Research

I am interested in the exciting problems arising in optimization, control and machine learning area. Here are some of my ongoing projects.

 

Optimization and Control Using Machine Learning

Traditional optimization problems often involve complicated modeling stages. In many control problems, problems are solved by first building an accurate enough model (e.g., parameter estimation), and then trying to find optimal solution via solving optimal control or other optimization problem.

Here we try to extract the potential of machine learning not only in regression, but also in are interested in using neural network not only as an accurate predictor, but also as an optimizer.

 

Generative Models in Time-Series Modeling and Forecasting

Recent years have seen the great advances in both the algorithm side and application side of generative models. We propose a general training and optimization framework based on Generative Adversarial Networks (GANs), which could be applied for scenario generation and forecasts problems of renewables generation as well as demand response profile analysis.

 

Adversarial Learning

This area is at the intersection of traditional computer security research and machine learning (ML). On the one hand, more and more newly-developed machine learning systems suffer from security issues. Research found that some trivial revisions on inputs would significantly reduce ML system’s performance. While on the contrary, the sophisticated structure of advanced attacks of threats have raised our interests in applying ML algorithms for anomaly or intrusion detection.