ECE Seminar Series (GSA) – Three Mini Talks by ECE PhD Students
Location: Engineering Science Building (ESB), Room 1001
Come at 1:30p for Reception and Conversation
LECTURE at the ECE SEMINAR SERIES
Three short talks by UCSB PhD students Roark Chao, Poorva Shukla, and Abdelrahman S. Abdelrahman presenting a wide array of ECE topics from photonics to dynamical systems to quantum algorithms. All three talks will be accessible to audiences of all backgrounds.
"Engineering Visible Luminescence via Metasurfaces," Roark Chao, ECE PhD Student, Schuller Lab
Abstract
Luminescence, the processes by which light is generated, is understood as a combination of electric and magnetic multipoles. This talk discusses various methods we use to exploit these multipolar mechanisms of light generation to achieve high performance optoelectronic devices. In particular, metasurface-enabled GaN microLEDs will be explored from the perspective of optical data centers and visible light communication.
"Geometrical Localization," Poorva Shukla, PhD Student, Bamieh Lab
Abstract
We study perturbations of a sparse circulant Laplacian L_C by a matrix E which has a small number of nonzero rows. The resulting operator L = L_C + E models a sparse, structured graph containing localized irregularities induced by E. We establish necessary and sufficient conditions for the localization of eigenvectors of L, where localization refers to the concentration of eigenvector mass on a small subset of nodes (i.e., pseudo-sparsity). This phenomenon—which we term geometrical localization—arises directly from the structural heterogeneities introduced by E. Our analysis shows that (i) the high-mass regions of localized eigenvectors are confined to the neighborhood of the support of E, and (ii) these regions remain invariant under changes in boundary conditions, provided the support of E is fixed. The results give a complete description of how finite-row perturbations of structured operators generate localized modes. From a theoretical standpoint, the work advances the spectral analysis of large structured operators subject to perturbations with a small support. On the applied side, it provides a framework for understanding how localized defects shape spectral behavior in physical, biological, and networked systems.
Shukla's Mechanical Engineering Profile
"Learning the Sampler: The Probabilistic Approximate Optimization Algorithm," Abdelrahman S. Abdelrahman, PhD Student, Çamsarı Lab
Abstract
Monte Carlo Markov Chain (MCMC) methods are standard for navigating glassy energy landscapes in combinatorial optimization and statistical physics. Classical simulated annealing (SA) employs a fixed sampling rule, which can struggle on rugged landscapes. Inspired by the Quantum Approximate Optimization Algorithm (QAOA), we introduce the Generalized Probabilistic Approximate Optimization Algorithm (PAOA), a variational MCMC framework that learns its sampling distribution instead of fixing it. We show that (i) under constrained parameterization, PAOA recovers SA; (ii) on the canonical N = 26 Sherrington–Kirkpatrick model, PAOA outperforms QAOA; and (iii) on SK with Lévy bonds, PAOA learns multiple temperature schedules, yielding improved performance over optimized SA. I will outline the framework and discuss its implications for adaptive, device-aware optimization.
Abdelrahman's Çamsarı Lab Profile
Hosted by: Lecture at the ECE Seminar Series
Submitted by: ECE Graduate Student Association
