Who We Are Sirani M. Perera
Sirani M. Perera
Associate Professor, Embry Riddle Aeronautical University
Citation
For excellence in applied linear algebra while converging research in applied mathematics, engineering, and theoretical computer science.
Presentation: Efficient Solutions for Real-world Challenges: Low-complexity Algorithms in Applied Linear Algebra for Engineering and Data Science
Addressing complex and large-scale systems, as well as data-intensive problems, presents a considerable challenge when employing conventional methods. Fortunately, one can examine the fundamental structures of these systems to obtain low-complexity machine learning (ML) algorithms that efficiently address real-world challenges.
In the first half of the presentation, we briefly address on AI-driven mathematical theories along with low-complexity ML algorithms tailored for AI chips. Our focus is on the efficient implementation of wideband multi-beam beamformers through the use of delay Vandermonde matrices (DVM). By leveraging the true time delay beamformers through the DVM structure, we significantly reduce spatial and computational complexities via a structured neural network (SNN) architecture. Our findings, both numerical and theoretical, demonstrate that the proposed SNN achieves a reduction of at least 70% in weights for configurations of 32, 64, 128, and 256 beams, all while preserving the accuracy of 10−4 in comparison to the feed-forward neural networks.
In the second half of the talk, we introduce a technique that leverages interpolation incorporating boundary and interior conditions to determine orbital trajectories with a novel low-complexity algorithm without the need for the spacecraft’s acceleration data. Once the algorithm is addressed, we will discuss its accuracy and performance across a variety of periodic orbital trajectories in the Earth-Moon system. We will talk about a comparative analysis to evaluate the time complexity of the proposed algorithm compared with conventional orbit propagators. We show that the proposed algorithm achieves approximately a 70% improvement in complexity compared to existing orbital propagation techniques. Finally, we will show that the proposed algorithm can be utilized to learn and update distant retrograde orbits (DRO) while training a neural network with initial conditions composing minimum predefined data. After the training is done, we show the effectiveness of the learning process to accurately predict DRO trajectories.
This work was a joint work with Hansaka Aluvihare, Brian Baker-McEvilly, Arjuna Madanayake, David Canales, and Xianqi Li
In the first half of the presentation, we briefly address on AI-driven mathematical theories along with low-complexity ML algorithms tailored for AI chips. Our focus is on the efficient implementation of wideband multi-beam beamformers through the use of delay Vandermonde matrices (DVM). By leveraging the true time delay beamformers through the DVM structure, we significantly reduce spatial and computational complexities via a structured neural network (SNN) architecture. Our findings, both numerical and theoretical, demonstrate that the proposed SNN achieves a reduction of at least 70% in weights for configurations of 32, 64, 128, and 256 beams, all while preserving the accuracy of 10−4 in comparison to the feed-forward neural networks.
In the second half of the talk, we introduce a technique that leverages interpolation incorporating boundary and interior conditions to determine orbital trajectories with a novel low-complexity algorithm without the need for the spacecraft’s acceleration data. Once the algorithm is addressed, we will discuss its accuracy and performance across a variety of periodic orbital trajectories in the Earth-Moon system. We will talk about a comparative analysis to evaluate the time complexity of the proposed algorithm compared with conventional orbit propagators. We show that the proposed algorithm achieves approximately a 70% improvement in complexity compared to existing orbital propagation techniques. Finally, we will show that the proposed algorithm can be utilized to learn and update distant retrograde orbits (DRO) while training a neural network with initial conditions composing minimum predefined data. After the training is done, we show the effectiveness of the learning process to accurately predict DRO trajectories.
This work was a joint work with Hansaka Aluvihare, Brian Baker-McEvilly, Arjuna Madanayake, David Canales, and Xianqi Li
Bio
Perera is an extraordinary force, effortlessly converging research in the fields of mathematics, engineering, data science, and theoretical computer science to boost groundbreaking discoveries and innovations. Her
excellence was recognized in 2023 when she was named the ”Convergence Research (CORE) Fellow” by the
NSF CORE Institute at the University of California San Diego. Continuing her stellar trajectory, she was
honored with the Research and Scholarship Excellence Award at the College of Arts and Sciences(COAS)
at ERAU in 2024. What sets Perera apart is her unparalleled expertise and pioneering spirit. She is the only
scholar at ERAU to have secured parallel multiple NSF awards across various divisions, serving as Principal
Investigators (PIs) and CoPIs. Her innovative contributions as an inventor and co-inventor of patents further
elevate her status as a truly exceptional candidate-a rising star in her field.
Among her most revolutionary achievements is her groundbreaking work on state-space dynamical systems. Perera has transcended the boundaries of classical methods to predict and understand future states by proposing novel machine learning (ML) algorithms. These algorithms enable the extraction, realization, and prediction of future dynamics within these systems using low-complexity ML models. Her work has not only advanced mathematical theories but also transformed the modeling of multiple spacecraft, generating time-advanced state trajectories that offer unprecedented insights into the future dynamics of space missions. As the space race intensifies with industry giants like SpaceX, OneWeb, and Blue Origin, Perera’s contributions have been rightfully acknowledged with the prestigious DMS NSF award (#2410676) for her leadership on the project titled Collaborative Research: Data-driven Realization of State-space Dynamical Systems via Low-complexity Algorithms.
Perera has also been a trailblazer in the development of AI-driven mathematical theories and low-complexity ML algorithms for AI-enabled RF sensing. Understanding the immense computational demands of processing massive datasets at GHz rates, she introduced innovative spatiotemporal frequency-based low-rank matrices, paving the way for optimized sensor analysis. Her work on AI-based mathematical theories for RF sensors earned her the esteemed ECCS NSF award (#2229473) for the project Collaborative Research: SWIFT: AI-based Sensing for Improved Resiliency via Spectral Adaptation with Lifelong Learning.
Perera has made significant strides in addressing the longstanding beam squint problem. Her elegant mathematical theories, combined with the development of low-complexity algorithms, led to the efficient realization of wideband multi-beam beamformers using the delay Vandermonde matrices (DVM), which she introduced. This breakthrough reduced the delays in N-wideband multi-beam beamforming from O(N2 ) to O(N log N), positioning it as a superclass to the fast Fourier transform (FFT) based narrowband beamforming. The FFT beams, based on the Discrete Fourier Transform matrix, often suffer from the beam squint problem in Butler Matrix type beamformers. Her exceptional work in DVM algorithms lead to solve the beam squint problem, and has been recognized with ECCS NSF awards #1902283 and #1711625
Beyond her technical achievements, Perera has distinguished herself as a dedicated mentor and advocate for diversity in STEM. She has played a pivotal role in empowering women and underrepresented minorities through her involvement with the Society of Asian Scientists and Engineers. Her commitment to fostering diversity and inclusion is evident in several research grants she has co-led, including Distributed Learning for Undergraduate Programs in Data Science at Diverse Universities (DUE NSF #2142514) and the Exchange of Mathematical Ideas Conference 2023 (DMS NSF #2322922). Additionally, she has served as senior personnel and mentor for the REU: Swarms of Unmanned Aircraft Systems in the Age of AI/Machine Learning (CNS NSF #2150213), nurturing the next generation of researchers.
Perera’s influence and legacy extend far beyond her own work, leaving an indelible mark on the scientific community. Her dedication, innovation, and leadership make her not only a deserving nominee but a true inspiration for future generations in STEM
Among her most revolutionary achievements is her groundbreaking work on state-space dynamical systems. Perera has transcended the boundaries of classical methods to predict and understand future states by proposing novel machine learning (ML) algorithms. These algorithms enable the extraction, realization, and prediction of future dynamics within these systems using low-complexity ML models. Her work has not only advanced mathematical theories but also transformed the modeling of multiple spacecraft, generating time-advanced state trajectories that offer unprecedented insights into the future dynamics of space missions. As the space race intensifies with industry giants like SpaceX, OneWeb, and Blue Origin, Perera’s contributions have been rightfully acknowledged with the prestigious DMS NSF award (#2410676) for her leadership on the project titled Collaborative Research: Data-driven Realization of State-space Dynamical Systems via Low-complexity Algorithms.
Perera has also been a trailblazer in the development of AI-driven mathematical theories and low-complexity ML algorithms for AI-enabled RF sensing. Understanding the immense computational demands of processing massive datasets at GHz rates, she introduced innovative spatiotemporal frequency-based low-rank matrices, paving the way for optimized sensor analysis. Her work on AI-based mathematical theories for RF sensors earned her the esteemed ECCS NSF award (#2229473) for the project Collaborative Research: SWIFT: AI-based Sensing for Improved Resiliency via Spectral Adaptation with Lifelong Learning.
Perera has made significant strides in addressing the longstanding beam squint problem. Her elegant mathematical theories, combined with the development of low-complexity algorithms, led to the efficient realization of wideband multi-beam beamformers using the delay Vandermonde matrices (DVM), which she introduced. This breakthrough reduced the delays in N-wideband multi-beam beamforming from O(N2 ) to O(N log N), positioning it as a superclass to the fast Fourier transform (FFT) based narrowband beamforming. The FFT beams, based on the Discrete Fourier Transform matrix, often suffer from the beam squint problem in Butler Matrix type beamformers. Her exceptional work in DVM algorithms lead to solve the beam squint problem, and has been recognized with ECCS NSF awards #1902283 and #1711625
Beyond her technical achievements, Perera has distinguished herself as a dedicated mentor and advocate for diversity in STEM. She has played a pivotal role in empowering women and underrepresented minorities through her involvement with the Society of Asian Scientists and Engineers. Her commitment to fostering diversity and inclusion is evident in several research grants she has co-led, including Distributed Learning for Undergraduate Programs in Data Science at Diverse Universities (DUE NSF #2142514) and the Exchange of Mathematical Ideas Conference 2023 (DMS NSF #2322922). Additionally, she has served as senior personnel and mentor for the REU: Swarms of Unmanned Aircraft Systems in the Age of AI/Machine Learning (CNS NSF #2150213), nurturing the next generation of researchers.
Perera’s influence and legacy extend far beyond her own work, leaving an indelible mark on the scientific community. Her dedication, innovation, and leadership make her not only a deserving nominee but a true inspiration for future generations in STEM