Event

Ph.D. Research Proposal Exam: Sahan Liyanaarachchi

Wednesday, July 30, 2025
10:30 a.m.

Souad Nejjar
301 405 8135
snejjar@umd.edu

ANNOUNCEMENT: Ph.D. Research Proposal Exam

Name: Sahan Liyanaarachchi

Committee:
Professor Sennur Ulukus (Chair)
Professor Behtash Babadi
Professor Sanghamitra Dutta 

Date/time: Wednesday, July 30, 2025 at 10:30 AM

Title: Age of Estimates: Freshness in Remote Estimation Applications

Abstract: Many remote estimation applications revolve closely around tracking and monitoring of a stochastic process across different spatial localities. For these applications,  mean squared error (MSE) of the estimates has been the most frequently used quantifier for the effectiveness of the tracking procedure. However, in the past decade there have been a growth in studying these systems from a freshness perspective. Indeed, some of the prior works highlight the direct corelation between freshness of estimates and the MSE. Since the communication channel between the remote monitor and the stochastic source is often modeled as a queue, this newly realized freshness based view point have revitalized and invigorated the synergy between queuing theory and remote estimation. In particular, freshness based metrics such as binary freshness, age of information (AoI) and age of incorrect information (AoII) are more analytically tractable and quite often lead to more theoretically rich solutions compared to the MSE.

In this work, we study several remote estimation applications mainly through a freshness lens and  devise freshness optimal sampling and scheduling strategies. Among our completed works, we first look at the problem of remote monitoring a stochastic process under delayed feedback where we show that periodic early sampling (i.e., sampling before ACKs) leads to an asymptotically optimal stationary policy. In our next completed work, we study sequence-based (i.e., cyclic) scheduling schemes for minimizing the average AoI of multiple heterogeneous sources in the presence of packet errors. In this regard, we give the optimal cyclic scheduling scheme for two sources and build upon it to develop low complexity  and scalable algorithms for constructing cyclic scheduling schemes for multiple sources. In our third completed work, we study the problem of multi-path transmissions and develop scheduling schemes to minimize the AoI. Here, we utilize an absorbing Markov chain to compute the exact distribution of AoI while accounting for complications such as out of order transmissions which are inherent to these multi-path settings.

Finally, we propose four new works, the first of which looks into source coding schemes for a Wiener process where we propose a novel event-driven sampling scheme with monotone function thresholds to efficiently transmit a Wiener process over a delay channel. In the next proposed work, we look into a set of novel estimators termed as structured estimators. Here, we aim to integrate the analytical feasibility of martingale estimators with that of maximum a posteriori (MAP) estimators which are known to be optimal especially in pull-based status update systems. In the next proposed work, we look into the problem of jointly tracking and utilizing a Markov source which oscillates between free and busy states. We term them as Markov machines (MM) and introduce two new metrics to evaluate the efficacy of the underlying decision process. Lastly, we propose an extension to the study of Markov machines  where we find the optimal decision process to maximize the utility of these machines which we define to be the average revenue generated by the MM.