Yang Li
PhD candidate
Electrical and Computer Engineering
University of Rochester
I′m currently a PhD student at the department of Electrical and Computer Engineering at University of Rochester. I′m now doing research in graph signal processing and network science with my academic supervisor Prof. Gonzalo Mateos. I received the degree of Bachelor in Engineering at Tianjin University in 2013. After that, I worked at Tianjin University as a research associate with the concentration in data analysis with magnetic sensor signals. I started as a graduate student in August 2014 at University of Rochester and received the degree of Master in Science in Electrical Engineering in May, 2016.
My current research is brain graph analytics, including mapping from fMRI imaging to brain graphs and applying machine learning algorithms for classification and inference. I′m also researching on community detection in networks with the possible application in modularity analysis in brain network. My research interests include
Research.
Brain graph construction with fMRI data
Overview:
fMRI focuses on brain functional connectivity. With fMRI signals, certain algorithms can be used to backtrack and locate those brain regions activated. The obtained regions can therefore consist the graph nodes. The dependencies between regions can be further used to construct the graph edges. Graphs derived from fMRI data show specifically how the brain operates to response such stimuli.
- fMRI data → brain graph
- Brain region → graph nodes
- Connectivity → graph edges
- Further analysis on graph properties
Year: 2017
More
Community detection on graphs
Overview:
Real world problems can be transferred into graph format for analysis, and one of the properties of such graphs is the community structure. It is of great interest and importance to detect such communities so that the overall network can be better analyzed and hidden properties can be explored.
- Community structures of network
- Graph construction from network
- Modularity optimization methods
- Multi-resolution community detection
Year: 2017
More
Data analysis with eddy current and GMR sensor
Overview:
GMR sensor is able to pick up the magnetic field signal generated by the eddy current on metal samples (intact or flawed). Metal defects will disturb the detected distribution of eddy current. Magnitude, bandwidth and phase of the signals are used as the features. Machine learning techniques such as SVM, decision tree, KNN, neural network have been applied for the classification.
- Metal quality detection
- Machine learning for classification
- Signal magnitude, phase, bandwidth
- Non-destructive testing
Year: 2013-2014
More
Brain graph construction with fMRI data
Community detection on graphs
Data analysis with eddy current and GMR sensor
Publications.
1. Gao P, Wang C, Li Y, et al. GMR-based eddy current probe for weld seam inspection and its non-scanning detection study[J]. Nondestructive Testing and Evaluation, 2017, 32(2): 133-151.
2. Gao P, Wang C, Li Y, et al. Electromagnetic and eddy current NDT in weld inspection: a review[J]. Insight-Non-Destructive Testing and Condition Monitoring, 2015, 57(6): 337-345.
3. Gao P, Wang C, Zhi Y, Li Y. Defect classification using phase lag information of EC-GMR output[J]. Nondestructive Testing and Evaluation, 2014, 29(3): 229-242.
4. Gao P, Wang C, Li Y, et al. Defect evaluation using the phase information of an EC-GMR sensor[C]. I2MTC Proceedings, 2014 IEEE International. IEEE, 2014: 25-29.
5. Gao P, Wang C, Zhi Y, Li Y, et al. Extraction of nonlinear characteristics from eddy current magnetic field of Al-alloy weld and their classification[J]. Acta Phys. Sin. , 2014, 63(13).
Patents.
1. Detection method and device for welding defect giant magneto-resistance eddy current based on decision-making tree, CN103760231A Issued Nov 23, 2016
2. Welding defect giant magneto-resistance eddy current testing method based on Bayesian network, CN103713043B, Issued May 4, 2016
Projects.
Community Detection in Graphs
Main content:
- Community structures of network
- Graph construction from network
- Modularity optimization methods
- Multi-resolution community detection
- Matlab programming
- Final presentation
Overview:
Real world systems can be modeled as network graphs and one vital property is the underlying community structure. In this project, multiple community detection methods, including the traditional ones based on modularity optimization, and the multi-resolution methods which are able to detect the hierarchical graph organization and overlapping communities. The comparison between the algorithms is given and the performance is visualized.
Year: 2017
Graph data recovery and classification
Main content:
- Graph data reconstruction
- Classification techniques
- Total variation minimization
- Total variation regularization
- Matlab programming
Overview:
Web blog data contains much information about social opinions and can be used to predict future trend in multiple areas. In this project, the web blog dataset regarding the 2004 US Presidential Election is used. By applying total variation minimization methods, the missing data from original dataset can be recovered with high accuracy. Different parameters are compared with plots to present the optimization process. The methods can also be used to predict social opinions, identify potential users for certain product.
Year: 2016
NBA statistics database system
Main content:
- Data collection regarding NBA
- Database entities, relations, attributes
- Functional dependency, relation norm
- ER diagram
- Programming in MySQL, HTML, PHP
- Database web interface construction
Overview:
In this project, a complete database regarding the statistics of NBA (players, games, seasons etc.) is established. The process includes data collection, raw data processing, database entity establishment and relation connection. Norms are used to provide consistency between data and constraints/triggers are applied to deal with anomalies. The database is built up based on MySQL and web interface programmed in HTML and PHP is also included for the access to the database via web browser.
Year: 2015
Defect classification based on GMR sensor signals
Main content:
- GMR sensor, metal board samples
- Magnitude/phase of sensor signals
- Machine learning (SVM, K-means etc.)
- Determin metal flaws/sample quality
- Non-destructive testing>/li>
Overview:
Embedded GMR sensors into optical mouse circuits to detect current distribution around cracks on metal boards. Analyzed the returned sensor signals in Matlab to determine the condition of the metal product. Extracted the magnitude and phase information of the detection signals to evaluate the location and the size of the defects. Applied machine learning techniques (SVM, K-means, Neural Networks and decision trees) for data pattern recognition and experiment result classification to identify the qualify of metal board samples (intact or flawed).