A Decoder-Free Approach for Unsupervised Clustering and Manifold Learning with Random Triplet Mining

Nina, Oliver; Moody, Jamison; Milligan, Clarissa

doi:10.1109/iccvw.2019.00493

Cited by 11 publications

(5 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Quantitative comparison. We first compare the proposed DCCS with several baseline methods as well as other state-of-the-art clustering approaches based [33] 0.572 0.500 0.365 0.474 0.512 0.348 SC [40] 0.696 0.663 0.521 0.508 0.575 -AC [11] 0.695 0.609 0.481 0.500 0.564 0.371 NMF [3] 0.545 0.608 0.430 0.434 0.425 -DEC [36] 0.843 0.772 0.741 0.590 0.601 0.446 JULE [37] 0.964 0.913 0.927 0.563 0.608 -VaDE [18] 0.945 0.876 -0.578 0.630 -DEPICT [9] 0.965 0.917 -0.392 0.392 -IMSAT [16] 0.984 0.956 0.965 0.736 0.696 0.609 DAC [4] 0.978 0.935 0.949 0.615 0.632 0.502 SpectralNet [32] 0.971 0.924 0.936 0.533 0.552 -ClusterGAN [27] 0.950 0.890 0.890 0.630 0.640 0.500 DLS-Clustering [8] 0.975 0.936 -0.693 0.669 -DualAE [39] 0.978 0.941 -0.662 0.645 -RTM [29] 0.968 0.933 0.932 0.710 0.685 0.578 NCSC [41] 0.941 0.861 0.875 0.721 0.686 0.592 IIC [17] 0.992 0.978 0.983 0.657 0.637 0.523 DCCS (Proposed) 0.989 0.970 0.976 0.756 0.704 0.623 on deep learning, as shown in Table 2 and Table 3. DCCS outperforms all the other methods by large margins on Fashion-MNIST, CIFAR-10, STL-10 and ImageNet-10.…”

Section: Resultsmentioning

confidence: 99%

Deep Image Clustering with Category-Style Representation

Zhao

et al. 2020

Preprint

View full text Add to dashboard Cite

Deep clustering which adopts deep neural networks to obtain optimal representations for clustering has been widely studied recently. In this paper, we propose a novel deep image clustering framework to learn a category-style latent representation in which the category information is disentangled from image style and can be directly used as the cluster assignment. To achieve this goal, mutual information maximization is applied to embed relevant information in the latent representation. Moreover, augmentation-invariant loss is employed to disentangle the representation into category part and style part. Last but not least, a prior distribution is imposed on the latent representation to ensure the elements of the category vector can be used as the probabilities over clusters. Comprehensive experiments demonstrate that the proposed approach outperforms state-of-the-art methods significantly on five public datasets. ‡

show abstract

Section: Resultsmentioning

confidence: 99%

Deep Image Clustering with Category-Style Representation

Zhao

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…While we focus on embedding using variational autoencoders, an open direction for could involve embedding hierarchical structure using other representation learning methods [3,10,11,15,57,63,67,79,85]. Another direction is to better understand the similarities and differences between learned embeddings, comparison-based methods, and ordinal relations [22,28,37,39].…”

Section: Discussionmentioning

confidence: 99%

Unsupervised Embedding of Hierarchical Structure in Euclidean Space

Zhao,

Hao,

Rashtchian

2020

Preprint

View full text Add to dashboard Cite

Deep embedding methods have influenced many areas of unsupervised learning. However, the best methods for learning hierarchical structure use non-Euclidean representations, whereas Euclidean geometry underlies the theory behind many hierarchical clustering algorithms. To bridge the gap between these two areas, we consider learning a non-linear embedding of data into Euclidean space as a way to improve the hierarchical clustering produced by agglomerative algorithms. To learn the embedding, we revisit using a variational autoencoder with a Gaussian mixture prior, and we show that rescaling the latent space embedding and then applying Ward's linkage-based algorithm leads to improved results for both dendrogram purity and the Moseley-Wang cost function. Finally, we complement our empirical results with a theoretical explanation of the success of this approach. We study a synthetic model of the embedded vectors and prove that Ward's method exactly recovers the planted hierarchical clustering with high probability. * Jinyu Zhao and Yi Hao contributed equally.

show abstract

“…We aim to train the style embedding network such that the output embeddings of similar classes are close together. While Meshry et al [25] used triplet loss to train the network by selecting triplets using style distance metric, there has been a series of work exploring different triplet mining techniques and losses [26,27,28,29]. In this work, we proceed to train the style encoder by using Easy Positive Hard Negative triplet mining proposed by Xuan et al [29].…”

Section: Frameworkmentioning

confidence: 99%

Unsupervised Image to Image Translation for Multiple Retinal Pathology Synthesis in Optical Coherence Tomography Scans

Hemanth¹,

Girish²

2021

Preprint

View full text Add to dashboard Cite

Image to Image Translation (I2I) is a challenging computer vision problem used in numerous domains for multiple tasks. Recently, ophthalmology became one of the major fields where the application of I2I is increasing rapidly. One such application is the generation of synthetic retinal optical coherence tomographic (OCT) scans. Existing I2I methods require training of multiple models to translate images from normal scans to a specific pathology: limiting the use of these models due to their complexity. To address this issue, we propose an unsupervised multi-domain I2I network with pre-trained style encoder that translates retinal OCT images in one domain to multiple domains. We assume that the image splits into domain-invariant content and domain-specific style codes, and pre-train these style codes. The performed experiments show that the proposed model outperforms state-of-the-art models like MUNIT and CycleGAN synthesizing diverse pathological scans.

show abstract

A Decoder-Free Approach for Unsupervised Clustering and Manifold Learning with Random Triplet Mining

Cited by 11 publications

References 18 publications

Deep Image Clustering with Category-Style Representation

Deep Image Clustering with Category-Style Representation

Unsupervised Embedding of Hierarchical Structure in Euclidean Space

Unsupervised Image to Image Translation for Multiple Retinal Pathology Synthesis in Optical Coherence Tomography Scans

Contact Info

Product

Resources

About