Generative Model With Dynamic Linear Flow
Flow-based generative models are a family of exact log-likelihood models with tractable sampling and latent-variable inference, hence conceptually attractive for modeling complex distributions. However, flow-based models are limited by density estimation performance issues as compared to state-of-th...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2019-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/8869769/ |
| _version_ | 1818877544890368000 |
|---|---|
| author | Huadong Liao Jiawei He Kunxian Shu |
| author_facet | Huadong Liao Jiawei He Kunxian Shu |
| author_sort | Huadong Liao |
| collection | DOAJ |
| description | Flow-based generative models are a family of exact log-likelihood models with tractable sampling and latent-variable inference, hence conceptually attractive for modeling complex distributions. However, flow-based models are limited by density estimation performance issues as compared to state-of-the-art autoregressive models. Autoregressive models, which also belong to the family of likelihood-based methods, however suffer from limited parallelizability. In this paper, we propose Dynamic Linear Flow (DLF), a new family of invertible transformations with partially autoregressive structure. Our method benefits from the efficient computation of flow-based methods and high density estimation performance of autoregressive methods. We demonstrate that the proposed DLF yields state-of-the-art performance on ImageNet 32 × 32 and 64 × 64 out of all flow-based methods. Additionally, DLF converges significantly faster than previous flow-based methods such as Glow. |
| first_indexed | 2024-12-19T13:59:59Z |
| format | Article |
| id | doaj.art-470694fb04064a5286959a5476b77c95 |
| institution | Directory Open Access Journal |
| issn | 2169-3536 |
| language | English |
| last_indexed | 2024-12-19T13:59:59Z |
| publishDate | 2019-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj.art-470694fb04064a5286959a5476b77c952022-12-21T20:18:29ZengIEEEIEEE Access2169-35362019-01-01715017515018310.1109/ACCESS.2019.29475678869769Generative Model With Dynamic Linear FlowHuadong Liao0https://orcid.org/0000-0002-3401-3032Jiawei He1Kunxian Shu2Chongqing Key Laboratory on Big Data for Bio Intelligence, Chongqing University of Posts and Telecommunications, Chongqing, ChinaSchool of Pharmaceutical and Food Sciences, Zhuhai College of Jilin University, Zhuhai, ChinaChongqing Key Laboratory on Big Data for Bio Intelligence, Chongqing University of Posts and Telecommunications, Chongqing, ChinaFlow-based generative models are a family of exact log-likelihood models with tractable sampling and latent-variable inference, hence conceptually attractive for modeling complex distributions. However, flow-based models are limited by density estimation performance issues as compared to state-of-the-art autoregressive models. Autoregressive models, which also belong to the family of likelihood-based methods, however suffer from limited parallelizability. In this paper, we propose Dynamic Linear Flow (DLF), a new family of invertible transformations with partially autoregressive structure. Our method benefits from the efficient computation of flow-based methods and high density estimation performance of autoregressive methods. We demonstrate that the proposed DLF yields state-of-the-art performance on ImageNet 32 × 32 and 64 × 64 out of all flow-based methods. Additionally, DLF converges significantly faster than previous flow-based methods such as Glow.https://ieeexplore.ieee.org/document/8869769/Exact likelihoodgenerative modelsinvertible transformation |
| spellingShingle | Huadong Liao Jiawei He Kunxian Shu Generative Model With Dynamic Linear Flow IEEE Access Exact likelihood generative models invertible transformation |
| title | Generative Model With Dynamic Linear Flow |
| title_full | Generative Model With Dynamic Linear Flow |
| title_fullStr | Generative Model With Dynamic Linear Flow |
| title_full_unstemmed | Generative Model With Dynamic Linear Flow |
| title_short | Generative Model With Dynamic Linear Flow |
| title_sort | generative model with dynamic linear flow |
| topic | Exact likelihood generative models invertible transformation |
| url | https://ieeexplore.ieee.org/document/8869769/ |
| work_keys_str_mv | AT huadongliao generativemodelwithdynamiclinearflow AT jiaweihe generativemodelwithdynamiclinearflow AT kunxianshu generativemodelwithdynamiclinearflow |