WU's page

Generative model (GAN) + inverse problem + compressive sensing + crack segmentation:

Reference:
Huang et al. (2020), Recovering compressed images for automatic crack segmentation using generative models.



 

Machine learning (transfer learning) + Bayesian inference + uncertainty quantification for polymer design:

Reference:
Wu et al. (2019), Machine-learning-assisted discovery of polymers with high thermal conductivity using a molecular design algorithm, npj Computational Materials.
Yamada&Liu et al. (2019), Predicting Materials Properties with Little Data Using Shotgun Transfer Learning, ACS Central Science.
Wu et al. (2019), iQSPR in XenonPy: A Bayesian Molecular Design Algorithm, Molecular Informatics.



 

A five-year-long cross-disciplinary and cross-institute project on studying whole brain neuronal activity of Caenorhabditis elegans (C. elegans) has taken part in Japan since 2013. The goal is to understand the relationship between the complicated neural system dynamics and animal behaviors through recent advancements of 3D imaging and data analysis techniques. Pan-neuronal calcium imaging is performed to capture neural activities of a large number of C. elegans samples. On top of the typical measurement noise issue, the heterogeneity of the samples posts significant challenges on image registration and annotation. My research develops a series of statistical techniques that are suitable for automatic annotation and data analysis. As a result, one can exploit the increasing availability of large data sets of high resolution pan-neuronal images for more reliable data analysis as compared to the existing studies. My methodology combines the advantages from the Motion Coherent Theory, Integer Linear Programing and Weighted Majority Voting to achieve accurate and computationally efficient annotation algorithm.

Reference: Wu et al. (2017), An ensemble learning approach to auto-annotation for whole-brain C. elegans imaging, bioRxiv:180430.



 

(Collaboration with Dr. Panagiotis Angelikopoulos, Dr. Georgios Arampatzis, Prof. Costas Papadimitriou and Prof. Petros Koumoutsakos)

Hierarchical model is used across many fields, both in scientific and engineering research. The earliest concept of a hierarchical model appears in statistics, known as random effect model. When putting the problem into a Bayesian framework, there can be different types of hierarchical models. In general, such a complex model significantly increases the computational demand. My research develops an efficient way of handling hierarchical models for Bayesian inference. Previous applications include Molecular Dynamics and Pharmacokinetics models.

References:
Wu et al. (2019), Hierarchical Stochastic Model in Bayesian Inference for Engineering Applications: Theoretical Implications and Efficient Approximation, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering, 5(1):011006.
Wu et al. (2017), Bayesian Annealed Sequential Importance Sampling (BASIS): an unbiased version of Transitional Markov Chain Monte Carlo, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering, 4(1):011008. (2018 best paper award)
Wu et al. (2016), Fusing heterogeneous data for the calibration of molecular dynamics force fields using hierarchical Bayesian models, Journal of Chemical Physics, 145(24):244112.
Wu et al. (2016), A hierarchical Bayesian framework for force field selection in Molecular Dynamics Simulations, Phil. Trans. R. Soc. A, 374(2060):20150032.



 

(Collaboration with Prof. Masumi Yamada)

Applying particle filter to extend the earthquake early warning algorithm for handling multiple event cases.

Reference: Wu et al. (2015), Multi-events Earthquake Early Warning algorithm using a Bayesian approach, Geophyical Journal International, 200, p.791-808.



 

Due to the high uncertainty of the stress and strength distributions within the tectonic plates on Earth, earthquakes are one of the most unpredictable natural hazards. Accurate prediction of when an earthquake will happen is still not possible, but the concept of EEW can be achieved because of the rapid development of computing power and network communication. A major challenge of developing applications for EEW is the short lead time, ranging from a few seconds to minutes depending on the geographics.As a result, a robust automated decision process about whether to initiate a mitigation action is essential. One of the recent approaches proposes taking an action upon exceedance of a fixed threshold for an intensity measure or damage or loss measure, but the determination of the threshold value remains as an open-ended question. Other approaches propose a decision-making framework based on cost-benefit analysis. However, a general framework that can handle multiple-action decisions, lead time and its uncertainty is still missing. My research focuses on developing a more robust decision criterion based on a new cost-benefit analysis procedure, which is proposed as part of an earthquake probability-based automated decision-making (ePAD) framework.

References:
Wu et al. (2013), Earthquake Probability-based Automated Decision-making Framework for Earthquake Early Warning Applications, Computer-Aided Civil and Infrastructure Engineering, 28, p.737-752.
Wu et al. (2015), An engineering application of earthquake early warning: ePAD-based decision framework for elevator control, ASCE-Structural Engineering, 142(1):04015092.



 

(Collaboration with Prof. Konstantin M. Zuev)

Reference: Zuev et al. (2015), General network reliability problem and its efficient solution by Subset Simulation, Probabilistic Engineering Mechanics, 40, p.25-35.

 

(Collaboration with Dr. Yong Huang)

Reference: Huang et al. (2014), Robust Bayesian compressive sensing for signals in structural health monitoring, Computer-Aided Civil and Infrastructure Engineering, 29(3-SI), p.160-179.

 

(Collaboration with Prof. Ikumasa Yoshida)

Reference: Yoshida et al. (2018), Optimal sampling placement in a Gaussian random field based on value of information, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, 4(3):04018018.