About

Shaowen Wang is a professor associated with the Geography department at the University of Illinois Urbana-Champaign. His research areas include Data and Information Systems, CyberGIS, Geospatial Data Science, and Geographic Information Systems (GIS). He has taught courses such as CyberGIS & Geospatial Data Science, CyberGIS, and GIS Professional Seminar. His work involves the application of geospatial technologies and data systems to address complex scientific and societal challenges, contributing to the advancement of geospatial data science and cyberinfrastructure.

Research topics

• Geography
• Computer Science
• Machine Learning
• Artificial Intelligence
• Cartography
• Data Mining
• Remote sensing
• Environmental science
• Business
• Meteorology
• Environmental health
• Economics
• Medicine
• Economic growth
• Agroforestry

Selected publications

• Target noise: A pre-training based neural network initialization for efficient high resolution learning

  ArXiv.org · 2026-02-06

  articleOpen access1st authorCorresponding

  Weight initialization plays a crucial role in the optimization behavior and convergence efficiency of neural networks. Most existing initialization methods, such as Xavier and Kaiming initializations, rely on random sampling and do not exploit information from the optimization process itself. We propose a simple, yet effective, initialization strategy based on self-supervised pre-training using random noise as the target. Instead of directly training the network from random weights, we first pre-train it to fit random noise, which leads to a structured and non-random parameter configuration. We show that this noise-driven pre-training significantly improves convergence speed in subsequent tasks, without requiring additional data or changes to the network architecture. The proposed method is particularly effective for implicit neural representations (INRs) and Deep Image Prior (DIP)-style networks, which are known to exhibit a strong low-frequency bias during optimization. After noise-based pre-training, the network is able to capture high-frequency components much earlier in training, leading to faster and more stable convergence. Although random noise contains no semantic information, it serves as an effective self-supervised signal (considering its white spectrum nature) for shaping the initialization of neural networks. Overall, this work demonstrates that noise-based pre-training offers a lightweight and general alternative to traditional random initialization, enabling more efficient optimization of deep neural networks.

  PublisherOA PDF

• Real-time sick fish detection system for aquaculture based on edge computing

  Aquaculture · 2026-04-08

  article
  PublisherDOI

• Defect 3D reconstruction with integrated bright-field and dark-field structured illumination microscopy based on Att-PU-Net

  Applied Optics · 2026-02-18 · 3 citations

  article

  Addressing the challenges in quantitative 3D inspection of micro-to-nanoscale surface defects in optical components for high-energy laser systems, this paper proposes a novel, to our knowledge, inspection framework integrating bright-field and dark-field structured illumination microscopy (BDSIM) with deep learning-based 3D reconstruction. To mitigate the limitations of sparse point clouds and inherent noise caused by the low luminous flux in BDSIM imaging, we developed the Att-PU-Net model, building upon the point cloud upsampling network (PU-Net) architecture. This model incorporates a self-attention mechanism to enhance the contextual perception of local geometric abruptness and employs a multi-scale feature fusion strategy to preserve fine topological details. To ensure robust generalization from simulation to reality, a hybrid training strategy combining procedurally generated and real-world defect samples is adopted. Furthermore, a composite loss function integrating chamfer distance, repulsion loss, and curvature consistency constraints was designed to significantly improve point distribution uniformity and edge sharpness. Simulations and comparisons between Att-PU-Net and the marching cubes, contour filter algorithms demonstrate that Att-PU-Net achieves an optimal balance between geometric accuracy and uniformity (P2S: 0.5720 µm, NUC: 0.3230). Experimental validation on actual optical damage reveals a reconstruction accuracy of 0.6343 µm and a maximum error of only 0.79 µm in defect depth compared with white light interferometry (WLI), confirming the method's effectiveness and reliability for high-precision 3D reconstruction of complex optical surface defects.

  PublisherDOI

• Target noise: A pre-training based neural network initialization for efficient high resolution learning

  Open MIND · 2026-02-06

  preprint1st authorCorresponding

  Weight initialization plays a crucial role in the optimization behavior and convergence efficiency of neural networks. Most existing initialization methods, such as Xavier and Kaiming initializations, rely on random sampling and do not exploit information from the optimization process itself. We propose a simple, yet effective, initialization strategy based on self-supervised pre-training using random noise as the target. Instead of directly training the network from random weights, we first pre-train it to fit random noise, which leads to a structured and non-random parameter configuration. We show that this noise-driven pre-training significantly improves convergence speed in subsequent tasks, without requiring additional data or changes to the network architecture. The proposed method is particularly effective for implicit neural representations (INRs) and Deep Image Prior (DIP)-style networks, which are known to exhibit a strong low-frequency bias during optimization. After noise-based pre-training, the network is able to capture high-frequency components much earlier in training, leading to faster and more stable convergence. Although random noise contains no semantic information, it serves as an effective self-supervised signal (considering its white spectrum nature) for shaping the initialization of neural networks. Overall, this work demonstrates that noise-based pre-training offers a lightweight and general alternative to traditional random initialization, enabling more efficient optimization of deep neural networks.

  DOI

• Ending the HIV Epidemic: Development and Evaluation of CyberGIS-HIV, a Web-Based Prediction Application

  American Journal of Public Health · 2025-08-14

  articleOpen access

  Officials from US state and county health departments perceive the need to adopt new methods for making epidemiological decisions about HIV. Hence, a web-based prediction modeling application, CyberGIS-HIV, was developed and systematically compared with currently used approaches (e.g., community consultation) in a sample of 42 state and federal public health officials as well as senior and junior modelers. Overall, CyberGIS-HIV had more favorable public health and modeling ratings than other approaches currently used by the participants. ( Am J Public Health. 2025;115(10):1589–1593. https://doi.org/10.2105/AJPH.2025.308194 )

  PublisherOA PDFDOI

• E. coli-mediated expedient biosynthesis of biotin-labeled ubiquitin probe enables activity-based profiling of deubiquitinases in cytoplasmic and nuclear compartments

  Chinese Chemical Letters · 2025-08-28

  article
  PublisherDOI

• Self-Improvement for Audio Large Language Model using Unlabeled Speech

  2025-08-17 · 6 citations

  article1st authorCorresponding
  PublisherDOI

• Enabling collaboration through data and model sharing with CUAHSI HydroShare; CIROH Devcon 25 presentation

  HydroShare Resources · 2025-07-20

  datasetOpen access
  PublisherDOI

• Integrating comprehensive geriatric assessment into routine nursing care for older adults with type 2 diabetes: implementation fidelity and clinical outcomes

  Frontiers in Public Health · 2025-09-19

  articleOpen accessSenior authorCorresponding

  Background Comprehensive geriatric assessment (CGA) offers promise for improving diabetes management in older adults; however, its real-world effectiveness depends on implementation fidelity, which remains poorly understood. This study examined fidelity variations and their associations with clinical outcomes in nurse-led CGA for older adults with type 2 diabetes at a tertiary care hospital in China. Methods This cross-sectional implementation study enrolled 3,351 adults aged ≥65 years with type 2 diabetes from Shanghai Jiading District Central Hospital between March 2021 and February 2025. Implementation fidelity was assessed using five validated dimensions yielding a composite score (mean 0.64, SD 0.19; range 0.28–0.94). Primary outcome was glycated hemoglobin (HbA1c); secondary outcomes included cardiometabolic parameters, patient-centered measures, healthcare utilization, and hypoglycemic events. Linear regression models with robust standard errors adjusted for confounders; mediation analysis examined functional status pathways. Results Fidelity demonstrated variation within the hospital (mean 0.64, SD 0.19; range 0.28–0.94), with higher educational attainment, provider experience, and CGA training completion associated with better implementation quality. Higher fidelity was associated with lower HbA1c (adjusted β −0.38 per 0.10-unit increase, 95% CI −0.47 to −0.29; p < 0.001), with a graded association across quartiles [7.89% (95% CI 7.78–8.00) in the lowest quartile vs. 7.16% (95% CI 7.04–7.28) in the highest quartile; p for trend < 0.001]. Benefits were associated with lower systolic blood pressure (−5.10 mm Hg, 95% CI −7.20 to −3.00), LDL cholesterol (−6.50 mg/dl, 95% CI −9.10 to −3.90), improved quality of life (EuroQol-5D: 0.061, 95% CI 0.041–0.081), and decreased depressive symptoms (−1.10, 95% CI −1.40 to −0.80; all p < 0.001). Healthcare utilization declined (hospitalization incidence rate ratio 0.61, 95% CI 0.51–0.73; p < 0.001), and odds of hypoglycemic events were lower (odds ratio 0.78, 95% CI 0.72–0.84; p < 0.001). Functional status was an estimated mediator of 31.6% of the fidelity–HbA1c association (indirect β −0.12, 95% CI −0.17 to −0.07; p < 0.001), with age and gait speed modifying associations ( p = 0.04 and 0.02, respectively). Conclusion High-fidelity CGA integration is associated with substantial clinical benefits and lower healthcare utilization; while suggestive of economic advantages, a formal cost-effectiveness evaluation was not undertaken. These associations support an institutional focus on provider training, experience development, and patient education to optimize geriatric diabetes care quality.

  PublisherOA PDFDOI

• Directional consistency of dark-field illumination for defect images

  Applied Optics · 2025-12-02 · 1 citations

  article1st authorCorresponding

  Scratch defects in dark-field imaging exhibit directional characteristics, with variations in incident light angles causing significant differences in images. This poses a challenge for the detection of randomly oriented scratches, as the illumination method directly impacts the missed detection and detection repeatability. The study establishes a dark-field imaging model, which is based on the finite-difference time-domain method. By varying illumination directions and superimposing uni-directional incoherent illumination imaging data, the dark-field images of scratches under both line-scan and annular light sources are obtained. The relationships between imaging intensities and imaging widths of scratches with incident angles are analyzed, which can quantify the directional consistency. Simulation results reveal that increasing the number of illumination beams effectively reduces fluctuations in both imaging intensity and imaging width, thereby improving the dark-field imaging consistency. Notably, annular light sources with odd-numbered beams offer an advantage in imaging consistency over the even-numbered configurations. The trend of the experimental results using dark-field systems is in accord with the simulations. A comparative analysis of defect detection between line-scan and annular light sources confirms that the annular source has superior performance in defect detectability.

  PublisherDOI

Recent grants

• CIF21 DIBBs: Scalable Capabilities for Spatial Data Synthesis

  NSF · $1.5M · 2014–2020

• Collaborative Research: SI2-SSI: Cyberinfrastructure for Advancing Hydrologic Knowledge through Collaborative Integration of Data Science, Modeling and Analysis

  NSF · $700k · 2017–2022

• EAGER: CISSDA: A Unified Cyberinfrastructure Framework for Scalable Spatiotemporal Data Analytics

  NSF · $300k · 2013–2016

• HDR Institute: Geospatial Understanding through an Integrative Discovery Environment

  NSF · $16.0M · 2021–2026

• MRI: Acquisition of a National CyberGIS Facility for Computing and Data-Intensive Geospatial Research and Education

  NSF · $1.8M · 2014–2018

Frequent coauthors

• Anand Padmanabhan

  105 shared

• Yan Liu

  75 shared

• Aiman Soliman

  44 shared

• Dandong Yin

  University of Illinois Urbana-Champaign

  40 shared

• Kiumars Soltani

  36 shared

• Junjun Yin

  Social Science Research Council

  32 shared

• Yizhao Gao

  University of Hong Kong

  30 shared

• Hao Hu

  25 shared

Labs

• Siebel School of Computing and Data SciencePI

Education

• Ph.D., Computer Science

  University of Illinois at Urbana-Champaign

  2006

• M.S., Computer Science

  University of Illinois at Urbana-Champaign

  2002

• B.S., Computer Science

  University of Science and Technology of China

  1999