Research

Our laboratory develops comprehensive artificial intelligence frameworks that integrate multiple disciplines including deep learning, machine learning, generative AI, large language models, natural language processing, and data science to revolutionize materials research and discovery. We focus on creating end-to-end AI-driven solutions that span the entire materials development pipeline, from inverse design approaches that identify materials with target functionalities to advanced prediction models for critical properties such as formation energies, phonon dispersions, and molecular dynamics behavior. Our research encompasses synthesizability prediction models that guide experimental feasibility, high-throughput screening methods that accelerate materials characterization, and intelligent recipe recommendation systems that optimize synthesis pathways. By combining first-principles density functional theory calculations with state-of-the-art machine learning algorithms, we bridge the gap between computational predictions and experimental validation, enabling rapid discovery and optimization of materials for energy storage, catalysis, electronics, and other technological applications. This interdisciplinary approach leverages the power of modern AI to transform traditional materials science workflows into efficient, predictive, and autonomous discovery platforms.

Our research focuses on the development of advanced hybrid quantum-classical algorithms that bridge traditional machine learning and emerging quantum computing paradigms. We design quantum machine learning frameworks that leverage variational circuits, quantum feature spaces, and entanglement-based representations to address complex problems in cybersecurity, medicine, and materials discovery. Current efforts involve algorithmic innovation in quantum-enhanced classification, with an emphasis on interpretable and resource-efficient designs. Moving forward, our work aims to extend these theoretical and simulation-based models onto real quantum hardware, validating their scalability and robustness under physical constraints. By integrating algorithmic theory, quantum information science, and applied AI, our research seeks to establish practical pathways toward quantum-accelerated intelligence for secure, data-driven, and discovery-oriented applications.

Our research explores the transformation from biological neural systems to engineered neural networks, with a focus on energy-efficient neuromorphic computing architectures. Inspired by the brain's synaptic processing, we develop novel hardware implementations that emulate neuron-synapse behavior for AI tasks.

We investigate two main directions:

• Electronic Neuromorphic: Design using memristor crossbars for in-memory analog matrix multiplication, enabling parallel, low-power computation.

• Photonic Neuromorphic: Design leveraging Mach-Zehnder Interferometers and microring resonators for ultrafast, light-based neural computation with programmable nonlinear activation.

By combining bio-inspired models with next-generation hardware—including both electronic and photonic substrates—we aim to accelerate AI performance while drastically reducing energy consumption.

For more details about each project, please check our publications.

Our research advances neuromorphic intelligence for autonomous systems through various complementary studies that collectively establish the foundation for fully event-driven estimation, perception, and control in aerial and robotic applications.

• The first study, Event-Based Heterogeneous Information Processing for Vision-Based Obstacle Detection and Localization, develops a hybrid ANN–SNN vision model that fuses continuous deep-learning pathways with spike-driven temporal encoding for dynamic scene understanding. By combining spatial feature extraction and event-based motion cues, this framework achieves real-time obstacle detection and localization, enabling energy-efficient perception for UAVs and robotic systems operating in complex and visually dynamic environments.

• The second study, Neuromorphic Robust Framework for Integrated Estimation and Control, introduces the SNN-LQR-EMSIF, a spiking neural network formulation of the Extended Modified Sliding Innovation Filter integrated with optimal control laws. This architecture performs simultaneous state estimation and control signal generation directly from neural spike dynamics, offering real-time adaptability, stability under uncertainty, and substantially reduced computational demand compared to conventional Kalman-based approaches.

• The third study, Neuromorphic Digital-Twin-Based Controller for Multi-UAV Deployment, establishes a distributed cloud-edge control architecture where each UAV is equipped with an individual SNN capable of learning and reproducing cloud-generated control commands. This digital-twin framework maintains formation control and collision avoidance even under communication loss, ensuring resilient, congestion-free coordination across swarms of UAVs operating in three-dimensional urban or indoor spaces.

Together, these research directions form a unified neuromorphic paradigm that bridges robust estimation, event-driven perception, and decentralized control. They demonstrate how biologically inspired neural computation can transform traditional robotic workflows into adaptive, scalable, and energy-efficient autonomous systems.

Our research focuses on developing advanced machine learning frameworks for materials discovery and design powered by generative artificial intelligence. We leverage variational autoencoders (VAEs) and generative adversarial networks (GANs) to capture complex representations of materials in both real and latent space, enabling the prediction and generation of novel structures with targeted properties.

To ensure the physical validity and stability of generated candidates, our computational pipeline integrates multiple validation approaches, including density functional theory (DFT) calculations and ALIGNN pre-evaluation methods. The framework combines training processes that minimize formation energy predictions with critic–discriminator networks that assess material feasibility using Wasserstein distance metrics.

This approach allows us to efficiently explore vast chemical spaces, reconstruct materials from compressed representations, and propose new candidates that can be systematically validated through first-principles calculations. Ultimately, our work accelerates the discovery of functional materials for energy, electronics, and other emerging technologies.

Our research develops an advanced multi-algorithm control framework for autonomous multi-UAV systems that integrates biological inspiration with rigorous mathematical modeling to achieve reliable, collision-free operation in dynamic 3D environments. The framework combines a probabilistic Lloyd’s algorithm for optimal formation control based on centroidal Voronoi tessellation, an elastic-field collision avoidance algorithm that models inter-vehicle interactions through adaptive spring-damper dynamics, and a pigeon-inspired 3D obstacle detection and avoidance algorithm that introduces a new dimension of maneuverability. Unlike conventional planar approaches that often experience congestion and temporary stagnation when agents share a single motion plane, the proposed algorithm enables UAVs to autonomously generate non-planar escape trajectories through conical field-of-view modeling, gradient-based potential fields, and spatial rotation matrices. This added vertical degree of freedom allows vehicles to resolve local conflicts, maintain formation integrity, and ensure smooth recovery after avoidance maneuvers. Extensive Monte-Carlo simulations demonstrate 100% mission success across complex urban scenarios, validating the scalability, adaptability, and safety of the proposed algorithmic framework for next-generation cooperative aerial missions.