I am Postdoc at UT Arlington in Prof. Prof. Junzhou Huang's lab. I am passionate at creating new machine learning algorithms to solve challenges met in the development of bioinformatics and biomedicine as well as developing real Artificial Agents to play challeging games, such as Atari, Alphago, Poker etc. Before the Postdoc at UTA, I spent one year in the Department of Statistics at the University of Michigan, starting September 2015. Prior to the working experience, I received my Ph.D. in the National Laboratory of Pattern Recognition (NLPR) at the Institute of Automation, Chinese Academy of Sciences (CASIA), advised by Prof. Chunhong Pan , and Prof. Shiming Xiang and Dr. Ying Wang. You can find my CV here.
Research Interests: Deep Reinforcement Learning, Unsupervised Learning, Manifold Learning, Remote Sensing Image Analysis, Hyperspectal Image Unmixing
- 04/2017 One paper on Actor-Critic Reinforcement Learning for Mobile Health Intervention is provided on ArXiv
- 03/2017 One paper accepted by CVPR 2017
Peer-reviewed Journal Papers
Guangliang Cheng, Feiyun Zhu, Shiming Xiang and Chunhong Pan. "Road Centerline Extraction via Semisupervised Segmentation and Multidirection Nonmaximum Suppression" , IEEE Transactions on Geoscience and Remote Sensing Letters, vol. 37, no. 1, pp. 191-211 (IEEE TGRSL), 2016.
Guangliang Cheng, Feiyun Zhu, Shiming Xiang, Ying Wang and Chunhong Pan. "Accurate urban road centerline extraction from VHR imagery via multiscale segmentation and tensor voting" , Elsevier Neurocomputing, vol. 205, pp. 407-420, 2016.
Haichang Li, Ying Wang, Shiming Xiang, Jiangyong Duan, Feiyun Zhu, Chunhong Pan. "A label propagation method using spatial-spectral consistency for hyperspectral image classification" , International Journal of Remote Sensing, vol. 37, no. 1, pp. 191-211, 2016.
Ying Wang, Chunhong Pan, Shiming Xiang and Feiyun Zhu. "Robust Hyperspectral Unmixing with Correntropy based Metric" , IEEE Transactions on Image Processing (IEEE TIP), vol. 24, no. 11, pp. 4027-4040, 2015.
Guangliang Cheng, Feiyun Zhu, Shiming Xiang, Ying Wang, Chunhong Pan. "Semisupervised Hyperspectral Image Classification via Discriminant Analysis and Robust Regression" , IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 2, pp. 595-608, 2015.
Feiyun Zhu, Ying Wang, Bin Fan, Shiming Xiang, Gaofeng Meng and Chunhong Pan. "Spectral Unmixing via Data-guided Sparsity" , IEEE Transactions on Image Processing (IEEE TIP), vol. 23, no. 12, pp. 5412-5427, 2014.
Feiyun Zhu, Ying Wang, Shiming Xiang, Bin Fan and Chunhong Pan. "Structured Sparse NMF for Hyperspectral Unmixing" , ISPRS Journal of Photogrammetry and Remote Sensing, vol. 88, no. 1, pp. 101-118, 2014.
Peer-reviewed Conference Papers
Xinliang Zhu, Jiawen Yao, Feiyun Zhu and Junzhou Huang. "WSISA: Making Survival Prediction from Whole Slide Pathology Images" , International Conference on Computer Vision and Pattern Recognition (CVPR), 2017.
Xiaoping Hu, Ying Wang, Feiyun Zhu and Chunhong Pan. "Learning-based fully 3D face reconstruction from a single image" , IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 1651-1655, 2016.
Feiyun Zhu, Bin Fan, Xinliang Zhu, Ying Wang, Shiming Xiang and Chunhong Pan. "10,000+ Times Accelerated Robust Subset Selection (ARSS)" , Proc. Association for the Advancement of Artificial Intelligence (AAAI), 2015. (AR: 531/1991~26.67%).
Guangliang Cheng, Ying Wang, Feiyun Zhu and Chunhong Pan. "Road extraction via adaptive graph cuts with multiple features" , IEEE Conference on Image Processing, pp. 3962-3966, 2015.
Guangliang Cheng, Ying Wang, Yongchao Gong, Feiyun Zhu and Chunhong Pan. "Urban road extraction via graph cuts based probability propagation" , IEEE Conference on Image Processing, pp. 5072-5076, 2014.
Feiyun Zhu, Peng Liao, Xinliang Zhu, Jiawen Yao and Junzhou Huang. "Cohesion-based Online Actor-Critic Reinforcement Learning for mHealth Intervention" , arXiv:1703.10039, 2017 (under review).
Feiyun Zhu, Ying Wang, Bin Fan, Gaofeng Meng and Chunhong Pan. "Effective Spectral Unmixing via Robust Representation and Learning-based Sparsity" , arXiv:1409.0685, 2014 (under review).
- One deep attention-based model was proposed to solve the challenging problem. Very promising results were achieved on two types of cancers datasets;
- We are trying to generalize our model to solve other fine-grained learning problems with weak labels.
- Efficient deep learning models with specific training strategies were developed to solve the first two problems;
- A novel framework was also created to solve the third challenge.
- Integrating features extracted from pathology image patches with genetic signature expressions to improve the survival prediction accuracy;
- Developing methods for imaging biomarker discovery;
- Mapping clinical outcome correlated Imaging-Genetic data by developing supervised conditional Gaussian graphical model (SuperCGGM).
- Created a novel method for RGB and depth images registration in outdoor scenes;
- Collected over 10,000 RGBD outdoor pedestrain RGBD images;
- Developed a fast pedestrian detection framework based on RGBD images.
Large Scale Learning for Complex Image-Omics Data Analytics
This project aims to develop computational tools for analyzing complex pathology image data as well genomics data. To solve the key and challenging problems in mining comprehensive heterogeneous image and genomic data, novel large scale learning tools and explore ways to integrate features from multiple data sources for clinical outcome prediction are developed. It will greatly support the Precision Medicine Initiative, which enables physicians to select individualized treatments. I work on developing feature learning from gigapixel whole slide pathology images and integrating imaging-omics data methods.
Miniature Autonomous Vehicle
The miniature autonomous vehicle was built from a toy car. It mainly consisted of a toy car, 3 web cameras, 1 ultrasonic radar and a motherboard equipped with Ubuntu 12.04LTS. It could do traffic signs recognition, traffic light recognition, road line detection and also be with the basic functions of a toy car. I was in charge of building the traffic signs recognition module and testing the hardware & software architecture.
SOAR-based Air Traffic Control Simulation System
Using SOAR as a cognition architecture to simulate an air traffic controller’s decision making could potentially make the training of an air traffic controller easier and with more fun. I was the software architect of the whole system and the main designer of the rules used in SOAR architecture