Selected work

Representative Papers

Neural Algorithmic Reasoning for Graph Saddle Point Problems

Oct 05, 2026

This study addresses the bottlenecks in solving saddle-point problems on graphs and improving combinatorial optimization efficiency by proposing the GraphPDHG framework. This method pioneers the integration of the primal-dual hybrid gradient (PDHG) algorithm into graph message-passing mechanisms, achieving deep alignment between neural network architectures and classical optimization paradigms through simulating the Chambolle-Pock algorithm. The results demonstrate that this network effectively learns and accelerates the underlying algorithm, significantly enhancing cross-size generalization while serving as an excellent warm-start initialization for second-order optimization. Furthermore, when combined with the neural algorithmic reasoning paradigm, its generalization performance substantially surpasses that of unaligned GNN baselines.

0 citationsRead paper

A comparison between geostatistical and machine learning models for spatio-temporal prediction of PM2.5 data

Sep 15, 2025

Traditional air quality monitoring networks suffer from insufficient spatiotemporal resolution, while low-cost PurpleAir sensor data exhibit systematic biases. Method: This study proposes a high-accuracy PM₂.₅ prediction framework integrating spatiotemporal dependency modeling and multi-algorithm ensemble learning. Conducting hourly modeling across California, we systematically benchmark kriging interpolation, land-use regression, neural networks, random forests, and support vector machines. We further introduce a novel spatiotemporally aware ensemble model that jointly enhances sensor data reliability and spatial heterogeneity representation via spatiotemporal graph convolution and bias correction. Contribution/Results: The proposed model significantly outperforms all individual baselines (RMSE reduced by 18.7%; R² increased by 0.12) and generates real-time, 1 km × 1 km resolution PM₂.₅ concentration maps—providing a scalable, cost-effective methodology for high-fidelity urban air quality monitoring.

0 citationsRead paper

Digital Contact Tracing: Examining the Effects of Understanding and Release Organization on Public Trust

Aug 13, 2025

This study investigates whether public trust in COVID-19 digital contact tracing applications is influenced by individuals’ understanding of their privacy-preserving mechanisms and the institutional identity of the app publisher (public vs. private sector). Analyzing online survey data from 101 U.S. adults, we employed correlation and regression analyses to test hypothesized relationships. Results indicate that: (1) comprehension of privacy mechanisms exhibits no statistically significant association with trust levels; and (2) the publisher’s institutional affiliation—whether governmental (e.g., U.S. federal agencies) or private-sector consortia (e.g., the Google–Apple Exposure Notification framework)—does not significantly predict trust. These findings challenge two prevailing assumptions in digital health governance: that privacy literacy drives trust, and that institutional provenance inherently determines credibility. Instead, trust appears contingent on unmeasured factors—such as perceived risk, social norms, or firsthand usability experiences. The study thus provides novel empirical evidence and prompts theoretical reconsideration of trust formation in digital public health interventions.

0 citationsRead paper

Recent publications

Latest Papers

Neural Algorithmic Reasoning for Graph Saddle Point Problems

Oct 05, 2026

This study addresses the bottlenecks in solving saddle-point problems on graphs and improving combinatorial optimization efficiency by proposing the GraphPDHG framework. This method pioneers the integration of the primal-dual hybrid gradient (PDHG) algorithm into graph message-passing mechanisms, achieving deep alignment between neural network architectures and classical optimization paradigms through simulating the Chambolle-Pock algorithm. The results demonstrate that this network effectively learns and accelerates the underlying algorithm, significantly enhancing cross-size generalization while serving as an excellent warm-start initialization for second-order optimization. Furthermore, when combined with the neural algorithmic reasoning paradigm, its generalization performance substantially surpasses that of unaligned GNN baselines.

0 citationsRead paper

A comparison between geostatistical and machine learning models for spatio-temporal prediction of PM2.5 data

Sep 15, 2025

Traditional air quality monitoring networks suffer from insufficient spatiotemporal resolution, while low-cost PurpleAir sensor data exhibit systematic biases. Method: This study proposes a high-accuracy PM₂.₅ prediction framework integrating spatiotemporal dependency modeling and multi-algorithm ensemble learning. Conducting hourly modeling across California, we systematically benchmark kriging interpolation, land-use regression, neural networks, random forests, and support vector machines. We further introduce a novel spatiotemporally aware ensemble model that jointly enhances sensor data reliability and spatial heterogeneity representation via spatiotemporal graph convolution and bias correction. Contribution/Results: The proposed model significantly outperforms all individual baselines (RMSE reduced by 18.7%; R² increased by 0.12) and generates real-time, 1 km × 1 km resolution PM₂.₅ concentration maps—providing a scalable, cost-effective methodology for high-fidelity urban air quality monitoring.

0 citationsRead paper

Digital Contact Tracing: Examining the Effects of Understanding and Release Organization on Public Trust

Aug 13, 2025

This study investigates whether public trust in COVID-19 digital contact tracing applications is influenced by individuals’ understanding of their privacy-preserving mechanisms and the institutional identity of the app publisher (public vs. private sector). Analyzing online survey data from 101 U.S. adults, we employed correlation and regression analyses to test hypothesized relationships. Results indicate that: (1) comprehension of privacy mechanisms exhibits no statistically significant association with trust levels; and (2) the publisher’s institutional affiliation—whether governmental (e.g., U.S. federal agencies) or private-sector consortia (e.g., the Google–Apple Exposure Notification framework)—does not significantly predict trust. These findings challenge two prevailing assumptions in digital health governance: that privacy literacy drives trust, and that institutional provenance inherently determines credibility. Instead, trust appears contingent on unmeasured factors—such as perceived risk, social norms, or firsthand usability experiences. The study thus provides novel empirical evidence and prompts theoretical reconsideration of trust formation in digital public health interventions.

0 citationsRead paper