Event

Ph.D. Research Proposal Exam: Ismail Cosandal

Thursday, July 31, 2025
3:30 p.m.

Souad Nejjar
301 405 8135
snejjar@umd.edu

ANNOUNCEMENT: Ph.D. Research Proposal Exam

Name: Ismail Cosandal

Committee:

Professor Sennur Ulukus (Chair)

Professor Zhambyl Shaikhanov

Professor Behtash Babadi

Date/time: July 31, 2025 at 3:30 PM

Title: Age of Incorrect Information (AoII) Optimization in Remote Monitoring

Abstract: 

In remote estimation problems, timeliness and correctness of the estimation play a crucial role. In this proposal, we consider several different system models, and aim to obtain optimal sampling policies for them. In our first completed work, we consider a sensor node that observes a random process while it experiences intermittent failures. We utilize a multi-goal communication approach where each successful transmission serves two purposes: i) timely tracking of the observed process, and ii) detecting the failure of the sensor. In our second completed work, we investigate the average age of incorrect information (AoII) minimization problem for push- and pull-based update models. In this work, we propose estimation-based policies for both update models and analyze them by utilizing a phase-type (PH) distribution. In our third completed work, we focus on push-based update model, and study optimum policies under sampling rate constraints. We show that the optimum transmission policy is a multi-threshold policy whose thresholds depend on the estimation and source processes. We extend PH to the multi-regime phase-type (MR-PH) distribution that allows us to find average AoII for a given multi-threshold policy. In our fourth completed work, we consider the AoII minimization problem with a pull-based update model under a pull-request rate constraint. We consider a fixed delay on the channel, thus, the monitor never has the exact knowledge of the state of the source and the AoII value. That is, we investigate an AoII minimization problem without knowing the AoII. The main contribution of this work is to propose a \emph{belief} that corresponds to the joint probability of both the state of the source process and the AoII value. We show that this belief is the best we have, and thus, the optimum policy should be a function of it. Finally, in our fifth completed work, we consider a joint-Markov process where each of its components is observed by a different sensor. From a similar motivation to the previous work, we formulate the belief, which is sufficient statistics for the problem. We also utilize two types of model predictive control (MPC) algorithms.

Next, we present three proposed works. In our first proposed work, we aim to adapt the DR-PH method for the discrete-time setting, which will be a first in the literature. Additionally, different from the minimization problem for the average AoII in our previous works, we aim to minimize a general function of the AoII. In our second proposed work, we consider a single-bit quantization method under periodic sampling and constant delay for the timely tracking of a Brownian process, where in each period, the difference between the process and its estimation will be quantized, and the estimation process will be reconstructed by the monitor upon receiving the update. In our third proposed work, we consider a source coding method for the timely tracking of a Markov process with finite number of states. The conventional methods for Markov processes only consider the last transmitted symbol to generate a codebook. However, in a timely tracking problem, the code length for each symbol affects the probability of the next symbol. Therefore, we propose a stochastic control problem formulation that considers this effect while designing joint codebooks.