Filters








5,526 Hits in 7.3 sec

Towards Better Understanding of Disentangled Representations via Mutual Information [article]

Xiaojiang Yang, Wendong Bi, Yitong Sun, Yu Cheng, Junchi Yan
2020 arXiv   pre-print
Together with the widely accepted independence assumption, we further bridge it with the conditional independence of factors in representations conditioned on data.  ...  In particular, we formulate such a relation through the concept of mutual information: the mutual information between each factor of the disentangled representations and data should be invariant conditioned  ...  This theorem bridges disentanglement with conditional independence and marginal independence of factors in representations.  ... 
arXiv:1911.10922v3 fatcat:mxqkpt4mkrbahpe6zaf2omnfhq

Learning Disentangled Representations for Time Series [article]

Yuening Li, Zhengzhang Chen, Daochen Zha, Mengnan Du, Denghui Zhang, Haifeng Chen, Xia Hu
2021 arXiv   pre-print
To bridge the gap, we propose Disentangle Time Series (DTS), a novel disentanglement enhancement framework for sequential data.  ...  Time-series representation learning is a fundamental task for time-series analysis.  ...  For example, learning interpretable and semantic-rich representations is useful for decomposing the ECG into cardiac cycles with recognizable phases as independent factors.  ... 
arXiv:2105.08179v2 fatcat:dyvhspxhtfd55mdi2kktfzlasi

Instance-Invariant Domain Adaptive Object Detection via Progressive Disentanglement [article]

Aming Wu, Yahong Han, Linchao Zhu, Yi Yang
2021 arXiv   pre-print
Particularly, base on disentangled learning used for feature decomposition, we devise two disentangled layers to decompose domain-invariant and domain-specific features.  ...  Most state-of-the-art methods of object detection suffer from poor generalization ability when the training and test data are from different domains, e.g., with different styles.  ...  Particularly, compared with the 'O-Base' and 'P-Base' used for disentanglement, the learned DIR and DSR separately contain much stronger object-relevant information and domainspecific information.  ... 
arXiv:1911.08712v4 fatcat:npgd5klwvrdz5jkue6zaqyntpe

A New Attention Mechanism to Classify Multivariate Time Series

Yifan Hao, Huiping Cao
2020 Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence  
CA-SFCN is compared with 16 approaches using 14 different MTS datasets.  ...  First, we introduce a temporal attention mechanism to extract long- and short-term memories across all time steps.  ...  Acknowledgments We thank IJCAI 2020 reviewers for their constructive feedback. This work is supported by the US National Science Foundation (NSF) under grants IIS-1652107 and IIS-1763620.  ... 
doi:10.24963/ijcai.2020/273 dblp:conf/ijcai/BeyazitTYT020 fatcat:zs672cmcgjb7dnh7rgjphed6ii

Unsupervised Disentanglement of Linear-Encoded Facial Semantics [article]

Yutong Zheng, Yu-Kai Huang, Ran Tao, Zhiqiang Shen, Marios Savvides
2021 arXiv   pre-print
Finally, we provide an analysis of our learned localized facial representations and illustrate that the semantic information is encoded, which surprisingly complies with human intuition.  ...  The method derives from linear regression and sparse representation learning concepts to make the disentangled latent representations easily interpreted as well.  ...  Intuitively, a more decorrelated latent space enforces more independent dimensions to encode information, therefore encourages disentanglement in representations.  ... 
arXiv:2103.16605v1 fatcat:yhddi74evzan3h4hd7xjnd2caa

Learning Decomposed Representation for Counterfactual Inference [article]

Anpeng Wu, Kun Kuang, Junkun Yuan, Bo Li, Runze Wu, Qiang Zhu, Yueting Zhuang, Fei Wu
2021 arXiv   pre-print
of both confounders and non-confounders, 2) balance confounder with sample re-weighting technique, and simultaneously 3) estimate the treatment effect in observational studies via counterfactual inference  ...  By modeling the different causal relations among observed pre-treatment variables, treatment and outcome, we propose a synergistic learning framework to 1) identify confounders by learning decomposed representations  ...  Besides, we can approximately achieve the conditional independence ( ) ⊥ | , by minimizing the mutual information between ( ) and (CLUB [8] ): L = [ , ( ( ))] + ={0,1} ( ( ) , ) : * = (16) Based on mutual  ... 
arXiv:2006.07040v2 fatcat:mw4twlenybevxgc6bflzzsqj5y

Hierarchical Disentanglement of Discriminative Latent Features for Zero-Shot Learning

Bin Tong, Chao Wang, Martin Klinkigt, Yoshiyuki Kobayashi, Yuuichi Nonaka
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Are image features trained with samples of seen classes expressive enough to capture the discriminative information for both seen and unseen classes?  ...  The discriminative latent features are learned through a group-wise disentanglement over feature groups with a hierarchical structure.  ...  Related work Our work is related to zero-shot learning and disentangled representation learning.  ... 
doi:10.1109/cvpr.2019.01173 dblp:conf/cvpr/TongWKKN19 fatcat:yz3y47fohvczhkx6hho75phl24

On Disentangled Representations Learned From Correlated Data [article]

Frederik Träuble, Elliot Creager, Niki Kilbertus, Francesco Locatello, Andrea Dittadi, Anirudh Goyal, Bernhard Schölkopf, Stefan Bauer
2021 arXiv   pre-print
We show and quantify that systematically induced correlations in the dataset are being learned and reflected in the latent representations, which has implications for downstream applications of disentanglement  ...  The focus of disentanglement approaches has been on identifying independent factors of variation in data.  ...  For the MIG score, one computes the mutual information between the latent representation and the ground truth factors and calculates the final score using a normalized gap between the two highest MI entries  ... 
arXiv:2006.07886v3 fatcat:irkinvphx5drdep6fe2kul2iby

DisCont: Self-Supervised Visual Attribute Disentanglement using Context Vectors [article]

Sarthak Bhagat, Vishaal Udandarao, Shagun Uppal
2020 arXiv   pre-print
Motivated by the recent surge in contrastive learning paradigms, our model bridges the gap between self-supervised contrastive learning algorithms and unsupervised disentanglement.  ...  Disentangling the underlying feature attributes within an image with no prior supervision is a challenging task.  ...  Saket Anand (IIIT Delhi) for his guidance in formulating the initial problem statement, and valuable comments and feedback on this paper.  ... 
arXiv:2006.05895v2 fatcat:bqml26d7ozb6dpi64zm5mqjqtm

Disentangled and Controllable Face Image Generation via 3D Imitative-Contrastive Learning

Yu Deng, Jiaolong Yang, Dong Chen, Fang Wen, Xin Tong
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Figure 1 : This paper presents a face image synthesis approach that generates realistic face images of virtual people with independent latent variables of identity, expression, pose, and illumination.  ...  The latent space is interpretable and highly disentangled, which allows precise control of the targeted images (e.g., degree of each pose angle, lighting intensity and direction), as shown in the top row  ...  A seminal GAN research for disentangled image generation is InfoGAN [6] , where the representation disentanglement is learned in an unsupervised manner via maximizing the mutual information between the  ... 
doi:10.1109/cvpr42600.2020.00520 dblp:conf/cvpr/DengYCWT20 fatcat:ijuehbe2v5gzlal5h6qsosgbju

Cluster-based Contrastive Disentangling for Generalized Zero-Shot Learning [article]

Yi Gao and Chenwei Tang and Jiancheng Lv
2022 arXiv   pre-print
The disentangled learning module with random swapping and semantic-visual alignment bridges the semantic gap.  ...  Moreover, we introduce contrastive learning on semantic-matched and class-unique variables to learn high intra-set and intra-class similarity, as well as inter-set and inter-class discriminability.  ...  Disentangled Representation Learning Disentangled representation learning means decomposing the feature representation into multiple factors that are independent of each other [2] , which is emulating  ... 
arXiv:2203.02648v1 fatcat:tlaz3bsmbfhfxfvwpcqom25r6i

Disentangled and Controllable Face Image Generation via 3D Imitative-Contrastive Learning [article]

Yu Deng, Jiaolong Yang, Dong Chen, Fang Wen, Xin Tong
2020 arXiv   pre-print
We propose DiscoFaceGAN, an approach for face image generation of virtual people with disentangled, precisely-controllable latent representations for identity of non-existing people, expression, pose,  ...  To deal with the generation freedom induced by the domain gap between real and rendered faces, we further introduce contrastive learning to promote disentanglement by comparing pairs of generated images  ...  A seminal GAN research for disentangled image generation is InfoGAN [6] , where the representation disentanglement is learned in an unsupervised manner via maximizing the mutual information between the  ... 
arXiv:2004.11660v2 fatcat:zaj3gkgyczdb3by7kvpcwjb3ne

Measuring the Biases and Effectiveness of Content-Style Disentanglement [article]

Xiao Liu, Spyridon Thermos, Gabriele Valvano, Agisilaos Chartsias, Alison O'Neil, Sotirios A. Tsaftaris
2021 arXiv   pre-print
Our findings, as well as the used task-independent metrics, can be used to guide the design and selection of new models for tasks where content-style representations are useful.  ...  While considerable effort has been made to measure disentanglement in vector representations, and assess its impact on task performance, such analysis for (spatial) content - style disentanglement is lacking  ...  , and achieves disentanglement with both design and learning biases.  ... 
arXiv:2008.12378v4 fatcat:mqo5s4gf2jbgrbt55vkqniujpa

SceneTrilogy: On Scene Sketches and its Relationship with Text and Photo [article]

Pinaki Nath Chowdhury and Ayan Kumar Bhunia and Tao Xiang and Yi-Zhe Song
2022 arXiv   pre-print
components via a conditional invertible neural network, and (ii) align cross-modalities information by maximising the mutual information between their modality-agnostic components using InfoNCE, with  ...  We importantly leverage the information bottleneck theory to achieve this goal, where we (i) decouple intra-modality information by minimising the mutual information between modality-specific and modality-agnostic  ...  We start by conducting intra-modality reasoning to decouple modality-agnostic information and modality-specific information, by minimising their mutual information using a conditional invertible neural  ... 
arXiv:2204.11964v1 fatcat:r7gaglcvqrhsni6cxqjieppfoy

Behavior-Driven Synthesis of Human Dynamics [article]

Andreas Blattmann, Timo Milbich, Michael Dorkenwald, Björn Ommer
2021 arXiv   pre-print
In this work, we present a model for human behavior synthesis which learns a dedicated representation of human dynamics independent of postures.  ...  In contrast, controlled behavior synthesis and transfer across individuals requires a deep understanding of body dynamics and calls for a representation of behavior that is independent of appearance and  ...  Acknowledgements The research leading to these results is funded by the German Federal Ministry for Economic Affairs and Energy within the project "KI-Absicherung -Safe AI for automated driving" and by  ... 
arXiv:2103.04677v2 fatcat:jtotlh5q6reqnnrviouptjkbni
« Previous Showing results 1 — 15 out of 5,526 results