Below, you can find a short summary of our current and previous research projects. If you want to learn about some of our new exciting projects, contact Prof. Daphney-Stavroula Zois ([email protected]).

Self-driving cars and home assistants provide just a small glimpse of the future cost-costrainted complex cyber-physical-human systems (CPHS) that will integrate engineering systems with the natural word and humans. This project will devise new mathematical tools and methods to systematically describe CPHS and optimize their operation. The application focus is on wireless body area networks, a natural CPHS representative with humans in the loop, heavily resource-constrained operation, and heterogeneous components that are intertwined with and altered by human behavior. The end result will help understand important factors related to the operation of CPHS and how to optimize their operation.

Those in need of help often do not know how to locate or access service providers. Likewise, service-providing agencies often work in silos. The lack of communication also applies to volunteers; people do not know who to help and how they can be resourceful. Response becomes even more problematic when a problem demands the coordination of service providers, volunteers, and government structures, and after business hours, when the communication channels that can aid people in need become sparse. The focus of this project is to (i) develop new data mining methods for uncovering complex interdependencies within a dynamic sociotechnical system, (ii) devise novel information processing, machine learning, and control methods to dynamically optimize delivery of human and physical services under uncertainty with humans in the decision-making loop, and (iii) shed light on the ability of communities to integrate emerging technologies to become more connected in human interactions.

Brain-computer interfaces (BCIs) are devices that enable people to control computer systems using brain activity. Since they require little to no voluntary motor control, they can help people with severe motor deficits (e.g., locked-in syndrome) to communicate, but can have applications for healthy individuals as well (e.g., multimedia and gaming). Steady-state visual evoked potential (SSVEP)-based BCIs are one common type of BCIs, where users are presented with a set of stimuli, each flashing at a unique frequency, and attention to one of these stimuli elicits changes in brain activity at the fundamental and higher harmonic frequencies of the flashing — an SSVEP — that can be measured using electroencephalography (EEG). The goal of this project is to design classifiers that meet various strict specification requirements for different (SSVEP)-based BCIs applications.

Bullying, once limited to physical spaces (e.g., schools, workplaces or sports fields) and particular times of the day (e.g., school hours), can now occur anytime, anywhere. Cyberbullying can take many forms, however, it typically refers to repeated and hostile behavior (e.g., hurtful comments, videos and images) performed in an effort to intentionally and repeatedly harass or harm individuals. The consequences can be devastating: learning difficulties, psychological suffering and isolation, escalated physical confrontations, suicide. While techniques to automatically detect cyberbullying incidents have been developed, the scalability and timeliness of existing cyberbullying detection approaches have largely been ignored. The goal of this project is to derive provably optimal, yet scalable online strategies to minimize the time-to-detection of cyberbullying incidents.

In recent years, urban mobility demand has become highly variable over time challenging the sustainability of transportation networks of major cities. At the same time, various types of incidents such as accidents, construction zone closures and weather hazards exacerbate the already congested transportation network. Timely detection of such events can offer an unprecedented opportunity to mitigate the consequences. Currently, we are developing a framework for continuous active collision detection in a road segment equipped with spatially distributed speed sensors of variable accuracy. Our goal is to design appropriate low-complexity algorithms that can quickly estimate the existence of a collision by appropriately selecting which sensors to query and when.

Fine–grained knowledge of human context information in terms of physical activity (e.g., sit, walk), emotional state (e.g., happy, sad), surroundings (e.g., home, work) and mobile phone usage (e.g., web search, listen to music) offers an unprecedented opportunity to objectively and accurately understand when, where and what type of behaviors are exhibited. Currently, we are building an off-the-self body sensing platform and designing an appropriate mathematical framework tha will formally integrate sensor capabilities, biomedical, contextual and other variables (e.g., channel information) and their interactions, and will enable accurate monitoring of human context information.

Wireless Body Area Networks (WBANs) that consist of heterogeneous biometric sensors (e.g., heart–rate monitors, accelerometers) and an energy–constrained personal device (e.g., mobile phone, PDA), have the transformative potential to influence a wide range of applications including health–care, sports, military and emergency applications. However, the practical realization of a WBAN is hindered by a number of unique challenges, including energy constraints that significantly impact its lifetime. To address this issue, we have proposed novel resource allocation strategies that maximize the network lifetime by minimizing energy spent at the energy–constrained fusion center. The effectiveness of our proposed methods have been evaluated using extensive simulations on real data collected from a real prototype WBAN, the KNOWME network, used for preventing obesity. We have showed that we can improve energy gains by as much as 68% compared to state–of–the–art schemes, without compromising detection accuracy.

Object detection is a fundamental, yet very challenging task in image analysis. A typical object detection algorithm first generates a set of region proposals. These can be either fixed and class–independent or generated on the fly while searching for a particular object. Each such proposal is then assigned a class label by running a set of detectors. Evaluating a large number of region proposals will certainly lead to high detection accuracy, but will incur high computational costs if each detector evaluation is computationally expensive. In this line of work, we proposed an object detection algorithm that models image regions as vertices and overlap relationships as edges in a directed weighted graph. Information is propagated from labeled vertices through graph edges that operate as noisy channels via message passing over locally informative trees that are extracted from the original graph using an information-theoretic criterion. Influential vertices are determined by an appropriate centrality index. Our algorithm can be applied on top of any state–of–the–art region proposal method as it treats it as a black box. We evaluated the performance of our proposed algorithm on different scenarios and showed that in some cases only 0.45% of the total regions is evaluated with maximum 21.45%.

Sensor heterogeneity necessitates the design of sensing algorithms to achieve inference, the quality of which drives the adaptation of the associated algorithm. Thus, sensing and tracking are tightly interconnected. In our work, we developed a generalized framework for active state tracking with heterogeneous observations, where the decision maker selects which observations to consider at each step based on accuracy and cost criteria. Various criteria were considered along with their effect on accuracy and were generalized appropriately to address the intertwined key problems of information characterization and unified sensing and tracking. Based on this framework, we developed fundamental theory that enabled us to characterize the structure of the optimal sensing algorithm and devise appropriate scalable algorithms.