Spatially-Coherent Segmentation Using Hierarchical Gaussian Mixture Reduction Based on Cauchy-Schwarz Divergence

Nouboukpo, Adama; Állili, Madjid

doi:10.1007/978-3-030-27202-9_35

Cited by 2 publications

(2 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While the former are usually trained upon a small amount of labelled data and a very large amount of unlabelled data (Van Engelen and Hoos 2019), the latter try to group data by incorporating side information from domain or user knowledge (Boulmerka and Allili 2018). Side information usually comes in the form of pairwise constraints (must-links and cannot-links) between samples of data (Shental et al 2004), which can be directly observed or inferred as background knowledge from user feedback (Nouboukpo and Allili 2019). The must-link establishes the samples which must be in the same cluster (or class) and the cannot-link refers to those samples that cannot be in the same cluster (or class).…”

Section: Introductionmentioning

confidence: 99%

“…The supervision can come in the form of pairwise or group relationships as well as partially labelled data. The group constraints and the number of data within each group can be application-driven or generated automatically by initially clustering the data into a large number of groups (Boulmerka, Allili, and Ait-Aoudia 2014;Nouboukpo and Allili 2019). The group constraints are defined at two levels.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Weakly Semi Supervised learning based Mixture Model With Two-Level Constraints

Nouboukpo

Állili

2021

FLAIRS

Self Cite

View full text Add to dashboard Cite

We propose a new weakly supervised approach for classification and clustering based on mixture models. Ourapproach integrates multi-level pairwise group and classconstraints between samples to learn the underlyinggroup structure of the data and propagate (scarce) initial labels to unlabelled data. Our algorithm assumes thenumber of classes is known but does not assume anyprior knowledge about the number of mixture components in each class. Therefore, our model : (1) allocatesmultiple mixture components to individual classes, (2)estimates automatically the number of components ofeach class, 3) propagates class labels to unlabelled datain a consistent way to predefined constraints. Experiments on several real-world and synthetic data datasetsshow the robustness and performance of our model overstate-of-the-art methods.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%