the must-link and cannot-link constraints. The network has two siamese neural networks with shared weights.…”

Section: Methodsmentioning

confidence: 99%

“…Here, we introduce a semi-supervised binning method, SemiBin (Semi-supervised metagenomic Binning), based on deep contrastive learning to take advantage of must-link and cannot-link constraints (Śmieja et al, 2020) (see Fig. 1a and Supplementary Fig.…”

Section: Mainmentioning

confidence: 99%

“…In particular, while cannot-link constraints could be robustly derived from contig annotations to the GTDB (Parks et al, 2020), must-link constraints are instead generated by breaking up longer contigs artificially (see Supplementary Text). The constraints serve as input to a semi-supervised siamese neural network (Śmieja et al, 2020) for feature embedding. SemiBin then generates a sparse graph from the embedded features and groups the contigs into bins with the Infomap (Rosvall and Bergstrom, 2008) community detection algorithm (see Methods).…”

Section: Mainmentioning

confidence: 99%

See 2 more Smart Citations

SemiBin: Incorporating information from reference genomes with semi-supervised deep learning leads to better metagenomic assembled genomes (MAGs)

Pan

Zhu

Zhao

et al. 2021

Preprint

View full text Add to dashboard Cite

Metagenomic binning is the step in building metagenome-assembled genomes (MAGs) when sequences predicted to originate from the same genome are automatically grouped together. The most widely-used methods for binning are reference-independent, operating de novo and allow the recovery of genomes from previously unsampled clades. However, they do not leverage the knowledge in existing databases. Here, we propose SemiBin, an open source tool that uses neural networks to implement a semi-supervised approach, i.e. SemiBin exploits the information in reference genomes, while retaining the capability of binning genomes that are outside the reference dataset. SemiBin outperforms existing state-of-the-art binning methods in simulated and real microbiome datasets across three different environments (human gut, dog gut, and marine microbiomes). SemiBin returns more high-quality bins with larger taxonomic diversity, including more distinct genera and species. SemiBin is available as open source software at https://github.com/BigDataBiology/SemiBin/.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Mainmentioning

confidence: 99%

Section: Mainmentioning

confidence: 99%

See 1 more Smart Citation

SemiBin: Incorporating information from reference genomes with semi-supervised deep learning leads to better metagenomic assembled genomes (MAGs)

Pan

Zhu

Zhao

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Śmieja 51 introduced a neural network framework for semi‐supervised clustering with pairwise (must‐link or cannot‐link) constraints to improve the classification. In contrast to existing approaches, this method decomposes the semi‐supervised clustering into two simpler classification tasks.…”

Section: Related Workmentioning

confidence: 99%

Deep semi‐supervised classification based in deep clustering and cross‐entropy

et al. 2021

View full text Add to dashboard Cite

Self‐labeled techniques, a semi‐supervised classification paradigm (SSC), are highly effective in alleviating the scarcity of labeled data used in classification tasks through an iterative process of self‐training. This problem was addressed by several approaches with different assumptions about the features of the input data, examples of these approaches being self‐training, co‐training, STRED, among others. This paper presents a framework for data self‐labeling based on deep autoencoder combined with a self‐labeled technique that takes into consideration cross‐entropy. The model uses the Encoder to reduce the dimensionality of the input that is submitted to a labeling layer. The weights of this layer are adjusted through the learning from a clustering performed in the Z space, which is the reduced dimensionality space. Results showed that the proposed method obtained competitive performance in relation to classic methods that are found in the literature.

show abstract

“…But, the class labels provide no information about must-link pairs because a class might have distinct subclasses. Most of the existing deep semi-supervised clustering methods[73][74][75][76][77] requires both must-link and cannot-link information to function.In general, a good embedding space should learn the similarity structures among samples. From a supervised learning perspective, classes should be well separated from each other.…”

mentioning

confidence: 99%

Multi-scale Deep Nearest Neighbors

Chauhan¹

View full text Add to dashboard Cite

The performance of the k-NN classifier is dependent on its input space. The predictive power of the k-NN classifier decreases in high dimensional spaces due to the curse of dimensionality where all pairs of samples have almost the same distance.k-NN has been used in a variety of applications in different areas such as computer vision, natural language processing, and recommender systems. Many approaches in the literature try to improve the input space of k-NN. These approaches include manually extracting a low dimensional feature space, or learning a mapping into a low-dimensional space. The mapping of the input space to a low-dimensional space is learned through linear or non-linear transformation.In this thesis, we aim to learn a deep embedding space suitable for k-NN. Our approach is based on minimizing the leave-one-out 1-NN classification error in the embedding space. That is, if we hold out a training sample, it should be correctly classified using 1-NN applied to the rest of the training data. Directly optimizing for such a rule is not tractable due to its discontinuous nature. We propose Multi-scale Deep Nearest Neighbor (MsDNN) which is a differentiable loss function that aims to maximize the expected sample margin for every training sample.The output of MsDNN is an embedding space. We evaluate the resulting space from two angles. From the classification view, during testing, we run a k-NN classifier and report the classification accuracy. But classification accuracy does not tell us v Table of Contents vi List of Tables ix List of Figures x

show abstract

A classification-based approach to semi-supervised clustering with pairwise constraints

Cited by 31 publications

References 40 publications

SemiBin: Incorporating information from reference genomes with semi-supervised deep learning leads to better metagenomic assembled genomes (MAGs)

SemiBin: Incorporating information from reference genomes with semi-supervised deep learning leads to better metagenomic assembled genomes (MAGs)

Deep semi‐supervised classification based in deep clustering and cross‐entropy

Multi-scale Deep Nearest Neighbors

Contact Info

Product

Resources

About