LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism

Adrian Stern; Vladislav Kravets

doi:10.1117/12.3053974

LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism

Adrian Stern
, Vladislav Kravets

School of Electrical and Computer Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

The Partial Random Ensemble (PRE) is one of the two main approaches for designing Compressive Sensing (CS) matrices, along with the random modulations approach. In traditional CS literature, various methods for PRE have been proposed to generate CS sensing matrices using different random sampling schemes. Recently, we introduced LPTNet, which uses a model-based deep learning approach to jointly optimize the PRE matrix and a corresponding reconstruction deep neural network (DNN). LPTNet has demonstrated unprecedented CS performance. In this paper, we provide a review of LPTNet and present an interpretable scheme for its inferences DNN by reformulating it as an unfolded DNN that implements an iterative proximal gradient descent algorithm.

Original language	English
Title of host publication	Machine Learning from Challenging Data 2025
Editors	Panagiotis Markopoulos, Bing Ouyang, George Sklivanitis
Publisher	SPIE
ISBN (Electronic)	9781510687097
DOIs	https://doi.org/10.1117/12.3053974
State	Published - 1 Jan 2025
Event	Machine Learning from Challenging Data 2025 - Orlando, United States Duration: 14 Apr 2025 → 15 Apr 2025

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	13460
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	Machine Learning from Challenging Data 2025
Country/Territory	United States
City	Orlando
Period	14/04/25 → 15/04/25

Keywords

Compressive Imaging
Deep Learning
Neural Networks
proximal gradient descent
unfolded neural networks
unrolled neural networks

ASJC Scopus subject areas

Electronic, Optical and Magnetic Materials
Condensed Matter Physics
Computer Science Applications
Applied Mathematics
Electrical and Electronic Engineering

Access to Document

10.1117/12.3053974

Cite this

@inproceedings{c6d34e1be5bc4297975e697f3c38cd45,

title = "LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism",

abstract = "The Partial Random Ensemble (PRE) is one of the two main approaches for designing Compressive Sensing (CS) matrices, along with the random modulations approach. In traditional CS literature, various methods for PRE have been proposed to generate CS sensing matrices using different random sampling schemes. Recently, we introduced LPTNet, which uses a model-based deep learning approach to jointly optimize the PRE matrix and a corresponding reconstruction deep neural network (DNN). LPTNet has demonstrated unprecedented CS performance. In this paper, we provide a review of LPTNet and present an interpretable scheme for its inferences DNN by reformulating it as an unfolded DNN that implements an iterative proximal gradient descent algorithm.",

keywords = "Compressive Imaging, Deep Learning, Neural Networks, proximal gradient descent, unfolded neural networks, unrolled neural networks",

author = "Adrian Stern and Vladislav Kravets",

year = "2025",

month = jan,

day = "1",

doi = "10.1117/12.3053974",

language = "English",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Panagiotis Markopoulos and Bing Ouyang and George Sklivanitis",

booktitle = "Machine Learning from Challenging Data 2025",

address = "United States",

}

Stern, A & Kravets, V 2025, LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism. in P Markopoulos, B Ouyang & G Sklivanitis (eds), Machine Learning from Challenging Data 2025., 134600C, Proceedings of SPIE - The International Society for Optical Engineering, vol. 13460, SPIE, Machine Learning from Challenging Data 2025, Orlando, United States, 14/04/25. https://doi.org/10.1117/12.3053974

LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism. / Stern, Adrian; Kravets, Vladislav.
Machine Learning from Challenging Data 2025. ed. / Panagiotis Markopoulos; Bing Ouyang; George Sklivanitis. SPIE, 2025. 134600C (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13460).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - LPTNet for Partial Ensemble Compressive Sensing

T2 - Machine Learning from Challenging Data 2025

AU - Stern, Adrian

AU - Kravets, Vladislav

PY - 2025/1/1

Y1 - 2025/1/1

N2 - The Partial Random Ensemble (PRE) is one of the two main approaches for designing Compressive Sensing (CS) matrices, along with the random modulations approach. In traditional CS literature, various methods for PRE have been proposed to generate CS sensing matrices using different random sampling schemes. Recently, we introduced LPTNet, which uses a model-based deep learning approach to jointly optimize the PRE matrix and a corresponding reconstruction deep neural network (DNN). LPTNet has demonstrated unprecedented CS performance. In this paper, we provide a review of LPTNet and present an interpretable scheme for its inferences DNN by reformulating it as an unfolded DNN that implements an iterative proximal gradient descent algorithm.

AB - The Partial Random Ensemble (PRE) is one of the two main approaches for designing Compressive Sensing (CS) matrices, along with the random modulations approach. In traditional CS literature, various methods for PRE have been proposed to generate CS sensing matrices using different random sampling schemes. Recently, we introduced LPTNet, which uses a model-based deep learning approach to jointly optimize the PRE matrix and a corresponding reconstruction deep neural network (DNN). LPTNet has demonstrated unprecedented CS performance. In this paper, we provide a review of LPTNet and present an interpretable scheme for its inferences DNN by reformulating it as an unfolded DNN that implements an iterative proximal gradient descent algorithm.

KW - Compressive Imaging

KW - Deep Learning

KW - Neural Networks

KW - proximal gradient descent

KW - unfolded neural networks

KW - unrolled neural networks

UR - https://www.scopus.com/pages/publications/105010473281

U2 - 10.1117/12.3053974

DO - 10.1117/12.3053974

M3 - Conference contribution

AN - SCOPUS:105010473281

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Machine Learning from Challenging Data 2025

A2 - Markopoulos, Panagiotis

A2 - Ouyang, Bing

A2 - Sklivanitis, George

PB - SPIE

Y2 - 14 April 2025 through 15 April 2025

ER -

LPTNet for Partial Ensemble Compressive Sensing: An Unfolded Formalism

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this