IUNets: Learnable invertible up-and downsampling for large-scale inverse problems

Christian Etmann; Rihuan Ke; Carola Bibiane Schonlieb

doi:10.1109/MLSP49062.2020.9231874

IUNets: Learnable invertible up-and downsampling for large-scale inverse problems

Christian Etmann^*, Rihuan Ke, Carola Bibiane Schonlieb

^*Corresponding author for this work

School of Mathematics

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

16 Citations (Scopus)

Abstract

U-Nets have been established as a standard neural network architecture for image-to-image problems such as segmentation and inverse problems in imaging. For high-dimensional applications, as they for example appear in 3D medical imaging, U-Nets however have prohibitive memory requirements. Here, we present a new fully-invertible U-Net-based architecture called the iUNet, which allows for the application of highly memory-efficient backpropagation procedures. As its main building block, we introduce learnable and invertible up- an downsampling operations. For this, we developed an open-source implementation in Pytorch for 1D, 2D and 3D data.

Original language	English
Title of host publication	Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020
Publisher	IEEE Computer Society
ISBN (Electronic)	9781728166629
DOIs	https://doi.org/10.1109/MLSP49062.2020.9231874
Publication status	Published - Sept 2020
Event	30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020 - Virtual, Espoo, Finland Duration: 21 Sept 2020 → 24 Sept 2020

Publication series

Name	IEEE International Workshop on Machine Learning for Signal Processing, MLSP
Volume	2020-September
ISSN (Print)	2161-0363
ISSN (Electronic)	2161-0371

Conference

Conference	30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020
Country/Territory	Finland
City	Virtual, Espoo
Period	21/09/20 → 24/09/20

Bibliographical note

Funding Information:
CE and CBS acknowledge support from the Wellcome Innovator Award RG98755. RK and CBS acknowledge support from the EPSRC grant EP/T003553/1. CE additionally acknowledges partial funding by the Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 281474342: ’RTG pi3 - Parameter Identification - Analysis, Algorithms, Applications’ for parts of the work done while being a member of RTG pi3. CBS additionally acknowledges support from the Leverhulme Trust project on ‘Breaking the non-convexity barrier’, the Philip Leverhulme Prize, the EPSRC grant EP/S026045/1, the EPSRC Centre Nr. EP/N014588/1, the RISE projects CHiPS and NoMADS, the Cantab Capital Institute for the Mathematics of Information and the Alan Turing Institute.

Publisher Copyright:
© 2020 IEEE.

Keywords

Inverse problems
Invertible networks
Neural networks
Segmentation
U-Net

Access to Document

10.1109/MLSP49062.2020.9231874

Handle.net

Persistent link

Cite this

Etmann, C., Ke, R., & Schonlieb, C. B. (2020). IUNets: Learnable invertible up-and downsampling for large-scale inverse problems. In Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020 Article 9231874 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2020-September). IEEE Computer Society. https://doi.org/10.1109/MLSP49062.2020.9231874

@inproceedings{08f4fdf4654a4fce941ed33f7380ff3f,

title = "IUNets: Learnable invertible up-and downsampling for large-scale inverse problems",

abstract = "U-Nets have been established as a standard neural network architecture for image-to-image problems such as segmentation and inverse problems in imaging. For high-dimensional applications, as they for example appear in 3D medical imaging, U-Nets however have prohibitive memory requirements. Here, we present a new fully-invertible U-Net-based architecture called the iUNet, which allows for the application of highly memory-efficient backpropagation procedures. As its main building block, we introduce learnable and invertible up- an downsampling operations. For this, we developed an open-source implementation in Pytorch for 1D, 2D and 3D data. ",

keywords = "Inverse problems, Invertible networks, Neural networks, Segmentation, U-Net",

author = "Christian Etmann and Rihuan Ke and Schonlieb, {Carola Bibiane}",

note = "Funding Information: CE and CBS acknowledge support from the Wellcome Innovator Award RG98755. RK and CBS acknowledge support from the EPSRC grant EP/T003553/1. CE additionally acknowledges partial funding by the Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 281474342: {\textquoteright}RTG pi3 - Parameter Identification - Analysis, Algorithms, Applications{\textquoteright} for parts of the work done while being a member of RTG pi3. CBS additionally acknowledges support from the Leverhulme Trust project on {\textquoteleft}Breaking the non-convexity barrier{\textquoteright}, the Philip Leverhulme Prize, the EPSRC grant EP/S026045/1, the EPSRC Centre Nr. EP/N014588/1, the RISE projects CHiPS and NoMADS, the Cantab Capital Institute for the Mathematics of Information and the Alan Turing Institute. Publisher Copyright: {\textcopyright} 2020 IEEE.; 30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020 ; Conference date: 21-09-2020 Through 24-09-2020",

year = "2020",

month = sep,

doi = "10.1109/MLSP49062.2020.9231874",

language = "English",

series = "IEEE International Workshop on Machine Learning for Signal Processing, MLSP",

publisher = "IEEE Computer Society",

booktitle = "Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020",

address = "United States",

}

Etmann, C, Ke, R & Schonlieb, CB 2020, IUNets: Learnable invertible up-and downsampling for large-scale inverse problems. in Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020., 9231874, IEEE International Workshop on Machine Learning for Signal Processing, MLSP, vol. 2020-September, IEEE Computer Society, 30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020, Virtual, Espoo, Finland, 21/09/20. https://doi.org/10.1109/MLSP49062.2020.9231874

IUNets: Learnable invertible up-and downsampling for large-scale inverse problems. / Etmann, Christian; Ke, Rihuan; Schonlieb, Carola Bibiane.
Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020. IEEE Computer Society, 2020. 9231874 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2020-September).

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

TY - GEN

T1 - IUNets

T2 - 30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020

AU - Etmann, Christian

AU - Ke, Rihuan

AU - Schonlieb, Carola Bibiane

N1 - Funding Information: CE and CBS acknowledge support from the Wellcome Innovator Award RG98755. RK and CBS acknowledge support from the EPSRC grant EP/T003553/1. CE additionally acknowledges partial funding by the Deutsche Forschungsgemeinschaft (DFG) - Projektnummer 281474342: ’RTG pi3 - Parameter Identification - Analysis, Algorithms, Applications’ for parts of the work done while being a member of RTG pi3. CBS additionally acknowledges support from the Leverhulme Trust project on ‘Breaking the non-convexity barrier’, the Philip Leverhulme Prize, the EPSRC grant EP/S026045/1, the EPSRC Centre Nr. EP/N014588/1, the RISE projects CHiPS and NoMADS, the Cantab Capital Institute for the Mathematics of Information and the Alan Turing Institute. Publisher Copyright: © 2020 IEEE.

PY - 2020/9

Y1 - 2020/9

N2 - U-Nets have been established as a standard neural network architecture for image-to-image problems such as segmentation and inverse problems in imaging. For high-dimensional applications, as they for example appear in 3D medical imaging, U-Nets however have prohibitive memory requirements. Here, we present a new fully-invertible U-Net-based architecture called the iUNet, which allows for the application of highly memory-efficient backpropagation procedures. As its main building block, we introduce learnable and invertible up- an downsampling operations. For this, we developed an open-source implementation in Pytorch for 1D, 2D and 3D data.

AB - U-Nets have been established as a standard neural network architecture for image-to-image problems such as segmentation and inverse problems in imaging. For high-dimensional applications, as they for example appear in 3D medical imaging, U-Nets however have prohibitive memory requirements. Here, we present a new fully-invertible U-Net-based architecture called the iUNet, which allows for the application of highly memory-efficient backpropagation procedures. As its main building block, we introduce learnable and invertible up- an downsampling operations. For this, we developed an open-source implementation in Pytorch for 1D, 2D and 3D data.

KW - Inverse problems

KW - Invertible networks

KW - Neural networks

KW - Segmentation

KW - U-Net

UR - http://www.scopus.com/inward/record.url?scp=85096499623&partnerID=8YFLogxK

U2 - 10.1109/MLSP49062.2020.9231874

DO - 10.1109/MLSP49062.2020.9231874

M3 - Conference Contribution (Conference Proceeding)

AN - SCOPUS:85096499623

T3 - IEEE International Workshop on Machine Learning for Signal Processing, MLSP

BT - Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020

PB - IEEE Computer Society

Y2 - 21 September 2020 through 24 September 2020

ER -

Etmann C, Ke R, Schonlieb CB. IUNets: Learnable invertible up-and downsampling for large-scale inverse problems. In Proceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020. IEEE Computer Society. 2020. 9231874. (IEEE International Workshop on Machine Learning for Signal Processing, MLSP). doi: 10.1109/MLSP49062.2020.9231874

IUNets: Learnable invertible up-and downsampling for large-scale inverse problems

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Handle.net

Other files and links

Fingerprint

Cite this