Most statistics and machine learning problems can be phrased as an optimization problem. Traditional convex optimization problems in statistics and machine learning are relatively well understood. By contrast, my research focuses on answering how can we solve non-convex statistical learning problems?
In the era of big data, high-dimensional graphical models have become a standard tool for inferring and understanding conditional independence relationships between variables in high-dimensional random vectors. When the data becomes extremely high dimensional relative to the sample size, it may become difficult to interpret, and even to detect, the conditional independence relationships between components in these vectors.