Dirichlet Variational Autoencoder

Joo, Weonyoung; Lee, Wonsung; Park, Sungrae; Moon, Il‐Chul

doi:10.1016/j.patcog.2020.107514

Cited by 69 publications

(26 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our application, it imposes a compact latent space, whose latent dimensions can be interpreted as mixture weights in a multinomial mixture model [49,50].…”

Section: Dirichlet-vaementioning

confidence: 99%

“…We then study a generalisation, in which the Gaussian prior is replaced by a Gaussian-Mixture prior (GMVAE) [42][43][44][45][46][47][48] in Section 3. In Section 4 we introduce the Dirichlet-VAE (DVAE), which uses a compact latent space with a Dirichlet prior [49,50]. Through a specific choice of decoder architecture we can interpret the decoder weights as the parameters of the mixture distributions in the probabilistic model, and can visualise these to directly interpret what the neural network is learning.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Better latent spaces for better autoencoders

et al. 2021

View full text Add to dashboard Cite

Autoencoders as tools behind anomaly searches at the LHC have the structural problem that they only work in one direction, extracting jets with higher complexity but not the other way around. To address this, we derive classifiers from the latent space of (variational) autoencoders, specifically in Gaussian mixture and Dirichlet latent spaces. In particular, the Dirichlet setup solves the problem and improves both the performance and the interpretability of the networks.

show abstract

“…In our application, it imposes a compact latent space, whose latent dimensions can be interpreted as mixture weights in a multinomial mixture model [49,50].…”

Section: Dirichlet-vaementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Better latent spaces for better autoencoders

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Although the latent variables follow a Gaussian distribution in the normal VAE, models have been proposed in which other distributions were assumed (Jang et al, 2017;Srivastava and Sutton, 2017;Joo et al, 2019). These models made it possible to sample from the distribution, except for a Gaussian distribution, by formulating the sampling procedure into a differentiable form.…”

Section: Related Workmentioning

confidence: 99%

“…Srivastava et al used the Laplace approximation, whereby the parameters of a Dirichlet distribution were approximated by the parameters of a Gaussian distribution, making it possible to sample from a Dirichlet distribution by adding a softmax layer to the normal VAE (Srivastava and Sutton, 2017). Joo et al applied parameter estimation of a multivariate Gamma distribution using an inverse Gamma cumulative distribution function approximation to enable sampling from a Dirichlet distribution (Joo et al, 2019). In this study, we used the Gumbel-Softmax distribution to obtain samples from a multinomial distribution.…”

Section: Related Workmentioning

confidence: 99%

Robot Concept Acquisition Based on Interaction Between Probabilistic and Deep Generative Models

Kuniyasu¹,

Nagaoka²,

Taniguchi³

et al. 2021

Front. Comput. Sci.

View full text Add to dashboard Cite

We propose a method for multimodal concept formation. In this method, unsupervised multimodal clustering and cross-modal inference, as well as unsupervised representation learning, can be performed by integrating the multimodal latent Dirichlet allocation (MLDA)-based concept formation and variational autoencoder (VAE)-based feature extraction. Multimodal clustering, representation learning, and cross-modal inference are critical for robots to form multimodal concepts from sensory data. Various models have been proposed for concept formation. However, in previous studies, features were extracted using manually designed or pre-trained feature extractors and representation learning was not performed simultaneously. Moreover, the generative probabilities of the features extracted from the sensory data could be predicted, but the sensory data could not be predicted in the cross-modal inference. Therefore, a method that can perform clustering, feature learning, and cross-modal inference among multimodal sensory data is required for concept formation. To realize such a method, we extend the VAE to the multinomial VAE (MNVAE), the latent variables of which follow a multinomial distribution, and construct a model that integrates the MNVAE and MLDA. In the experiments, the multimodal information of the images and words acquired by a robot was classified using the integrated model. The results demonstrated that the integrated model can classify the multimodal information as accurately as the previous model despite the feature extractor learning in an unsupervised manner, suitable image features for clustering can be learned, and cross-modal inference from the words to images is possible.

show abstract

“…Calculating the gradient for the sampling process from the Dirichlet distribution is difficult. Researchers propose approximation methods [38,39,40,41] in order to apply Dirichlet distribution to the neural topic model. We follow the rejection sampling method [42] in this work.…”

Section: Overview Of Our Frameworkmentioning

confidence: 99%

Exploiting Domain-Aware Aspect Similarity for Multi-Source Cross-Domain Sentiment Classification

Lai¹,

Ho²,

Lam³

2021

Adv. sci. technol. eng. syst. j.

View full text Add to dashboard Cite

We propose a novel framework exploiting domain-aware aspect similarity for solving the multisource cross-domain sentiment classification problem under the constraint of little labeled data. Existing works mainly focus on identifying the common sentiment features from all domains with weighting based on the coarse-grained domain similarity. We argue that it might not provide an accurate similarity measure due to the negative effect of domain-specific aspects. In addition, existing models usually involve training sub-models using a small portion of the labeled data which might not be appropriate under the constraint of little labeled data. To tickle the above limitations, we propose a domain-aware topic model to exploit the fine-grained domain-aware aspect similarity. We utilize the novel domain-aware linear layer to control the exposure of various domains to latent aspect topics. The model discovers latent aspect topics and also captures the proportion of latent aspect topics of the input. Next, we utilize the proposed topicattention network for training aspect models capturing the transferable sentiment knowledge regarding particular aspect topics. The framework finally makes predictions according to the aspect proportion of the testing data for adjusting the contribution of various aspect models. Experimental results show that our proposed framework achieves the state-of-the-art performance under the constraint of little labeled data. The framework has 71% classification accuracy when there are only 40 labeled data. The performance increases to around 82% with 200 labeled data. This proves the effectiveness of the fine-grained domain-aware aspect similarity measure.

show abstract

Dirichlet Variational Autoencoder

Cited by 69 publications

References 7 publications

Better latent spaces for better autoencoders

Better latent spaces for better autoencoders

Robot Concept Acquisition Based on Interaction Between Probabilistic and Deep Generative Models

Exploiting Domain-Aware Aspect Similarity for Multi-Source Cross-Domain Sentiment Classification

Contact Info

Product

Resources

About