Research
Current
My research interest mainly involves Bayesian inference in signal processing and machine learning, and its applications for understanding the complex world. I am working with Prof. Djurić on the study of new machine learning techniques for tackling the practical classification and prediction problems by using statistical and stochastic methods, E.G. non-parametric Gaussian process models. We focus on the challenging problems in Bayesian inference of latent variables, models (regimes), as well as the functions (mappings) between inputs and outputs of complex systems, such as perinatal outcomes and cerebral circulation.
1. Class-imbalanced learning using Ensemble Gaussian Processes
Classification with imbalanced data is a common and challenging problem in many practical machine learning problems. Ensemble learning is a popular solution where the results from multiple base classifiers are synthesized to reduce the effect of a possibly skewed distribution of the training set. In this work, we attempt to employ Gaussian process models for inferring the predictive distributions of test samples in an ensemble structure.
2. Time-Series Anomaly Detection using Phase Space Reconstruction
Anomaly detection of rare events or entire segments of time series data is an important but still challenging task in applications. Without prior knowledge of signs, features, and labels, the unsupervised approach of time series anomaly detection is studied based on the concept of phase space reconstruction and manifolds. By exploring the "trajectory" of signal in latent space, we are developing new methods for differentiating time-series samples and detecting outliers from a data set.
Application: Electronic Fetal Monitoring and Assessment
Computer-aided fetal monitoring has continuously developed over the past decade. Electronic signals assist clinicians to reduce the risk of fetal hypoxia and acidosis by timing surgical interventions during labor. However, human subjective judgments are often unreliable. The purpose of this work is to develop new approaches of interpreting electronic fetal monitoring signals to predict fetal status with the aid of machine learning methods, thereby leading to better estimation of risk to the fetus and the decrease of operative vaginal delivery and cesarean delivery.
Past
During my Master's studies, Dr. Le Yang and Dr. K. C. Ho brought me into my first research project "multistatic sonar localization" in which we investigated the effects of Doppler measurements on estimating moving target velocity and position in multistatic sonar. We evaluated theoretically the contribution of Doppler measurements via Cramer-Rao bound analysis and developed computationally attractive algorithms for multistatic sonar localization.
