Adversarial masking for self-supervised learning
We propose ADIOS, a masked image modeling (MIM) framework for self-supervised learning, which simultaneously learns a masking function and an image encoder using an adversarial objective. The image encoder is trained to minimise the distance between representations of the original and that of a mask...
| Main Authors: | , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Journal of Machine Learning Research
2022
|
| _version_ | 1826310574666940416 |
|---|---|
| author | Shi, Y Siddharth, N Torr, PHS Kosiorek, AR |
| author_facet | Shi, Y Siddharth, N Torr, PHS Kosiorek, AR |
| author_sort | Shi, Y |
| collection | OXFORD |
| description | We propose ADIOS, a masked image modeling (MIM) framework for self-supervised learning, which simultaneously learns a masking function and an image encoder using an adversarial objective. The image encoder is trained to minimise the distance between representations of the original and that of a masked image. The masking function, conversely, aims at maximising this distance. ADIOS consistently improves on state-ofthe-art self-supervised learning (SSL) methods on a variety of tasks and datasets-including classification on ImageNet100 and STL10, transfer learning on CIFAR10/100, Flowers102 and iNaturalist, as well as robustness evaluated on the backgrounds challenge (Xiao et al., 2021)-while generating semantically meaningful masks. Unlike modern MIM models such as MAE, BEiT and iBOT, ADIOS does not rely on the image-patch tokenisation construction of Vision Transformers, and can be implemented with convolutional backbones. We further demonstrate that the masks learned by ADIOS are more effective in improving representation learning of SSL methods than masking schemes used in popular MIM models. |
| first_indexed | 2024-03-07T07:53:58Z |
| format | Conference item |
| id | oxford-uuid:103e0dec-e729-4366-9041-c4c1fbf0dacc |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T07:53:58Z |
| publishDate | 2022 |
| publisher | Journal of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:103e0dec-e729-4366-9041-c4c1fbf0dacc2023-08-10T11:07:44ZAdversarial masking for self-supervised learningConference itemhttp://purl.org/coar/resource_type/c_5794uuid:103e0dec-e729-4366-9041-c4c1fbf0daccEnglishSymplectic ElementsJournal of Machine Learning Research2022Shi, YSiddharth, NTorr, PHSKosiorek, ARWe propose ADIOS, a masked image modeling (MIM) framework for self-supervised learning, which simultaneously learns a masking function and an image encoder using an adversarial objective. The image encoder is trained to minimise the distance between representations of the original and that of a masked image. The masking function, conversely, aims at maximising this distance. ADIOS consistently improves on state-ofthe-art self-supervised learning (SSL) methods on a variety of tasks and datasets-including classification on ImageNet100 and STL10, transfer learning on CIFAR10/100, Flowers102 and iNaturalist, as well as robustness evaluated on the backgrounds challenge (Xiao et al., 2021)-while generating semantically meaningful masks. Unlike modern MIM models such as MAE, BEiT and iBOT, ADIOS does not rely on the image-patch tokenisation construction of Vision Transformers, and can be implemented with convolutional backbones. We further demonstrate that the masks learned by ADIOS are more effective in improving representation learning of SSL methods than masking schemes used in popular MIM models. |
| spellingShingle | Shi, Y Siddharth, N Torr, PHS Kosiorek, AR Adversarial masking for self-supervised learning |
| title | Adversarial masking for self-supervised learning |
| title_full | Adversarial masking for self-supervised learning |
| title_fullStr | Adversarial masking for self-supervised learning |
| title_full_unstemmed | Adversarial masking for self-supervised learning |
| title_short | Adversarial masking for self-supervised learning |
| title_sort | adversarial masking for self supervised learning |
| work_keys_str_mv | AT shiy adversarialmaskingforselfsupervisedlearning AT siddharthn adversarialmaskingforselfsupervisedlearning AT torrphs adversarialmaskingforselfsupervisedlearning AT kosiorekar adversarialmaskingforselfsupervisedlearning |