Projects
Streaming Classical Multidimensional Scaling
Proposed a one-pass, limited memory streaming classical Multidimensional Scaling (MDS) algorithm with higher approximation accuracy. It can be used as a dimension reduction tool to sequentially represent high dimensional data such as text and images.
Streaming Representation for Online Corpus
Proposed a representation framework for streaming text data. It can be utilized to semantically reconstruct text data and summarize the important information for data understanding over time. Productized as an internal tool for data representation in Brookhaven National Lab KBase project.
Time Series Anomaly Detection for Unusual Status in Data Center
Generated a pattern discovery method using Long Short Term Memory (LSTM) for anomaly detection among multivariate time series data. The method was adopted to detect unusual status (e.g. overheating) for RACF Data Center in Brookhaven National Lab.
Sequence Modeling for Weather Prediction
Implemented Recursive Neural Network (RNN) using TensorFlow backend for weather forecasting, which applied among multivariate time series data collected by sensors, such as temperature, air pressure, humanity etc.
Image Detection for Human Upper Bodies
Implemented a Support Vector Machine (SVM) classifier to detect images with human upper bodies. The classifier is used to distinguish upper-body patches from non-upper-body patches and used Histogram of Oriented Gradient (HOG) algorithm for feature selection.
Predictive Model for Biological Data Analysis
Developed a LASSO regression solver to predict and improve experimental design. It was integrated into a bioinformatics system that enables the scientist to upload their own data and analyze it in School of Medicine of Stony Brook University.