Research

Selected publications and research areas across computer vision, signal processing, large language models, geoscience, and healthcare.

My research focuses on efficient and structure-aware learning methods for scientific and real-world sensing problems, with recurring themes in efficiency, limited supervision, and geometry-aware modeling for deployment in ecology and healthcare.

Full publication list on Google Scholar. † indicates first authorship, and * indicates corresponding authorship.

Research Topics

Geospatial AI for Ecological Monitoring and Environmental Change

I develop efficient pipelines for large-scale ecological and environmental questions, including palm detection and spatial distribution modeling, illegal mining detection, and remote sensing understanding under limited labels and deployment constraints.

NBC News The Today Show

AwakeForest: An Interactive Geospatial Platform for Large-Scale Forest Imagery|[IJCAI Demo'26]

Learning Where to Embed: Noise-Aware Positional Embedding for Query Retrieval in Small-Object Detection|[ICMR'26]|Paper|Preprint|Code

Center-guided Classifier for Semantic Segmentation of Remote Sensing Images|[TGRS'26]|Paper|Preprint

From Orthomosaics to Raw UAV Imagery: Enhancing Palm Detection and Crown-Center Localization|[IGARSS'26]|Preprint

MoSAiC: Multi-Modal Multi-Label Supervision-Aware Contrastive Learning for Remote Sensing|[arXiv'25]|Preprint

Efficient Localization and Spatial Distribution Modeling of Canopy Palms Using UAV Imagery|[TGRS'25]|Paper|Preprint

Detection and Geographic Localization of Natural Objects in the Wild: A Case Study on Palms|[IJCAI'25, Oral]|Paper

PalmProbNet: A Probabilistic Approach to Understanding Palm Distributions in Ecuadorian Tropical Forest via Transfer Learning|[ACMSE'24]|Paper|Preprint

Change Detection of Amazonian Alluvial Gold Mining Using Deep Learning and Sentinel-2 Imagery|[RS'22]|Paper

Label-Efficient and Geometry-Aware Learning for Hyperspectral Imaging

I study how spatial structure, diffusion geometry, superpixels, and multi-view representations can improve learning from high-dimensional hyperspectral data when supervision is limited or unavailable.

Superpixel-Level Hypergraph Regularized Multi-View Clustering for Hyperspectral Images|[JSTARS'26]|Paper

Superpixel-based and Spatially-regularized Diffusion Learning for Unsupervised Hyperspectral Image Clustering|[TGRS'24]|Paper|Preprint

Unsupervised Diffusion and Volume Maximization-Based Clustering of Hyperspectral Images|[RS'23]|Paper

Unsupervised Spatial-Spectral Hyperspectral Image Reconstruction and Clustering With Diffusion Geometry|[WHISPERS'22]|Preprint

Unsupervised Detection of Ash Dieback Disease Using Diffusion-Based Hyperspectral Image Clustering|[IGARSS'22]|Paper|Preprint

Classification of Hyperspectral Images Using SVM With Shape-Adaptive Reconstruction and Smoothed Total Variation|[IGARSS'22]|Paper

Active Diffusion and VCA-Assisted Image Segmentation of Hyperspectral Images|[IGARSS'22]|Paper|Preprint

Unsupervised Classification of AVIRIS-NG Hyperspectral Images|[WHISPERS'21]|Paper

Medical Imaging and Biomedical Signal Analysis

I work on annotation-efficient and clinically meaningful methods for ultrasound, echocardiography, ECG, and cell imaging, with an emphasis on reconstruction, motion understanding, and practical downstream analysis.

Latent Motion Profiling for Annotation-free Cardiac Phase Detection in Adult and Fetal Echocardiography Videos|[MICCAI'25, Oral]|Paper

Blind Restoration of High-Resolution Ultrasound Video|[MICCAI'25]|Paper

A Deep Learning Framework for Fetal Heart Tracking in Ultrasound Videos: Toward Enhanced Congenital Heart Defects Detection|[FIMH'25]|Paper

Bilateral Signal Warping for Left Ventricular Hypertrophy Diagnosis|[ISBI'25, Oral]|Paper

A Method for Cardiovascular Disorder Identification via Prototype-Driven Biomedical Signal Evaluation|[US Patent App., filed '25]

Accurate Detection and Instance Segmentation of Unstained Living Adherent Cells in Differential Interference Contrast Images|[CIBM'24]|Paper

Method for Accurate Segmentation of Unstained Living Adherent Cells in Differential Interference Contrast Images|[US Patent, filed '23; published '25]|Patent

Optimized Hard Exudate Detection with Supervised Contrastive Learning|[ISBI'24]|Paper

LLM Compression, Efficiency, and Interpretability

I study efficient and interpretable LLMs, focusing on low-rank/SVD compression, fair evaluation, recovery in compressed subspaces, and decode-time attribution.

FlashSVD v1.5: Making Low-Rank Transformers Inference Actually Fast|[arXiv'26]|Preprint

DecodeShare: Tracing the Shared Subspace of LLM Decode-Time Decisions|[ICML'26, Spotlight]