Interpolating convolutional neural networks using batch normalization
Perceiving a visual concept as a mixture of learned ones is natural for humans, aiding them to grasp new concepts and strengthening old ones. For all their power and recent success, deep convolutional networks do not have this ability. Inspired by recent work on universal representations for neural...
| मुख्य लेखकों: | , , , |
|---|---|
| स्वरूप: | Conference item |
| प्रकाशित: |
Springer
2018
|
| _version_ | 1826268087791386624 |
|---|---|
| author | Data, G Ngu, K Murray, D Prisacariu, V |
| author_facet | Data, G Ngu, K Murray, D Prisacariu, V |
| author_sort | Data, G |
| collection | OXFORD |
| description | Perceiving a visual concept as a mixture of learned ones is natural for humans, aiding them to grasp new concepts and strengthening old ones. For all their power and recent success, deep convolutional networks do not have this ability. Inspired by recent work on universal representations for neural networks, we propose a simple emulation of this mechanism by purposing batch normalization layers to discriminate visual classes, and formulating a way to combine them to solve new tasks. We show that this can be applied for 2-way few-shot learning where we obtain between 4% and 17% better accuracy compared to straightforward full fine-tuning, and demonstrate that it can also be extended to the orthogonal application of style transfer. |
| first_indexed | 2024-03-06T21:04:13Z |
| format | Conference item |
| id | oxford-uuid:3be8efd8-cb06-458e-9f4b-62dd2d0693a9 |
| institution | University of Oxford |
| last_indexed | 2024-03-06T21:04:13Z |
| publishDate | 2018 |
| publisher | Springer |
| record_format | dspace |
| spelling | oxford-uuid:3be8efd8-cb06-458e-9f4b-62dd2d0693a92022-03-26T14:10:18ZInterpolating convolutional neural networks using batch normalizationConference itemhttp://purl.org/coar/resource_type/c_5794uuid:3be8efd8-cb06-458e-9f4b-62dd2d0693a9Symplectic Elements at OxfordSpringer2018Data, GNgu, KMurray, DPrisacariu, VPerceiving a visual concept as a mixture of learned ones is natural for humans, aiding them to grasp new concepts and strengthening old ones. For all their power and recent success, deep convolutional networks do not have this ability. Inspired by recent work on universal representations for neural networks, we propose a simple emulation of this mechanism by purposing batch normalization layers to discriminate visual classes, and formulating a way to combine them to solve new tasks. We show that this can be applied for 2-way few-shot learning where we obtain between 4% and 17% better accuracy compared to straightforward full fine-tuning, and demonstrate that it can also be extended to the orthogonal application of style transfer. |
| spellingShingle | Data, G Ngu, K Murray, D Prisacariu, V Interpolating convolutional neural networks using batch normalization |
| title | Interpolating convolutional neural networks using batch normalization |
| title_full | Interpolating convolutional neural networks using batch normalization |
| title_fullStr | Interpolating convolutional neural networks using batch normalization |
| title_full_unstemmed | Interpolating convolutional neural networks using batch normalization |
| title_short | Interpolating convolutional neural networks using batch normalization |
| title_sort | interpolating convolutional neural networks using batch normalization |
| work_keys_str_mv | AT datag interpolatingconvolutionalneuralnetworksusingbatchnormalization AT nguk interpolatingconvolutionalneuralnetworksusingbatchnormalization AT murrayd interpolatingconvolutionalneuralnetworksusingbatchnormalization AT prisacariuv interpolatingconvolutionalneuralnetworksusingbatchnormalization |