Athanasia Panousopoulou

Feature selection for performance characterization in multi-hop wireless sensor networks

Current trends in Wireless Sensor Networks are faced with the challenge of shifting from testbeds in controlled environments to real-life deployments, characterized by unattended and long-term operation. The network performance in such settings depends on various factors, ranging from the operational space, the behavior of the protocol stack, the intra-network dynamics, and the status of each individual node. As such, characterizing the networks high-level performance based exclusively on link-quality estimation, can yield episodic snapshots on the performance of specific, point-to-point links. The objective of this work is to provide an integrated framework for the unsupervised selection of the dominant features that have crucial impact on the performance of end-to-end links, established over a multi-hop topology. Our focus is on compressing the original feature vector of network parameters, by eliminating redundant network attributes with predictable behavior. The proposed approach is implemented alongside different cases of protocol stacks and evaluated on data collected from real-life deployments in rural and industrial environments. Discussions on the efficacy of the proposed scheme, and the dominant network characteristics per deployment are offered.

An online feature selection architecture for Human Activity Recognition

Human Activity Recognition (HAR) must currently face up to the challenge of rethinking analytics from the perspective of real-time operation, wherein biophysical sensing streams are efficiently intertwined at close vicinity to the point of sensing. As such, feature selection techniques, traditionally employed for off-line data processing, should be evaluated with respect to their ability to filter out redundant information in real-time. In this work, we propose an online architecture for implementing feature selection on mobile devices, and we evaluate popular feature selection methods against constantly alternating activity labels. We perform a qualitative analysis to determine the dominant sensing modality that dictates the activities of a certain time duration. The results indicate that online feature selection performance changes among consecutive data partitions, leading to the conclusion that the type of available activity influences significantly the feature selection procedure.

Matrix and Tensor Completion on a Human Activity Recognition Framework

Sensor-based activity recognition is encountered in innumerable applications of the arena of pervasive healthcare and plays a crucial role in biomedical research. Nonetheless, the frequent situation of unobserved measurements impairs the ability of machine learning algorithms to efficiently extract context from raw streams of data. In this paper, we study the problem of accurate estimation of missing multimodal inertial data and we propose a classification framework that considers the reconstruction of subsampled data during the test phase. We introduce the concept of forming the available data streams into low-rank two-dimensional (2-D) and 3-D Hankel structures, and we exploit data redundancies using sophisticated imputation techniques, namely matrix and tensor completion. Moreover, we examine the impact of reconstruction on the classification performance by experimenting with several state-of-the-art classifiers. The system is evaluated with respect to different data structuring scenarios, the volume of data available for reconstruction, and various levels of missing values per device. Finally, the tradeoff between subsampling accuracy and energy conservation in wearable platforms is examined. Our analysis relies on two public datasets containing inertial data, which extend to numerous activities, multiple sensing parameters, and body locations. The results highlight that robust classification accuracy can be achieved through recovery, even for extremely subsampled data streams.

A benchmark study on feature selection for human activity recognition

Human Activity Recognition (HAR) currently confronts the challenge of interpreting massive data streams to a significantly smaller number of activities. Thus, feature selection should be treated as an inseparable aspect of the HAR chain. In this work we perform an integrated study on feature selection, considering: (a) the generation of an expanded HAR dataset; (b) the development of a software tool that covers the entire feature-level fusion chain; (c) the calculation of performance metrics that go beyond machine learning terms. The results yield guidelines on the preferable feature selection technique that should be considered for adoption in the HAR domain.

Effects of matrix completion on the classification of undersampled human activity data streams

Classification of activities of daily living is of paramount importance in modern healthcare applications. However, hardware monitoring constraints lead frequently to missing raw values, dramatically affecting the performance of machine learning algorithms. In this work, we study the problem of efficient estimation of missing linear acceleration and angular velocity measurements, experimenting on a public Human Activity Recognition (HAR) dataset. We exploit the data correlation to formulate the problem as an instance of low-rank Matrix Completion (MC) within a general classification framework. We consider the effects of our proposed reconstruction method on the classification accuracy as related to the size of the training and test sets, and the single versus collective recovery. Additionally, we compare the performance of our approach with popular imputation and expectation maximization algorithms for treating missing measurements, in conjunction with several state-of-the-art classifiers. The results highlight that robust and efficient classification is feasible even with a substantially reduced amount of measurements.

Application of Matrix Completion on Water Treatment Data

Recent advances in Cyber-Physical Systems (CPS) have revolutionized water management in urban areas. Nevertheless, literature reports minor progress in introducing CPS-based systems at industrial water treatment plants, responsible for water purification. Such environments would greatly benefit by adopting CPS technologies in general, and Wireless Sensor Networks (WSNs) in particular. However, WSNs would suffer from a series of industrial monitoring constraints, which would inevitably lead to missing measurements. In this work, we study the problem of efficient estimation of missing water treatment data, collected over a WSN deployed at a water desalination plant, and we examine how redundancies can be used for the recovery of extremely undersampled matrices. We propose the formulation of the problem as an instance of low rank Matrix Completion (MC), and we employ the inexact Augmented Lagrange Multipliers (ALM) algorithm. We consider three key questions related to the performance of our method; namely, recovery from artificially introduced missing entries, single versus collective recovery of measurements matrices, and the real-valued problem of temporal super-resolution. The results highlight that MC is a valid method for estimating missing water-treatment data, even from a very limited number of measurements.

On realizing distributed topology control in low-power IoT platforms

As current trends in distributed computing are enabling the vision of the Internet of Things (IoT), the necessity intensifies for addressing real-life aspects in massive scales. Energy efficiency remains a key design requirement for low-power IoT platforms, and distributed topology control techniques can yield the guidelines for optimizing the transmission power-connectivity nexus. In this work, we exploit previous theoretical results on necessary and sufficient conditions for establishing end-to-end connectivity, by introducing the notion of relative Delaunay neighbourhoods to computationally constrained hardware platforms. We implement the proposed approach on Contiki OS and we offer extensive emulation results, which highlight the scalability of our approach. Comparisons with benchmark solutions are offered to evaluate the performance of our framework in terms of achieved connectivity, memory demands, and energy efficiency.

mMamee: A mHealth Platform for Monitoring and Assessing Maternal Environmental Exposure

Over the last years significant efforts have been made to prevent and/or minimize exposure to a wide range of environmental risks (e.g. air pollution and nutrition) that adverse health effects especially among vulnerable populations including pregnant women. Towards this direction, mHealth approaches can provide the means for remotely capturing the environmental factors that affect maternal health, and replace traditional, tedious ways of logging, e.g. face-to-face interviews. This work presents mMamee, a mHealth platform for monitoring and assessing maternal environmental exposure. mMamee employs a client-server architecture and addresses the integration of sensing data and descriptive input on maternal daily habits. The future application of this platform to monitor environmental exposure during pregnancy is outlined. The conclusions derived highlight the feasibility of mMamee for the realization of long-term epidemiological studies.

On graph-based feature selection for multi-hop performance characterization in industrial smart water networks

Recent deployments of Smart Water Networks in urban environments are causing a paradigm shift towards sustainable water resources management. Nevertheless, there exists a substantial gap on respective solutions for industrial water treatment. In such deployments the wireless network backbone would have to overcome limiting factors that span across different layers of a protocol stack. Incorporating data analytics for capturing multi-dimensional correlations could be extremely beneficial to the design of reconfigurable network protocols for industrial Smart Water Networks. In this work, we exploit recent findings in the arena of network measurements and we propose a graph-based unsupervised feature selection approach for extracting the dominant network conditions that affect the performance of user-defined links. We employ a real- life industrial Smart Water Network deployed in a desalination plant to evaluate the efficacy of our approach. Finally, we provide useful insights on how different locations in a desalination plant affect the performance of the network backbone.

Building a Smart City Ecosystem for Third Party Innovation in the City of Heraklion

This paper describes the implementation of an Internet of Things (IoT) and Open Data infrastructure by the Institute of Computer Science of the Foundation for Research and Technology—Hellas (FORTH-ICS) for the city of Heraklion, focusing on the application of mature research and development outcomes in a Smart City context. These outcomes mainly fall under the domains of Telecommunication and Networks, Information Systems, Signal Processing and Human Computer Interaction. The infrastructure is currently being released and becoming available to the municipality and the public through the Heraklion Smart City web portal. It is expected that in the future such infrastructure will act as one of the pillars for sustainable growth and prosperity in the city, supporting enhanced overview of the municipality over the city that will foster better planning, enhanced social services and improved decision-making, ultimately leading to improved quality of life for all citizens and visitors.