Seung-Jun Kim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Seung-Jun Kim has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Georgios B. Giannakis
Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, USA
6 shared publications - Daniel Romero
Department of Information and Communication Technology, University of Agder, Grimstad, Norway
2 shared publications - Rami Mowakeaa
Dept. of Computer Science and Electrical Engineering, University of Maryland, Baltimore, MD, USA
2 shared publications Swayambhoo Jain
2 shared publications- William Clarence Walton
University of Maryland, Baltimore County, CSEE Department, Baltimore, USA
2 shared publications - Beibei Chen
Department of Clinical Sciences, University of Texas Southwestern Medical Center, Dallas, United States
1 shared publication
Latest Publications
- Open Access
- 23 September 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38  (3) , 1829-1845
Uncertainty Estimation for Dual View X-ray Mammographic Image Registration Using Deep Ensembles
Techniques are developed for generating uncertainty estimates for convolutional neural network (CNN)-based methods for registering the locations of lesions between the craniocaudal (CC) and mediolateral oblique (MLO) mammographic X-ray image views. Multi-view lesion correspondence is an important task that clinicians perform for characterizing lesions during routine mammographic exams. Automated registration tools can aid in this task, yet if the tools also provide confidence estimates, they can be of greater value to clinicians, especially in cases involving dense tissue where lesions may be difficult to see. A set of deep ensemble-based techniques, which leverage a negative log-likelihood (NLL)-based cost function, are implemented for estimating uncertainties. The ensemble architectures involve significant modifications to an existing CNN dual-view lesion registration algorithm. Three architectural designs are evaluated, and different ensemble sizes are compared using various performance metrics. The techniques are tested on synthetic X-ray data, real 2D X-ray data, and slices from real 3D X-ray data. The ensembles generate covariance-based uncertainty ellipses that are correlated with registration accuracy, such that the ellipse sizes can give a clinician an indication of confidence in the mapping between the CC and MLO views. The results also show that the ellipse sizes can aid in improving computer-aided detection (CAD) results by matching CC/MLO lesion detects and reducing false alarms from both views, adding to clinical utility. The uncertainty estimation techniques show promise as a means for aiding clinicians in confidently establishing multi-view lesion correspondence, thereby improving diagnostic capability.
- 23 September 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38  (3) , 1829-1845
Open AccessUncertainty Estimation for Dual View X-ray Mammographic Image Registration Using Deep Ensembles
Techniques are developed for generating uncertainty estimates for convolutional neural network (CNN)-based methods for registering the locations of lesions between the craniocaudal (CC) and mediolateral oblique (MLO) mammographic X-ray image views. Multi-view lesion correspondence is an important task that clinicians perform for characterizing lesions during routine mammographic exams. Automated registration tools can aid in this task, yet if the tools also provide confidence estimates, they can be of greater value to clinicians, especially in cases involving dense tissue where lesions may be difficult to see. A set of deep ensemble-based techniques, which leverage a negative log-likelihood (NLL)-based cost function, are implemented for estimating uncertainties. The ensemble architectures involve significant modifications to an existing CNN dual-view lesion registration algorithm. Three architectural designs are evaluated, and different ensemble sizes are compared using various performance metrics. The techniques are tested on synthetic X-ray data, real 2D X-ray data, and slices from real 3D X-ray data. The ensembles generate covariance-based uncertainty ellipses that are correlated with registration accuracy, such that the ellipse sizes can give a clinician an indication of confidence in the mapping between the CC and MLO views. The results also show that the ellipse sizes can aid in improving computer-aided detection (CAD) results by matching CC/MLO lesion detects and reducing false alarms from both views, adding to clinical utility. The uncertainty estimation techniques show promise as a means for aiding clinicians in confidently establishing multi-view lesion correspondence, thereby improving diagnostic capability.
- Open Access
- 15 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 27816-27828
Online Robust Subspace Clustering With Application to Power Grid Monitoring
In this work, a robust subspace clustering algorithm is developed to exploit the inherent union-of-subspaces structure in the data for reconstructing missing measurements and detecting anomalies. Our focus is on processing an incessant stream of large-scale data such as synchronized phasor measurements in the power grid, which is challenging due to computational complexity, memory requirement, and missing and corrupt observations. In order to mitigate these issues, a low-rank representation (LRR) model-based subspace clustering problem is formulated that can handle missing measurements and sparse outliers in the data. Then, an efficient online algorithm is derived based on stochastic approximation. The convergence property of the algorithm is established. Strategies to maintain a representative yet compact dictionary for capturing the subspace structure are also proposed. The developed method is tested on both simulated and real phasor measurement unit (PMU) data to verify the effectiveness and is shown to significantly outperform existing algorithms based on simple low-rank structure of data.
- 15 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 27816-27828
Open AccessOnline Robust Subspace Clustering With Application to Power Grid Monitoring
In this work, a robust subspace clustering algorithm is developed to exploit the inherent union-of-subspaces structure in the data for reconstructing missing measurements and detecting anomalies. Our focus is on processing an incessant stream of large-scale data such as synchronized phasor measurements in the power grid, which is challenging due to computational complexity, memory requirement, and missing and corrupt observations. In order to mitigate these issues, a low-rank representation (LRR) model-based subspace clustering problem is formulated that can handle missing measurements and sparse outliers in the data. Then, an efficient online algorithm is derived based on stochastic approximation. The convergence property of the algorithm is established. Strategies to maintain a representative yet compact dictionary for capturing the subspace structure are also proposed. The developed method is tested on both simulated and real phasor measurement unit (PMU) data to verify the effectiveness and is shown to significantly outperform existing algorithms based on simple low-rank structure of data.
- Top Cited Papers
- 27 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Magazine
- Vol. 39  (6) , 53-72
Radio Map Estimation: A data-driven approach to spectrum cartography
Radio maps characterize quantities of interest in radio communication environments, such as the received signal strength and channel attenuation, at every point of a geographical region. Radio map estimation (RME) typically entails interpolative inference based on spatially distributed measurements. In this tutorial article, after presenting some representative applications of radio maps, the most prominent RME methods are discussed. Starting from simple regression, the exposition gradually delves into more sophisticated algorithms, eventually touching upon state-of-the-art techniques. To gain insight into this versatile toolkit, illustrative toy examples will also be presented.
- 27 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Magazine
- Vol. 39  (6) , 53-72
Top Cited PapersRadio Map Estimation: A data-driven approach to spectrum cartography
Radio maps characterize quantities of interest in radio communication environments, such as the received signal strength and channel attenuation, at every point of a geographical region. Radio map estimation (RME) typically entails interpolative inference based on spatially distributed measurements. In this tutorial article, after presenting some representative applications of radio maps, the most prominent RME methods are discussed. Starting from simple regression, the exposition gradually delves into more sophisticated algorithms, eventually touching upon state-of-the-art techniques. To gain insight into this versatile toolkit, illustrative toy examples will also be presented.
- 6 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 69  (11) , 3538-3550
Automated Registration for Dual-View X-Ray Mammography Using Convolutional Neural Networks
Objective: Automated registration algorithms for a pair of 2D X-ray mammographic images taken from two standard imaging angles, namely, the craniocaudal (CC) and the mediolateral oblique (MLO) views, are developed. Methods: A fully convolutional neural network, a type of convolutional neural network (CNN), is employed to generate a pixel-level deformation field, which provides a mapping between masses in the two views. Novel distance-based regularization is employed, which contributes significantly to the performance. Results: The developed techniques are tested using real 2D mammographic images, slices from real 3D mammographic images, and synthetic mammographic images. Architectural variations of the neural network are investigated and the performance is characterized from various aspects including image resolution, breast density, lesion size, lesion subtlety, and lesion Breast Imaging-Reporting and Data System (BI-RADS) category. Our network outperformed the state-of-the-art CNN-based and non-CNN-based registration techniques, and showed robust performance across various tissue/lesion characteristics. Conclusion: The proposed methods provide a useful automated tool for co-locating lesions between the CC and MLO views even in challenging cases. Significance: Our methods can aid clinicians to establish lesion correspondence quickly and accurately in the dual-view X-ray mammography, improving diagnostic capability.
- 6 May 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 69  (11) , 3538-3550
Automated Registration for Dual-View X-Ray Mammography Using Convolutional Neural Networks
Objective: Automated registration algorithms for a pair of 2D X-ray mammographic images taken from two standard imaging angles, namely, the craniocaudal (CC) and the mediolateral oblique (MLO) views, are developed. Methods: A fully convolutional neural network, a type of convolutional neural network (CNN), is employed to generate a pixel-level deformation field, which provides a mapping between masses in the two views. Novel distance-based regularization is employed, which contributes significantly to the performance. Results: The developed techniques are tested using real 2D mammographic images, slices from real 3D mammographic images, and synthetic mammographic images. Architectural variations of the neural network are investigated and the performance is characterized from various aspects including image resolution, breast density, lesion size, lesion subtlety, and lesion Breast Imaging-Reporting and Data System (BI-RADS) category. Our network outperformed the state-of-the-art CNN-based and non-CNN-based registration techniques, and showed robust performance across various tissue/lesion characteristics. Conclusion: The proposed methods provide a useful automated tool for co-locating lesions between the CC and MLO views even in challenging cases. Significance: Our methods can aid clinicians to establish lesion correspondence quickly and accurately in the dual-view X-ray mammography, improving diagnostic capability.
- Open Access
- 15 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 141107-141120
Robust RF Mixture Signal Recognition Using Discriminative Dictionary Learning
RF signal recognition is an important element toward RF situational awareness and dynamic spectrum management. In this work, machine learning-based signal recognition algorithms are proposed. Our key contribution is to engineer feature learning such that the classifiers can perform robustly even when a mixture of heterogeneous signal classes is observed, although the training is still done using non-mixture single-label samples. To achieve this, discriminative dictionary learning algorithms are developed with various feature-shaping constraints. The signal detection can then be done in a way reminiscent of the multi-user detection in wireless communication, employing linear equalizers. The algorithms are tested using real wideband RF measurement data. It is verified that the proposed algorithms can robustly classify the component signals even when their powers are widely different and their number is not known a priori.
- 15 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 141107-141120
Open AccessRobust RF Mixture Signal Recognition Using Discriminative Dictionary Learning
RF signal recognition is an important element toward RF situational awareness and dynamic spectrum management. In this work, machine learning-based signal recognition algorithms are proposed. Our key contribution is to engineer feature learning such that the classifiers can perform robustly even when a mixture of heterogeneous signal classes is observed, although the training is still done using non-mixture single-label samples. To achieve this, discriminative dictionary learning algorithms are developed with various feature-shaping constraints. The signal detection can then be done in a way reminiscent of the multi-user detection in wireless communication, employing linear equalizers. The algorithms are tested using real wideband RF measurement data. It is verified that the proposed algorithms can robustly classify the component signals even when their powers are widely different and their number is not known a priori.
- Open Access
- 6 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 3500-3504
Kernearl-Based Lifelong Policy Gradient Reinforcement Learning
Policy gradient methods have been widely used in reinforcement learning (RL), especially thanks to their facility to handle continuous state spaces, strong convergence guarantees, and low-complexity updates. Training of the methods for individual tasks, however, can still be taxing in terms of the learning speed and the sample trajectory collection. Lifelong learning aims to exploit the intrinsic structure shared among a suite of RL tasks, akin to multitask learning, but in an efficient online fashion. In this work, we propose a lifelong RL algorithm based on the kernel method to leverage nonlinear features of the data based on a popular union-of-subspace model. Experimental results on a set of simple related tasks verify the advantage of the proposed strategy, compared to the single-task and the parametric counterparts.
- 6 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 3500-3504
Open AccessKernearl-Based Lifelong Policy Gradient Reinforcement Learning
Policy gradient methods have been widely used in reinforcement learning (RL), especially thanks to their facility to handle continuous state spaces, strong convergence guarantees, and low-complexity updates. Training of the methods for individual tasks, however, can still be taxing in terms of the learning speed and the sample trajectory collection. Lifelong learning aims to exploit the intrinsic structure shared among a suite of RL tasks, akin to multitask learning, but in an efficient online fashion. In this work, we propose a lifelong RL algorithm based on the kernel method to leverage nonlinear features of the data based on a popular union-of-subspace model. Experimental results on a set of simple related tasks verify the advantage of the proposed strategy, compared to the single-task and the parametric counterparts.
- 4 May 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Signal Processing
- Vol. 14  (6) , 1265-1279
Dictionary Learning-Based fMRI Data Analysis for Capturing Common and Individual Neural Activation Maps
In this paper, a novel dictionary learning (DL) method is proposed to estimate sparse neural activations from multi-subject fMRI data sets. By exploiting the label information such as the patient and the normal healthy groups, the activation maps that are commonly shared across the groups as well as those that can explain the group differences are both captured. The proposed method was tested using real fMRI data sets consisting of schizophrenic subjects and healthy controls. The DL approach not only reproduced most of the maps obtained from the conventional independent component analysis (ICA), but also identified more maps that are significantly group-different, including a number of novel ones that were not revealed by ICA. The stability analysis of the DL method and the correlation analysis with separate neuropsychological test scores further strengthen the validity of our analysis.
- 4 May 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Signal Processing
- Vol. 14  (6) , 1265-1279
Dictionary Learning-Based fMRI Data Analysis for Capturing Common and Individual Neural Activation Maps
In this paper, a novel dictionary learning (DL) method is proposed to estimate sparse neural activations from multi-subject fMRI data sets. By exploiting the label information such as the patient and the normal healthy groups, the activation maps that are commonly shared across the groups as well as those that can explain the group differences are both captured. The proposed method was tested using real fMRI data sets consisting of schizophrenic subjects and healthy controls. The DL approach not only reproduced most of the maps obtained from the conventional independent component analysis (ICA), but also identified more maps that are significantly group-different, including a number of novel ones that were not revealed by ICA. The stability analysis of the DL method and the correlation analysis with separate neuropsychological test scores further strengthen the validity of our analysis.
- Open Access
- 28 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 82131-82141
Building Energy Management for Demand Response Using Kernel Lifelong Learning
Demand response (DR) aims at improving the reliability and efficiency of the power grids by shaping the power demand over time. Given that building energy consumption constitutes a significant portion of the overall grid load, building energy management is a critical component for the DR portfolio. In this study, DR control policies for lighting and air-conditioner systems for the individual spaces in buildings are proposed. The policies are designed to achieve the energy reduction amount specified in the DR request while minimizing the user discomfort. A significant challenge is to cope with the uncertainty of various environmental factors such as the solar illuminance and ambient temperature, as well as the psycho-economic factors such as the energy usage preferences of the occupants. We employ a data-driven machine learning approach to tackle this challenge. Our novel idea is to take advantage of the structural similarity of the control policies across the spaces in a lifelong multi-task learning framework. To accommodate significant nonlinearity in efficient policies, a kernel-based learning approach is pursued. The dual decomposition method is employed to relax the constraint coupled across the spaces, which allows solving the overall learning problem via a series of unconstrained subproblems. The efficacy of the proposed method is verified by numerical experiments based on semi-real data sets.
- 28 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 82131-82141
Open AccessBuilding Energy Management for Demand Response Using Kernel Lifelong Learning
Demand response (DR) aims at improving the reliability and efficiency of the power grids by shaping the power demand over time. Given that building energy consumption constitutes a significant portion of the overall grid load, building energy management is a critical component for the DR portfolio. In this study, DR control policies for lighting and air-conditioner systems for the individual spaces in buildings are proposed. The policies are designed to achieve the energy reduction amount specified in the DR request while minimizing the user discomfort. A significant challenge is to cope with the uncertainty of various environmental factors such as the solar illuminance and ambient temperature, as well as the psycho-economic factors such as the energy usage preferences of the occupants. We employ a data-driven machine learning approach to tackle this challenge. Our novel idea is to take advantage of the structural similarity of the control policies across the spaces in a lifelong multi-task learning framework. To accommodate significant nonlinearity in efficient policies, a kernel-based learning approach is pursued. The dual decomposition method is employed to relax the constraint coupled across the spaces, which allows solving the overall learning problem via a series of unconstrained subproblems. The efficacy of the proposed method is verified by numerical experiments based on semi-real data sets.
- Open Access
Learning Power Spectrum Maps from Quantized Power Measurements
Power spectral density (PSD) maps providing the distribution of RF power across space and frequency are constructed using power measurements collected by a network of low-cost sensors. By introducing linear compression and quantization to a small number of bits, sensor measurements can be communicated to the fusion center with minimal bandwidth requirements. Strengths of data- and model-driven approaches are combined to develop estimators capable of incorporating multiple forms of spectral and propagation prior information while fitting the rapid variations of shadow fading across space. To this end, novel nonparametric and semiparametric formulations are investigated. It is shown that PSD maps can be obtained using support vector machine-type solvers. In addition to batch approaches, an online algorithm attuned to real-time operation is developed. Numerical tests assess the performance of the novel algorithms.
Open AccessLearning Power Spectrum Maps from Quantized Power Measurements
Power spectral density (PSD) maps providing the distribution of RF power across space and frequency are constructed using power measurements collected by a network of low-cost sensors. By introducing linear compression and quantization to a small number of bits, sensor measurements can be communicated to the fusion center with minimal bandwidth requirements. Strengths of data- and model-driven approaches are combined to develop estimators capable of incorporating multiple forms of spectral and propagation prior information while fitting the rapid variations of shadow fading across space. To this end, novel nonparametric and semiparametric formulations are investigated. It is shown that PSD maps can be obtained using support vector machine-type solvers. In addition to batch approaches, an online algorithm attuned to real-time operation is developed. Numerical tests assess the performance of the novel algorithms.
- Open Access
Backhaul-Constrained Multi-Cell Cooperation Leveraging Sparsity and Spectral Clustering
Multi-cell cooperative processing with limited backhaul traffic is studied for cellular uplinks. Aiming at reduced backhaul overhead, a sparsity-regularized multi-cell receive-filter design problem is formulated. Both unstructured distributed cooperation as well as clustered cooperation, in which base station groups are formed for tight cooperation, are considered. Dynamic clustered cooperation, where the sparse equalizer and the cooperation clusters are jointly determined, is solved via alternating minimization based on spectral clustering and group-sparse regression. Furthermore, decentralized implementations of both unstructured and clustered cooperation schemes are developed for scalability, robustness and computational efficiency. Extensive numerical tests verify the efficacy of the proposed methods.
Open AccessBackhaul-Constrained Multi-Cell Cooperation Leveraging Sparsity and Spectral Clustering
Multi-cell cooperative processing with limited backhaul traffic is studied for cellular uplinks. Aiming at reduced backhaul overhead, a sparsity-regularized multi-cell receive-filter design problem is formulated. Both unstructured distributed cooperation as well as clustered cooperation, in which base station groups are formed for tight cooperation, are considered. Dynamic clustered cooperation, where the sparse equalizer and the cooperation clusters are jointly determined, is solved via alternating minimization based on spectral clustering and group-sparse regression. Furthermore, decentralized implementations of both unstructured and clustered cooperation schemes are developed for scalability, robustness and computational efficiency. Extensive numerical tests verify the efficacy of the proposed methods.