In this talk, we discuss a caching problem that resembles a lossy Gray-Wyner network. A source produces vector samples with K components from Gaussian distribution (K-Gaussians) and the user is interested in decoding only one component. As in practical scenario, the encoder first places a message in the user’s cache without the knowledge of her preference and later sends an update after knowing user’s request. The goal is to compute the trade-off between cache and average update rate required to satisfy certain end distortion constraint. This involves solving a non-convex optimization problem and we consider two alternative approaches. In the first approach, we focus on Gaussian vector sources with circulant correlation matrix and we analytically compute an inner bound to the trade-off and argue about its tightness in the high cache rate region. In the second approach, we perform numerical analysis of the caching problem and introduce two techniques, namely - Domain Re-Parametrization and Optimization Function modification to deal with the non-convex nature of the original problem. In this analysis, we do NOT restrict ourselves to sources with circulant correlation matrix and consider the general case. These modifications make the problem convex for 2-Gaussians and we provide partial convexity proofs for K(>2)-Gaussians. We compute an approximation to the trade-off for both circulant and non-circulant case. We show that the computed inner bound is tight in the low cache rate region as well for the circulant case.
Rohan Pote completed his master's study in Communication Systems at École Polytechnique Fédérale de Lausanne (EPFL) in February, 2017. He wrote his master thesis on coded caching with Gaussians, under the supervision of Prof. Michael Gastpar. He was a recipient of M.Sc Research Scholarship at EPFL and worked under Prof. Bixio Rimoldi on WiFi device localization using flying ad-hoc networks. He was also a graduate student researcher at the European Technology Center of SONY, Germany in 2016 and worked on Indoor Localization based on Geomagnetic Fingerprinting. He did his B.Tech in Electrical Engineering at IIT Jodhpur in 2014.