@article{
 title = {Diagnosing and enhancing VAE models},
 type = {article},
 year = {2019},
 keywords = {deep generative model,variational autoencoder},
 pages = {1-44},
 id = {b2139d5d-4090-349a-bc75-88db0efdace7},
 created = {2022-03-23T06:17:59.504Z},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {1ff583c0-be37-34fa-9c04-73c69437d354},
 last_modified = {2022-03-28T09:45:10.605Z},
 read = {true},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {Dai2019a},
 folder_uuids = {1853f94b-7af1-40fa-b068-4758e9a02bc4},
 private_publication = {false},
 abstract = {Although variational autoencoders (VAEs) represent a widely influential deep generative model, many aspects of the underlying energy function remain poorly understood. In particular, it is commonly believed that Gaussian encoder/decoder assumptions reduce the effectiveness of VAEs in generating realistic samples. In this regard, we rigorously analyze the VAE objective, differentiating situations where this belief is and is not actually true. We then leverage the corresponding insights to develop a simple VAE enhancement that requires no additional hyperparameters or sensitive tuning. Quantitatively, this proposal produces crisp samples and stable FID scores that are actually competitive with a variety of GAN models, all while retaining desirable attributes of the original VAE architecture. The code for our model is available at https://github.com/daib13/TwoStageVAE.},
 bibtype = {article},
 author = {Dai, Bin and Wipf, David},
 journal = {7th International Conference on Learning Representations, ICLR 2019}
}

{"_id":"5gqYSN4wMxmeQeii9","bibbaseid":"dai-wipf-diagnosingandenhancingvaemodels-2019","authorIDs":[],"author_short":["Dai, B.","Wipf, D."],"bibdata":{"title":"Diagnosing and enhancing VAE models","type":"article","year":"2019","keywords":"deep generative model,variational autoencoder","pages":"1-44","id":"b2139d5d-4090-349a-bc75-88db0efdace7","created":"2022-03-23T06:17:59.504Z","file_attached":"true","profile_id":"235249c2-3ed4-314a-b309-b1ea0330f5d9","group_id":"1ff583c0-be37-34fa-9c04-73c69437d354","last_modified":"2022-03-28T09:45:10.605Z","read":"true","starred":false,"authored":false,"confirmed":"true","hidden":false,"citation_key":"Dai2019a","folder_uuids":"1853f94b-7af1-40fa-b068-4758e9a02bc4","private_publication":false,"abstract":"Although variational autoencoders (VAEs) represent a widely influential deep generative model, many aspects of the underlying energy function remain poorly understood. In particular, it is commonly believed that Gaussian encoder/decoder assumptions reduce the effectiveness of VAEs in generating realistic samples. In this regard, we rigorously analyze the VAE objective, differentiating situations where this belief is and is not actually true. We then leverage the corresponding insights to develop a simple VAE enhancement that requires no additional hyperparameters or sensitive tuning. Quantitatively, this proposal produces crisp samples and stable FID scores that are actually competitive with a variety of GAN models, all while retaining desirable attributes of the original VAE architecture. The code for our model is available at https://github.com/daib13/TwoStageVAE.","bibtype":"article","author":"Dai, Bin and Wipf, David","journal":"7th International Conference on Learning Representations, ICLR 2019","bibtex":"@article{\n title = {Diagnosing and enhancing VAE models},\n type = {article},\n year = {2019},\n keywords = {deep generative model,variational autoencoder},\n pages = {1-44},\n id = {b2139d5d-4090-349a-bc75-88db0efdace7},\n created = {2022-03-23T06:17:59.504Z},\n file_attached = {true},\n profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},\n group_id = {1ff583c0-be37-34fa-9c04-73c69437d354},\n last_modified = {2022-03-28T09:45:10.605Z},\n read = {true},\n starred = {false},\n authored = {false},\n confirmed = {true},\n hidden = {false},\n citation_key = {Dai2019a},\n folder_uuids = {1853f94b-7af1-40fa-b068-4758e9a02bc4},\n private_publication = {false},\n abstract = {Although variational autoencoders (VAEs) represent a widely influential deep generative model, many aspects of the underlying energy function remain poorly understood. In particular, it is commonly believed that Gaussian encoder/decoder assumptions reduce the effectiveness of VAEs in generating realistic samples. In this regard, we rigorously analyze the VAE objective, differentiating situations where this belief is and is not actually true. We then leverage the corresponding insights to develop a simple VAE enhancement that requires no additional hyperparameters or sensitive tuning. Quantitatively, this proposal produces crisp samples and stable FID scores that are actually competitive with a variety of GAN models, all while retaining desirable attributes of the original VAE architecture. The code for our model is available at https://github.com/daib13/TwoStageVAE.},\n bibtype = {article},\n author = {Dai, Bin and Wipf, David},\n journal = {7th International Conference on Learning Representations, ICLR 2019}\n}","author_short":["Dai, B.","Wipf, D."],"urls":{"Paper":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c/file/03923c9a-dcac-9437-3a90-5e40286dc3a6/Dai_and_Wipf___2019___Diagnosing_and_Enhancing_VAE_Models.pdf.pdf"},"biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","bibbaseid":"dai-wipf-diagnosingandenhancingvaemodels-2019","role":"author","keyword":["deep generative model","variational autoencoder"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","creationDate":"2020-01-27T02:13:33.870Z","downloads":0,"keywords":["deep generative model","variational autoencoder"],"search_terms":["diagnosing","enhancing","vae","models","dai","wipf"],"title":"Diagnosing and enhancing VAE models","year":2019,"dataSources":["hEoKh4ygEAWbAZ5iy","ya2CyA73rpZseyrZ8","e6KiFpst4JLfXwT7X","2252seNhipfTmjEBQ"]}