Skip to content. Skip to main navigation.

Faculty Highlight- Assistant Professor Junzhou Huang

Junzhou Huang February 5, 2015
Personal website: http://ranger.uta.edu/~huang/

Junzhou Huang, PhD is an Assistant Professor at the CSE Department. His major research interests include machine learning, computer vision and biomedical imaging informatics. The general goal of his research is to investigate scalable models and algorithms for data-intensive applications. To be more specific, he is working on statistical learning and imaging informatics with the focus on scalable algorithms and software systems. His interest is to develop efficient algorithms with nice theoretical guarantees to solve practical problems involved huge scale data.

Because of the enormous and unexplored potential that big data analysis has in many real-world applications, since joining UTA Dr. Huang has established important multidisciplinary collaborations with experts in math, statistics, surgeons, radiologists and biologist inside and outside of UTA. Despite his ultimate focus on real-world applications, most of his recent work deals with researching algorithms with a solid theoretical foundation, and that address unsolved research challenges in this age of big data. His publication record is broad and covers diverse areas such as statistics, machine learning, computer vision, medical imaging informatics, pattern recognition and image processing. Dr. Huang has published 90+ papers on top tier Journals and conferences in his research areas. According to Google scholar, his citations are around 1918, h-index is 20, and i-10 index is 30. He has been globally selected as one of the 10 emerging leaders in multimedia and signal processing by the IBM T.J. Watson Research Center in 2010. He won the MICCAI Young Scientist Award 2010, the FIMH Best Paper Award 2011, the STMI Best Paper Award 2012 and the NIPS Best Reviewer Award 2013.

Heart Paraview
Heart Paraview

Dr. Huang is currently leading the Scalable Modeling & Imaging & Learning Lab (SMILE) at University of Texas at Arlington. His lab is working actively on developing scalable algorithms and applying high performance computing technologies, such as multicore processors, GPU. and Grid/Cloud computing environment to process and analysis big data together with many other collaborators. The research projects in his lab are broad and interdisciplinary, such as Robust Materials Genome Data Mining for Nanoparticle Synthesis, Adaptive Interdisciplinary Pain Management for Personalized Medicine, Large Scale Non-intrusive Energy Monitoring, Compressed Sensing Magnetic Resonance Imaging (MRI), 3D Modeling and Simulation, Visual Tracking and Event Detection and so on.

The Materials Genome Initiative research has been launched by U.S. government to discover, manufacture, and deploy advanced materials fast and low-cost, which holds great opportunities to address the challenges in clean energy, national security, and human welfare. However, the major computational challenges are the bottlenecks for comprehensive materials genome data analysis. There is a critical need for new data mining and machine learning strategies to bridge the gap and facilitate the new materials discovery. The overall goal of this project is to address the computational challenges for comprehensive materials genome data analysis due to unprecedented scale and complexity, and develop new data mining and machine learning strategies for facilitating the new materials discovery. This project will develop new computational tools to automate the material genome data processing, investigate the new learning model to integrate heterogeneous material characterizations for predicting the catalytic capabilities and associations to theoretical modeling measurements, design novel robust learning techniques to predict the catalytic capabilities of the new synthesized nanoparticles. This project is funded by NSF IIS core program from 08/2014 to 08/2017 with a total amount of an award as $250,000.

Adaptive Interdisciplinary Pain Management is an important and challenging task for clinical treatment with Personalized Medicine. The overall goal of this project is to develop new statistics learning and optimization methods for adaptive interdisciplinary pain management that use patient data to recommend an interdisciplinary treatment regime with personalized medicine for controlling pain outcomes. This project will investigate machine learning method to impute missing pain records, develop an inverse-probability-of-treatment weighted estimator method for complex pain data, study how to coordinate the development of statistical meta-models and optimization algorithms and evaluated how the interdisciplinary pain management system could be accurately represented or globally optimized for personalized treatment and medicine. This project is funded by NSF CMMI core program from 09/2014 to 08/2017 with a total amount of an award as $374,998.

Large scale Non-Intrusive Load Monitoring (NILM) is an attractive method for energy disaggregation, as it can discern devices from the aggregated data acquired from a single point of measurement. Although numerous studies have been devoted to developing effective models for NILM from high-sample-rate data with higher costs, limited progress has been made in low frequency energy disaggregation by exploring discriminant features of low-sample-rate data for different appliances. There is a lack of research work considering inherent characteristics of the data. Due to these special characteristics, how to maximally exploit them to improve energy disaggregation is the key challenge in NILM. This project aims at discovering discriminant features from the low-sample-rate power measurements to characterize appliance functional modes, developing new hierarchical probabilistic models to imbed discriminative features, working states and aggregated consumption into a unified framework and investigating efficient inference algorithms to learn the latent states from the measured aggregation data. Our result was published in NILM’14. This project is funded by Samsung Research America from 08/2013 to 07/2016 with a total amount of an award as $149,998.

MRI Flow
MRI Flow

Magnetic Resonance Imaging (MRI) has been widely used in medical diagnosis because of its non-invasive manner and excellent depiction of soft tissue changes. Compressive sensing can help to accurately reconstruct Magnetic Resonance (MR) images from highly under-sampled K-space data and therefore significantly reduce the scanning duration. However, current solvers for it are still very slow and impractical for real MR images, which prevented it from being used in practice. Moreover, current algorithms depend on the standard sparsity assumption, which is not good enough. Building upon the foundation of his structured sparsity (SS) theory, Dr. Huang has developed a new MRI system to dramatically improve the current MRI techniques with higher resolution and short imaging duration. The initial results have been published on top tier Journals and conferences in medical imaging. This work helps Dr. Huang win the Young Scientist Award in the 13th International Conference on Medical Image Computing and Computer Assisted Intervention, Beijing, China, 2010 and the Best Paper Award in the MICCAI Workshop on Sparsity Techniques in Medical Imaging, Nice, France, 2012. This project has been funded by Research Enhancement Program (REP) at University of Texas.

A long-term research question in image/video processing is how parsimoniously the images/signals can be represented. Dr. Huang is developing sparsity techniques to answer it by exploiting more structure information to learn sparse representations in high-dimensional images/signals. A real-time tracking systems and dynamic event detection system are developed in this lab. In visual tracking, object occlusions can be detected as anomaly, which are clustered sparse. Inspired by this, a robust tracking system is developed to handle sparse occlusion and illumination changes in long-term tracking. Dr. Huang has also developed a framework to track face shapes by using both color and depth information. The low-resolution depth image is captured by using Microsoft Kinect. By exploiting the depth information, the performance of the tracking system is significantly improved. A non-intrusive system has been developed for monitoring fatigue by tracking eyelids with a single web camera based on the accurate face tracking system. The system is able to track face movement and fit eyelids reliably in real time.

Dr. Huang is also working actively on sparse learning and compressed imaging. These findings are reported in leading conferences and journals, including International Conference on Machine Learning, International Conference on Computer Vision and Pattern Recognition, Annals of Statistics, Journal of Machine Learning Research, IEEE Transaction on Pattern Analysis and Machine Intelligence, Medical Imaging, and, etc. His recent research is heavily focused on scalable imaging and learning with parallel and distributed computation techniques.


Video Introducing Face Tracking Demo