I. Time series comparison/Anomaly detection/Time series classification (supervised learning) & clustering (unsupervised learning) in terms of dynamics
Learning from data in both the time & frequency domain
Health monitoring from continuous health data from sensors such as wearable devices.
Biomedical study
Pattern of hormone levels at different stages
Damage detection of a mechanical system
Heart beat rates:
differences in the frequency patterns
II. Streaming data: a continuous flow of data
Online algorithms for variance/covariance matrix estimation
Adaptive spectral variance estimation for large scale Monte Carlo simulation data
III. Statistical finance: financial time series are often nonlinear and heavy tailed.
1). to develop new distance metric which is able to capture non-linear dynamics in financial time series for financial time series classification and clustering.
2). to extract new dynamic features and integrate them into learning models for volatility modeling & prediction.
3). to detect structure change/stationarity over a period
Gulf of Mexico
Multiple data sources
IV. Use statistical and physical models with data science methods such as data assimilation, adaptive filtering to analyze a suite of chemical variables from multiple sources (cruises, Wave Glider, In-situ measurements, satellite, etc.) at different temporal and spatial resolutions to understand the temporal (seasonal) and spatial (4 dimensional) dynamics of pH, Ωarag, and the relationships between environmental factors in the Gulf of Mexico.
Longitudinal data analysis via semi-parametric partially linear models to study how CD4 cell numbers depends on different factors.
