Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks
Summary: Due to lack of the kernel awareness, some popular deep image reconstruction networks are unstable. To address this problem, here we introduce the bounded relative error norm (BREN) property, which is a special case of the Lipschitz continuity. Then, we perform a convergence study consisting...
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-05-01
|
| Series: | Patterns |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666389922000551 |
| _version_ | 1828553821417111552 |
|---|---|
| author | Weiwen Wu Dianlin Hu Wenxiang Cong Hongming Shan Shaoyu Wang Chuang Niu Pingkun Yan Hengyong Yu Varut Vardhanabhuti Ge Wang |
| author_facet | Weiwen Wu Dianlin Hu Wenxiang Cong Hongming Shan Shaoyu Wang Chuang Niu Pingkun Yan Hengyong Yu Varut Vardhanabhuti Ge Wang |
| author_sort | Weiwen Wu |
| collection | DOAJ |
| description | Summary: Due to lack of the kernel awareness, some popular deep image reconstruction networks are unstable. To address this problem, here we introduce the bounded relative error norm (BREN) property, which is a special case of the Lipschitz continuity. Then, we perform a convergence study consisting of two parts: (1) a heuristic analysis on the convergence of the analytic compressed iterative deep (ACID) scheme (with the simplification that the CS module achieves a perfect sparsification), and (2) a mathematically denser analysis (with the two approximations: [1] AT is viewed as an inverse A-1 in the perspective of an iterative reconstruction procedure and [2] a pseudo-inverse is used for a total variation operator H). Also, we present adversarial attack algorithms to perturb the selected reconstruction networks respectively and, more importantly, to attack the ACID workflow as a whole. Finally, we show the numerical convergence of the ACID iteration in terms of the Lipschitz constant and the local stability against noise. The bigger picture: For deep tomographic reconstruction to realize its full potential in practice, it is critically important to address the instabilities of deep reconstruction networks, which were identified in a recent PNAS paper. Our analytic compressed iterative deep (ACID) framework has provided an effective solution to address this challenge by synergizing deep learning and compressed sensing through iterative refinement. Here, we provide an initial convergence analysis, describe an algorithm to attack the entire ACID workflow, and establish not only its capability of stabilizing an unstable deep reconstruction network but also its stability against adversarial attacks dedicated to ACID as a whole. Although our theoretical results are under approximations, they shed light on the converging mechanism of ACID, serving as a basis for further investigation. |
| first_indexed | 2024-12-12T05:25:16Z |
| format | Article |
| id | doaj.art-00be61a3a58543d095d8f90f4e171dc3 |
| institution | Directory Open Access Journal |
| issn | 2666-3899 |
| language | English |
| last_indexed | 2024-12-12T05:25:16Z |
| publishDate | 2022-05-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Patterns |
| spelling | doaj.art-00be61a3a58543d095d8f90f4e171dc32022-12-22T00:36:28ZengElsevierPatterns2666-38992022-05-0135100475Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacksWeiwen Wu0Dianlin Hu1Wenxiang Cong2Hongming Shan3Shaoyu Wang4Chuang Niu5Pingkun Yan6Hengyong Yu7Varut Vardhanabhuti8Ge Wang9Biomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA; School of Biomedical Engineering, Sun Yat-sen University, Shenzhen, Guangdong, China; Department of Diagnostic Radiology, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong SAR, ChinaThe Laboratory of Image Science and Technology, Southeast University, Nanjing, ChinaBiomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USABiomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA; Institute of Science and Technology for Brain-inspired Intelligence, Fudan University, Shanghai, ChinaDepartment of Electrical & Computer Engineering, University of Massachusetts Lowell, Lowell, MA, USABiomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USABiomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USADepartment of Electrical & Computer Engineering, University of Massachusetts Lowell, Lowell, MA, USA; Corresponding authorDepartment of Diagnostic Radiology, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong SAR, China; Corresponding authorBiomedical Imaging Center, Center for Biotechnology and Interdisciplinary Studies, Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY, USA; Corresponding authorSummary: Due to lack of the kernel awareness, some popular deep image reconstruction networks are unstable. To address this problem, here we introduce the bounded relative error norm (BREN) property, which is a special case of the Lipschitz continuity. Then, we perform a convergence study consisting of two parts: (1) a heuristic analysis on the convergence of the analytic compressed iterative deep (ACID) scheme (with the simplification that the CS module achieves a perfect sparsification), and (2) a mathematically denser analysis (with the two approximations: [1] AT is viewed as an inverse A-1 in the perspective of an iterative reconstruction procedure and [2] a pseudo-inverse is used for a total variation operator H). Also, we present adversarial attack algorithms to perturb the selected reconstruction networks respectively and, more importantly, to attack the ACID workflow as a whole. Finally, we show the numerical convergence of the ACID iteration in terms of the Lipschitz constant and the local stability against noise. The bigger picture: For deep tomographic reconstruction to realize its full potential in practice, it is critically important to address the instabilities of deep reconstruction networks, which were identified in a recent PNAS paper. Our analytic compressed iterative deep (ACID) framework has provided an effective solution to address this challenge by synergizing deep learning and compressed sensing through iterative refinement. Here, we provide an initial convergence analysis, describe an algorithm to attack the entire ACID workflow, and establish not only its capability of stabilizing an unstable deep reconstruction network but also its stability against adversarial attacks dedicated to ACID as a whole. Although our theoretical results are under approximations, they shed light on the converging mechanism of ACID, serving as a basis for further investigation.http://www.sciencedirect.com/science/article/pii/S2666389922000551DSML2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem |
| spellingShingle | Weiwen Wu Dianlin Hu Wenxiang Cong Hongming Shan Shaoyu Wang Chuang Niu Pingkun Yan Hengyong Yu Varut Vardhanabhuti Ge Wang Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks Patterns DSML2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem |
| title | Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks |
| title_full | Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks |
| title_fullStr | Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks |
| title_full_unstemmed | Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks |
| title_short | Stabilizing deep tomographic reconstruction: Part B. Convergence analysis and adversarial attacks |
| title_sort | stabilizing deep tomographic reconstruction part b convergence analysis and adversarial attacks |
| topic | DSML2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem |
| url | http://www.sciencedirect.com/science/article/pii/S2666389922000551 |
| work_keys_str_mv | AT weiwenwu stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT dianlinhu stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT wenxiangcong stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT hongmingshan stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT shaoyuwang stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT chuangniu stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT pingkunyan stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT hengyongyu stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT varutvardhanabhuti stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks AT gewang stabilizingdeeptomographicreconstructionpartbconvergenceanalysisandadversarialattacks |