My research area is mainly related to future networks which include SDN, NFV, and IRS. I work on wireless networks on the physical layer and network layer both. I worked on several research projects related to SDN, WBAN and IRS. Currently, I am working on 5G network slices with the assistance of IRS to improve the data offloading process between IOT devices and MEC server. We are trying to build a network simulator which allows the user to deploy the wireless network and test the network performance.

My research interests focus around studying the energy consumption of various processes on mini clusters residing on the fog layer consisting of heterogeneous nodes. Also, energy consumption is a crucial factor on edge devices where there are energy restrictions for battery-based devices. For this to be done, we need to have the ability to run workloads with different hardware demands in an automated way and being able to extract useful information about energy consumption on process level. Given that we have this information we can work towards finding ways reducing the overall energy spend for the cluster and ways into incorporating green energy in cases where this is available.

The world we live in can be often described in terms of networks: social networks, computer networks, networks of Internet of Things or Internet of Sports devices. Understanding the phenomena underlying these networks is fundamental in order to advance our society. My work within the RAIS project focuses on two key issues related to this topic.
The first issue is how to use Machine Learning to understand the structure of complex networks and predict their evolution. The second issue is how to safely and effectively perform Machine Learning tasks on top of an existing network of devices, in a decentralized fashion, exploiting the computational power of all these devices, while still maintaining their data private. The final goal of my research is to be able to combine the solutions of these two issues, known as Graph Representation Learning and Decentralized Privacy-Preserving Machine Learning, in a way that can be directly applied to a variety of tasks.
As an example application, consider an Internet of Sports sensor/app that can record the user activity and his/her interactions with other users. With these technologies, it would be possible for the app to analyze the behaviour of the user and provide exercise recommendations, without sharing any raw information with any central entity, thus providing the best user experience without compromising privacy.

The project I am working on aims at designing and developing a model and a system to protect user personal data in IoT ecosystems by moving from a service-centric way of securing personal data (which is today the de facto standard) to a user-centric approach. This will empower individuals with tools to manage their privacy preferences over data produced through different IoT/wearable devices. Adopting a user-centric model will allow us to design a system that offers individuals more fine-grained control on how their data are released, we plan to achieve an information-sharing approach driven also by trust relationships with data consumers. The system will be complemented with tools leveraging on data mining and machine learning techniques able to: automatically adjust user privacy settings when new devices enter the personal pervasive ecosystem and adapt privacy settings to change with minimal user intervention.

Here at FORTH within RAIS project we are trying to discover and address security and privacy limitations of centralized handling for sensitive IoT/IoS health data. In particular, we are investigating ways to disclose sensitive personal information by intercepting and analyzing data (potentially encrypted) sent by wearable devices to the cloud. To do that we are employing cryptoanalysis and various machine learning techniques. Cryptanalytic approaches include comparing pairs of plain text and encrypted data, as well as analyzing specific bits of encrypted payload. Machine learning approaches are based on identifying relevant features of encrypted data: DNS, IP, ports, time intervals, etc. Furthermore, we are studying existing cybersecurity attacks on wearable devices, and ways to mitigate them. We are replicating these attacks, and designing pipelines to neutralize them. Finally, we are trying to scrutinize impact of General Data Protection Regulation (GDPR) on IoS. We are surveying whether major wearable devices manufacturers are complying with GDPR.

The emergence of new data-driven technologies for the Internet of Sports creates new ways of helping people in their everyday activities, but on the other hand it also raises privacy concerns with regard to their sensitive data. Nowadays, most of the IoS data are sent from users’ devices towards proprietary clouds and databases, where users lose control over them. The main topic of my research at FORTH, within the RAIS project, consists in assessing the privacy risks related the use of these centralized systems in the Internet of Sports domain, with a focus on wearable devices, and evaluating the possible countermeasures. Moreover, I study decentralized alternatives that reduce such risks without impacting on the efficiency of IoS services.

At the University of Insubria, we are currently working on security and privacy for IoT/IoS devices. We explore privacy-preservation techniques for processes involving data sharing in IoT environment. Besides, we are working on decentralized malware detection in IoT networks by analyzing network flows. For this matter, we adopt blockchain as a distributed solution to ensure transparency and immutability of malware detection results in an IoT setting. Our research takes into consideration the exclusive aspects of IoT, such as low computational power, in order to develop efficient algorithms to ensure the operability of the IoT devices while protecting them.

Networks are popular data structures that embed relationships between entities in edges between nodes. Although they have many applications, including, but not limited to, Internet of Sports, it is hard to extract insights into the roles of the nodes. Roles are able to capture the higher-order connectivity patterns of the nodes and thus to emphasize the behaviors of the entities.
Within RAIS, I am focusing on role discovery, which has many applications ranging from data anonymization to online anomaly detection. Particularly, I will research into possible ways of carrying out an analysis of roles in networks in a distributed manner.
Currently, I am reviewing the topic of role discovery in order that strengths and weaknesses in current methods may be recognized.

My present research work focusses on motion modelling using artificial intellgience. Presently I am exploring the domain of human pose estimation in a minimalistic approach. Primarily, we are focussing on estimating human pose only from one IMU sensor providing accelerometer readings. This approach would enable a low cost pose estimation technique that can easily be integrated with any IOT ecosystem revolving around accelerometer readings (e.g. mobile phone) and will greatly contribute towards personalized coaching feedback for athletes (for whom a good pose equivalents to good technique. Furthermore, it can be extended to multiple other use cases of health and well being. We will build a system leveraging machine learning techniques that would take in accelerometer signals from running users of Racefox and estimate poses for those users in a fast and effective manner. Furthermore, we plan to extend our research to explore the possibility of training our deep learning/machine learning models over a distributed infrastructure and experiment with different edge settings for deployment of our model.  Additionally, we also plan to explore the scope of reinforcement learning for personalization and recommendation.

My current research focuses on designing distributed/decentralized algorithms for personalized machine Learning for IoT, where only access to partial information is assumed. Supervised and unsupervised ML algorithms for decentralized setting are being investigated. Gossip learning will be compared to federated learning under different proposed conditions such as homogeneity and security vs trustiness in the network. Furthermore, we plan to extend our research to explore decentralized context-aware learning will be explored. Leveraging information from the surrounding context can be very useful. For this part of the project, a key assumption and motivation is that incorporating contextual information can improve the results of the model globally for the network, and more personally for the nodes. Towards that end, different applications and tasks are intended to be explored such as personalized recommendation and graph representation learning in decentralized fashion. Among the machine learning algorithms, the graph neural networks are to be explored, as well.   

The World Health Organization aims to increase physical activity by 15% by 2030. Since exercise is socially contagious, interdependent rather than independent interventions need to be studied. My research within the RAIS project will amalgamate the interdisciplinary fields of data analysis, machine learning and marketing towards identifying habitual patterns among fit individuals and determining strategies in which exercise can be promoted. The impact of COVID-19 on physical exercise will also be studied for determining in what form the habit of exercise was replaced. This knowledge will guide us towards more effective decision-making.

Machine learning is undoubtedly a revolutionary technology in closing the gap between personalized services and health care. At the moment, the research activity undertaken at the RAIS program is developing advanced (deep) machine learning systems capable of early prediction of potential deterioration of leg (knee) status.  This continuous risk prediction of upcoming injuries, especially in athletes who have undergone in the past some knee surgery, could mitigate any re-occurrence of adverse events at the same time comforts the recovery period.
The current roadmap for this issue contains a combination of recurrent neural networks and model-based reinforcement learning. A follow-up within the scope of injury prevention is also a personalized guidance system able to provide continuous recommendations for potentially risky situations.

The gap between wearable devices in the research world and commercial technologies has been decreasing in the last few years. My research interests within the RAIS project lie in understanding how peoples’ emotions change, focusing on stress and the severity of depressive symptoms. My goal is to study the between- and within-person variability regarding emotions, stress, and mental outbreaks. I aim to connect my research to different disciplinary fields: data analysis, machine learning, and mental health studies. This approach aims to predict daily mood and the presence of depressive symptoms to detect risks and possibly help mitigate anxiety and depression.

My research activity within the RAIS project is focused on the analysis of behavioral patterns to achieve sustained user engagement. Wearable devices and technological solutions for physical well-being suffer from high attrition rates, while having immense unexplored potential in terms of personalization at the same time. To unveil the unexploited possibilities of wearable devices, I am currently conducting an in-depth, interdisciplinary literature review, to identify current research gaps in the field of healthy habit formation and user engagement. Hence, my work lies in the intersection between Computer Science, Behavioral Sciences, and Sports Science. My goal is to utilize technology as a driver of change, rather than a secondary tool in the Internet of Sports (IoT) domain. To this end, I am planning to exploit state-of-the-art, privacy-preserving Machine Learning and analytics, to explore the habits, behaviors, motivations and barriers of people when it comes to physical well-being, and help them run more active lives in the long-term through the use of wearable technology.