{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/dirichlet-variational-autoencoder","title":"Dirichlet Variational Autoencoder","arxiv_id":"1901.02739","date":"2019-01-09","proceeding":"ICLR 2019 5","authors":["Weonyoung Joo","Wonsung Lee","Sungrae Park","Il-Chul Moon"],"abstract":"This paper proposes Dirichlet Variational Autoencoder (DirVAE) using a\nDirichlet prior for a continuous latent variable that exhibits the\ncharacteristic of the categorical probabilities. To infer the parameters of\nDirVAE, we utilize the stochastic gradient method by approximating the Gamma\ndistribution, which is a component of the Dirichlet distribution, with the\ninverse Gamma CDF approximation. Additionally, we reshape the component\ncollapsing issue by investigating two problem sources, which are decoder weight\ncollapsing and latent value collapsing, and we show that DirVAE has no\ncomponent collapsing; while Gaussian VAE exhibits the decoder weight collapsing\nand Stick-Breaking VAE shows the latent value collapsing. The experimental\nresults show that 1) DirVAE models the latent representation result with the\nbest log-likelihood compared to the baselines; and 2) DirVAE produces more\ninterpretable latent values with no collapsing issues which the baseline models\nsuffer from. Also, we show that the learned latent representation from the\nDirVAE achieves the best classification accuracy in the semi-supervised and the\nsupervised classification tasks on MNIST, OMNIGLOT, and SVHN compared to the\nbaseline VAEs. Finally, we demonstrated that the DirVAE augmented topic models\nshow better performances in most cases.","url_abs":"http://arxiv.org/abs/1901.02739v1","url_pdf":"http://arxiv.org/pdf/1901.02739v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"dirichlet-variational-autoencoder","repo_url":"https://github.com/sophieburkhardt/dirichlet-vae-topic-models","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"topic-models","task_name":"Topic Models"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1901.02739","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}