Speaker: Lekshmi Ramesh
Affiliation: PhD, ECE Dept, IISc Bangalore
Date and Time : October 29, 2021 (Friday), 5:45 PM to 6:30 PM
Talk Recording YouTube Link: To be updated
Talk Abstract : In this talk, I will describe the problem of multiple support recovery, where we are given access to linear measurements of multiple sparse samples. These samples can be partitioned into l groups, with samples belonging to the same group having a common support. For a given budget of m linear measurements per sample, the goal is to recover each of the underlying supports, in the absence of the knowledge of group labels. We study this problem with a focus on the measurement-constrained regime where m is smaller than the support size k of each sample. I will describe a two-step spectral algorithm, where we first estimate the union of the underlying supports, and then use spectral clustering on a fourth order statistic of the data to estimate the individual supports. Under a general, generative model assumption on the samples and measurement matrices, we will see that this estimator can recover the supports with fewer than k measurements per sample, from roughly k^4 l^4/m^4 samples.
Speaker Biography:
Lekshmi Ramesh recently completed her PhD from the department of ECE, IISc, where she worked with
Prof. Chandra R. Murthy and Prof. Himanshu Tyagi. Her interests are in the areas of compressed sensing,
statistics, information theory, and privacy.