Karelia Pena-Pena

Mrs. Karelia Pena-Pena

University of Delaware, Newark, California, USA
orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 March 2024 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
FWFuli Wang
Fuli Wang
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
WQWei Qian
Wei Qian
Gonzalo ArceGonzalo Arce
Gonzalo Arce
Hypergraph neural networks (HyperGNNs) are a family of deep neural networks designed to perform inference on hypergraphs. HyperGNNs follow either a spectral or a spatial approach, in which a convolution or message-passing operation is conducted based on a hypergraph algebraic descriptor. While many HyperGNNs have been proposed and achieved state-of-the-art performance on broad applications, there have been limited attempts at exploring high-dimensional hypergraph descriptors (tensors) and joint node interactions carried by hyperedges. In this article, we depart from hypergraph matrix representations and present a new tensor-HyperGNN (T-HyperGNN) framework with cross-node interactions (CNIs). The T-HyperGNN framework consists of T-spectral convolution, T-spatial convolution, and T-message-passing HyperGNNs (T-MPHN). The T-spectral convolution HyperGNN is defined under the t-product algebra that closely connects to the spectral space. To improve computational efficiency for large hypergraphs, we localize the T-spectral convolution approach to formulate the T-spatial convolution and further devise a novel tensor-message-passing algorithm for practical implementation by studying a compressed adjacency tensor representation. Compared to the state-of-the-art approaches, our T-HyperGNNs preserve intrinsic high-order network structures without any hypergraph reduction and model the joint effects of nodes through a CNI layer. These advantages of our T-HyperGNNs are demonstrated in a wide range of real-world hypergraph datasets. The implementation code is available at https://github.com/wangfuli/T-HyperGNNs.git.

16 May 2023 in IEEE Transactions on Signal and Information Processing over Networks
Verified
Journal Article
Closed access
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
DLDaniel L. Lau
Daniel L. Lau
Gonzalo ArceGonzalo Arce
Gonzalo Arce
Graph signal processing (GSP) techniques are powerful tools that model complex relationships within large datasets, being now used in a myriad of applications in different areas including data science, communication networks, epidemiology, and sociology. Simple graphs can only model pairwise relationships among data which prevents their application in modeling networks with higher-order relationships. For this reason, some efforts have been made to generalize well-known graph signal processing techniques to more complex graphs such as hypergraphs, which allow capturing higher-order relationships among data. In this paper, we provide a new hypergraph signal processing framework (t-HGSP) based on a novel tensor-tensor product algebra that has emerged as a powerful tool for preserving the intrinsic structures of tensors. The proposed framework allows the generalization of traditional GSP techniques while keeping the dimensionality characteristic of the complex systems represented by hypergraphs. To this end, the core elements of the t-HGSP framework are introduced, including the shifting operators and the hypergraph signal. The hypergraph Fourier space is also defined, followed by the concept of bandlimited signals and sampling. In our experiments, we demonstrate the benefits of our approach in applications such as clustering and denoising.

29 January 2024 in IEEE Transactions on Geoscience and Remote Sensing
Verified
Journal Article
Open access
ARAndres Ramirez-Jaime
Andres Ramirez-Jaime
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
Gonzalo ArceGonzalo Arce
Gonzalo Arce
DHDavid Harding
David Harding
MSMark Stephen
Mark Stephen
JMJames MacKinnon
James MacKinnon
Light detection and ranging (LiDAR) remote sensing systems are deployed in various platforms including satellites, airplanes, and drones—which, in essence, determines the sampling characteristics of the underlying imaging system. Low-altitude LiDARs provide high photon count and high spatial resolution but only in very localized patches. Satellite LiDARs, on the other hand, provide measurements at a global scale but are limited by low photon count and their samples are sparsely apart along swath line trajectories that are far in between. This article describes a new class of satellite remote sensing LiDARs, aimed at overcoming the limitations of current satellite imaging systems. It exploits the principles of compressive sensing and machine learning (ML) to compressively sense Earth from hundreds of kilometers above Earth to then reconstruct the 3-D imagery with resolution and coverage, as if the data were collected from airborne platforms at just hundreds of meters in height. We introduce a novel representation of waveform altimetry profiles, coined hyperheight data cubes (HHDCs), which encompass rich information about the 3-D structure of a scene. Canopy height models (CHMs), digital terrain models (DTMs), and many other features of a scene that are embedded in HHDC are easily extracted with simple statistical quantiles. We introduce ML methods to reconstruct the compressive LiDAR measurements so as to attain high-resolution, dense coverage, and broad field-of-view per swath pass. ML training data are attained from NASA’s G-LiHT imaging missions. Simulations with various types of forests across the US illustrate the power of the new LiDAR imaging systems.

1 December 2021 in Signal Processing
Verified
Journal Article
Closed access
XMXu Ma
Xu Ma
SZShengen Zhang
Shengen Zhang
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
GAGonzalo R. Arce
Gonzalo R. Arce

16 February 2022 in Multimedia Tools and Applications
Verified
Journal Article
Closed access
PDF file
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
DLDaniel L. Lau
Daniel L. Lau
AAAndrew J. Arce
Andrew J. Arce
Gonzalo ArceGonzalo Arce
Gonzalo Arce

29 October 2020 in IEEE Signal Processing Magazine
Verified
Journal Article
Closed access
DLDaniel L. Lau
Daniel L. Lau
Gonzalo ArceGonzalo Arce
Gonzalo Arce
APAlejandro Parada-Mayorga
Alejandro Parada-Mayorga
DDDaniela Dapena
Daniela Dapena
Karelia Pena-PenaKarelia Pena-Pena
Karelia Pena-Pena
With the surge in the volumes and dimensions of data defined in non-Euclidean spaces, graph signal processing (GSP) techniques are emerging as important tools in our understanding of these domains [1]. A fundamental problem for GSP is to determine which nodes play the most important role; so, graph signal sampling and recovery thus become essential [2]. In general, most of the current sampling methods are based on graph spectral decompositions where the graph Fourier transform (GFT) plays a central role [2]. Although adequate in many cases, they are not applicable when the graphs are large and where spectral decompositions are computationally difficult [3]. After years of beautiful and useful theoretical insights developed in this problem, the interest has now centered on finding more efficient methods for the computation of good sampling sets. Looking to the spatial domain for inspiration, substantial research has been performed that looks at the use of spatial point processes to define stochastic sampling grids with a particular interest at point processes that generate "blue noise."