The Variational Homoencoder: Learning to learn high capacity generative models from few examples
© 34th Conference on Uncertainty in Artificial Intelligence 2018. All rights reserved. Hierarchical Bayesian methods can unify many related tasks (e.g. k-shot classification, conditional and unconditional generation) as inference within a single generative model. However, when this generative model...
| Main Authors: | , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137610 |
| _version_ | 1826197950768873472 |
|---|---|
| author | Hewitt, Luke B. Nye, Maxwell I. Gane, Andreea Jaakkola, Tommi Tenenbaum, Joshua B. |
| author2 | MIT-IBM Watson AI Lab |
| author_facet | MIT-IBM Watson AI Lab Hewitt, Luke B. Nye, Maxwell I. Gane, Andreea Jaakkola, Tommi Tenenbaum, Joshua B. |
| author_sort | Hewitt, Luke B. |
| collection | MIT |
| description | © 34th Conference on Uncertainty in Artificial Intelligence 2018. All rights reserved. Hierarchical Bayesian methods can unify many related tasks (e.g. k-shot classification, conditional and unconditional generation) as inference within a single generative model. However, when this generative model is expressed as a powerful neural network such as a PixelCNN, we show that existing learning techniques typically fail to effectively use latent variables. To address this, we develop a modification of the Variational Autoencoder in which encoded observations are decoded to new elements from the same class. This technique, which we call a Variational Homoencoder (VHE), produces a hierarchical latent variable model which better utilises latent variables. We use the VHE framework to learn a hierarchical PixelCNN on the Omniglot dataset, which outperforms all existing models on test set likelihood and achieves strong performance on one-shot generation and classification tasks. We additionally validate the VHE on natural images from the YouTube Faces database. Finally, we develop extensions of the model that apply to richer dataset structures such as factorial and hierarchical categories. |
| first_indexed | 2024-09-23T10:56:16Z |
| format | Article |
| id | mit-1721.1/137610 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T10:56:16Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1376102022-09-27T16:02:16Z The Variational Homoencoder: Learning to learn high capacity generative models from few examples Hewitt, Luke B. Nye, Maxwell I. Gane, Andreea Jaakkola, Tommi Tenenbaum, Joshua B. MIT-IBM Watson AI Lab © 34th Conference on Uncertainty in Artificial Intelligence 2018. All rights reserved. Hierarchical Bayesian methods can unify many related tasks (e.g. k-shot classification, conditional and unconditional generation) as inference within a single generative model. However, when this generative model is expressed as a powerful neural network such as a PixelCNN, we show that existing learning techniques typically fail to effectively use latent variables. To address this, we develop a modification of the Variational Autoencoder in which encoded observations are decoded to new elements from the same class. This technique, which we call a Variational Homoencoder (VHE), produces a hierarchical latent variable model which better utilises latent variables. We use the VHE framework to learn a hierarchical PixelCNN on the Omniglot dataset, which outperforms all existing models on test set likelihood and achieves strong performance on one-shot generation and classification tasks. We additionally validate the VHE on natural images from the YouTube Faces database. Finally, we develop extensions of the model that apply to richer dataset structures such as factorial and hierarchical categories. 2021-11-05T20:11:24Z 2021-11-05T20:11:24Z 2018 2019-05-31T16:35:52Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137610 Hewitt, Luke B., Nye, Maxwell I., Gane, Andreea, Jaakkola, Tommi and Tenenbaum, Joshua B. 2018. "The Variational Homoencoder: Learning to learn high capacity generative models from few examples." en http://auai.org/uai2018/accepted.php Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf arXiv |
| spellingShingle | Hewitt, Luke B. Nye, Maxwell I. Gane, Andreea Jaakkola, Tommi Tenenbaum, Joshua B. The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title | The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title_full | The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title_fullStr | The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title_full_unstemmed | The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title_short | The Variational Homoencoder: Learning to learn high capacity generative models from few examples |
| title_sort | variational homoencoder learning to learn high capacity generative models from few examples |
| url | https://hdl.handle.net/1721.1/137610 |
| work_keys_str_mv | AT hewittlukeb thevariationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT nyemaxwelli thevariationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT ganeandreea thevariationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT jaakkolatommi thevariationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT tenenbaumjoshuab thevariationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT hewittlukeb variationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT nyemaxwelli variationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT ganeandreea variationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT jaakkolatommi variationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples AT tenenbaumjoshuab variationalhomoencoderlearningtolearnhighcapacitygenerativemodelsfromfewexamples |