"Accurate Inference in High-Dimensions: Structure and Fundamental Limits"

Christos Thrampoulidis, Postdoctoral Research Associate,MIT

March 15th (Thursday), 10:00am
Harold Frank Hall (HFH), Rm 4164

The increasing availability of data and of computational resources has shifted the boundaries of what is deemed possible, creating an expectation for high-dimensional information extraction from observations that are incomplete and indirect, as well as, corrupted by noise, outliers and interference. Importantly, extracting useful signals in such instances relies critically on exploiting available structural information in ways that are statistically accurate, computationally efficient, and robust.

In this talk, I will describe a framework of analysis to accurately quantify the statistical performance and the robustness of non-smooth convex-relaxation methods that are used as powerful tools to extract low-dimensional signal structures in high-dimensions. In particular, the framework sheds light on the relative performance of different algorithms and allows the engineer to fine-tune the involved parameters to improve the system as a whole. Also, it applies to a wide range of measurement models under generic observations. To demonstrate the generality of the above framework, I will present its applications to massive MIMO detection, quantized compressed sensing, and phase-retrieval. At the core of the framework, lies a new theorem on Gaussian process comparisons, which I will also highlight.

Finally, I will discuss new opportunities in contemporary imaging problems that call for exploiting structure beyond signal-structure. In particular, I will demonstrate an efficient computational imaging technique that exploits opportunistically the presence of occluding objects, enabling for the first time imaging of hidden scenes without reliance on ultrafast time-of-flight measurements.

About Christos Thrampoulidis:

Photo of Christos Thrampoulidis Christos Thrampoulidis received his Diploma in electrical and computer engineering from the University of Patras, Greece in 2011. He received a M.Sc. and a Phd. degree in electrical engineering in 2012 and 2016, respectively, both from the California Institute of Technology (Caltech), with a minor in applied and computational mathematics. He is currently a Postdoctoral Research Associate at the Research Laboratory of Electronics (RLE) at the Massachusetts Institute of Technology (MIT). His research interests include convex optimization, high-dimensional probability and statistics, signal processing, and computational imaging. He is a recipient of the 2014 Qualcomm Innovation Fellowship, and, of the Andreas Mentzelopoulos Scholarship and the I. Milias awards (2011) from the University of Patras.

Hosted by: BS Manjunath