Understanding the role of individual units in a deep neural network
Deep neural networks excel at finding hierarchical representations that solve complex tasks over large datasets. How can we humans understand these learned representations? In this work, we present network dissection, an analytic framework to systematically identify the semantics of individual hidde...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Proceedings of the National Academy of Sciences
2021
|
| Online Access: | https://hdl.handle.net/1721.1/130269 |
| _version_ | 1811095898050002944 |
|---|---|
| author | Bau, David Zhu, Jun-Yan Strobelt, Hendrik Lapedriza Garcia, Agata Zhou, Bolei Torralba, Antonio |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Bau, David Zhu, Jun-Yan Strobelt, Hendrik Lapedriza Garcia, Agata Zhou, Bolei Torralba, Antonio |
| author_sort | Bau, David |
| collection | MIT |
| description | Deep neural networks excel at finding hierarchical representations that solve complex tasks over large datasets. How can we humans understand these learned representations? In this work, we present network dissection, an analytic framework to systematically identify the semantics of individual hidden units within image classification and image generation networks. First, we analyze a convolutional neural network (CNN) trained on scene classification and discover units that match a diverse set of object concepts. We find evidence that the network has learned many object classes that play crucial roles in classifying scene classes. Second, we use a similar analytic method to analyze a generative adversarial network (GAN) model trained to generate scenes. By analyzing changes made when small sets of units are activated or deactivated, we find that objects can be added and removed from the output scenes while adapting to the context. Finally, we apply our analytic framework to understanding adversarial attacks and to semantic image editing. |
| first_indexed | 2024-09-23T16:33:04Z |
| format | Article |
| id | mit-1721.1/130269 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T16:33:04Z |
| publishDate | 2021 |
| publisher | Proceedings of the National Academy of Sciences |
| record_format | dspace |
| spelling | mit-1721.1/1302692022-09-29T20:06:58Z Understanding the role of individual units in a deep neural network Bau, David Zhu, Jun-Yan Strobelt, Hendrik Lapedriza Garcia, Agata Zhou, Bolei Torralba, Antonio Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Media Laboratory MIT-IBM Watson AI Lab Deep neural networks excel at finding hierarchical representations that solve complex tasks over large datasets. How can we humans understand these learned representations? In this work, we present network dissection, an analytic framework to systematically identify the semantics of individual hidden units within image classification and image generation networks. First, we analyze a convolutional neural network (CNN) trained on scene classification and discover units that match a diverse set of object concepts. We find evidence that the network has learned many object classes that play crucial roles in classifying scene classes. Second, we use a similar analytic method to analyze a generative adversarial network (GAN) model trained to generate scenes. By analyzing changes made when small sets of units are activated or deactivated, we find that objects can be added and removed from the output scenes while adapting to the context. Finally, we apply our analytic framework to understanding adversarial attacks and to semantic image editing. Defense Advanced Research Projects Agency (Award FA8750-18-C-0004) NSF (Grants 1524817 and BIGDATA-1447476) 2021-03-29T21:05:50Z 2021-03-29T21:05:50Z 2020-09 2019-08 2021-03-16T15:02:16Z Article http://purl.org/eprint/type/JournalArticle 0027-8424 1091-6490 https://hdl.handle.net/1721.1/130269 Bau, David et al. "Understanding the role of individual units in a deep neural network." Proceedings of the National Academy of Sciences 117, 48 (September 2020): 30071-30078 © 2020 National Academy of Sciences en http://dx.doi.org/10.1073/pnas.1907375117 Proceedings of the National Academy of Sciences Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Proceedings of the National Academy of Sciences PNAS |
| spellingShingle | Bau, David Zhu, Jun-Yan Strobelt, Hendrik Lapedriza Garcia, Agata Zhou, Bolei Torralba, Antonio Understanding the role of individual units in a deep neural network |
| title | Understanding the role of individual units in a deep neural network |
| title_full | Understanding the role of individual units in a deep neural network |
| title_fullStr | Understanding the role of individual units in a deep neural network |
| title_full_unstemmed | Understanding the role of individual units in a deep neural network |
| title_short | Understanding the role of individual units in a deep neural network |
| title_sort | understanding the role of individual units in a deep neural network |
| url | https://hdl.handle.net/1721.1/130269 |
| work_keys_str_mv | AT baudavid understandingtheroleofindividualunitsinadeepneuralnetwork AT zhujunyan understandingtheroleofindividualunitsinadeepneuralnetwork AT strobelthendrik understandingtheroleofindividualunitsinadeepneuralnetwork AT lapedrizagarciaagata understandingtheroleofindividualunitsinadeepneuralnetwork AT zhoubolei understandingtheroleofindividualunitsinadeepneuralnetwork AT torralbaantonio understandingtheroleofindividualunitsinadeepneuralnetwork |