Source author record

Daphney-Stavroula Zois

Daphney-Stavroula Zois appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Systems and Control math.OC Machine Learning

Catalog footprint

What is connected

4works

3topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2020arXiv

On-the-Fly Joint Feature Selection and Classification

Joint feature selection and classification in an online setting is essential for time-sensitive decision making. However, most existing methods treat this coupled problem independently. Specifically, online feature selection methods can handle either streaming features or data instances offline to produce a fixed set of features for classification, while online classification methods classify incoming instances using full knowledge about the feature space. Nevertheless, all existing methods utilize a set of features, common for all data instances, for classification. Instead, we propose a framework to perform joint feature selection and classification on-the-fly, so as to minimize the number of features evaluated for every data instance and maximize classification accuracy. We derive the optimum solution of the associated optimization problem and analyze its structure. Two algorithms are proposed, ETANA and F-ETANA, which are based on the optimum solution and its properties. We evaluate the performance of the proposed algorithms on several public datasets, demonstrating (i) the dominance of the proposed algorithms over the state-of-the-art, and (ii) its applicability to broad range of application domains including clinical research and natural language processing.

preprint2014arXiv

Controlled Sensing: A Myopic Fisher Information Sensor Selection Algorithm

This paper considers the problem of state tracking with observation control for a particular class of dynamical systems. The system state evolution is described by a discrete-time, finite-state Markov chain, while the measurement process is characterized by a controlled multi-variate Gaussian observation model. The computational complexity of the optimal control strategy proposed in our prior work proves to be prohibitive. A suboptimal, lower complexity algorithm based on the Fisher information measure is proposed. Toward this end, the preceding measure is generalized to account for multi-valued discrete parameters and control inputs. A closed-form formula for our system model is also derived. Numerical simulations are provided for a physical activity tracking application showing the near-optimal performance of the proposed algorithm.

preprint2014arXiv

Nonlinear POMDPs for Active State Tracking with Sensing Costs

Active state tracking is needed in object classification, target tracking, medical diagnosis and estimation of sparse signals among other various applications. Herein, active state tracking of a discrete-time, finite-state Markov chain is considered. Noisy Gaussian observations are dynamically collected by exerting appropriate control over their information content, while incurring a related sensing cost. The objective is to devise sensing strategies to optimize the trade-off between tracking performance and sensing cost. A recently proposed Kalman-like estimator \cite{ZoisTSP14} is employed for state tracking. The associated mean-squared error and a generic sensing cost metric are then used in a partially observable Markov decision process formulation, and the optimal sensing strategy is derived via a dynamic programming recursion. The resulting recursion proves to be non-linear, challenging control policy design. Properties of the related cost functions are derived and sufficient conditions are provided regarding the structure of the optimal control policy enabling characterization of when passive state tracking is optimal. To overcome the associated computational burden of the optimal sensing strategy, two lower complexity strategies are proposed, which exploit the aforementioned properties. The performance of the proposed strategies is illustrated in a wireless body sensing application, where cost savings as high as $60\%$ are demonstrated for a $4\%$ detection error with respect to a static equal allocation sensing strategy.

preprint2013arXiv

Active Classification for POMDPs: a Kalman-like State Estimator

The problem of state tracking with active observation control is considered for a system modeled by a discrete-time, finite-state Markov chain observed through conditionally Gaussian measurement vectors. The measurement model statistics are shaped by the underlying state and an exogenous control input, which influence the observations' quality. Exploiting an innovations approach, an approximate minimum mean-squared error (MMSE) filter is derived to estimate the Markov chain system state. To optimize the control strategy, the associated mean-squared error is used as an optimization criterion in a partially observable Markov decision process formulation. A stochastic dynamic programming algorithm is proposed to solve for the optimal solution. To enhance the quality of system state estimates, approximate MMSE smoothing estimators are also derived. Finally, the performance of the proposed framework is illustrated on the problem of physical activity detection in wireless body sensing networks. The power of the proposed framework lies within its ability to accommodate a broad spectrum of active classification applications including sensor management for object classification and tracking, estimation of sparse signals and radar scheduling.