PUBLICATIONS

Skeleton-based human action recognition with Graph Convolutional Networks is an active research field that has gained increased popularity over the last few years. A challenge in skeleton-based action recognition is the design of a model in a way that captures fine-grained motions and the relations between the movements of different parts of the skeleton towards the recognition of specific actions. In this paper, the use of a set of part-aware graphs for the skeleton representation is proposed aiming to enhance discrimination between actions in the recognition task since each action put emphasis on specific parts of the skeleton. Extensive experimental work has been carried out in a consistent evaluation framework taking into account different combinations of part-aware graphs and feature representations leading to a configuration that achieves the optimal balance. Based upon two well-established datasets, namely NTU RGB+D and NTU RGB+D 120, we demonstrate that the proposed methodology compares favourably with the state-of-the-art.

Zois Tsakiris, Lazaros Tsochatzopoulos, Ioannis Pratikakis

The efficient control of HVAC devices in building structures is mandatory for achieving energy savings and comfort. To balance these objectives efficiently, it is essential to incorporate adequate advanced control strategies to adapt to varying environmental conditions and occupant preferences. Model-free control approaches for building HVAC systems have gained significant interest due to their flexibility and ability to adapt to complex, dynamic systems without relying on explicit mathematical models. The current review presents the recent advancements in HVAC control, with an emphasis on reinforcement learning, artificial neural networks, fuzzy logic control, and their hybrid integration with other model-free algorithms. The main focus of this study is a literature review of the most notable research from 2015 to 2023, highlighting the most highly cited applications and their contributions to the field. After analyzing the concept of each work according to its control strategy, a detailed evaluation across different thematic areas is conducted. To this end, the prevalence of methodologies, utilization of different HVAC equipment, and diverse testbed features, such as building zoning and utilization, are further discussed considering the entire body of work to identify different patterns and trends in the field of model-free HVAC control. Last but not least, based on a detailed evaluation of the research in the field, the current work provides future directions for model-free HVAC control considering different aspects and thematic areas.

Panagiotis Michailidis, Iakovos Michailidis, Dimitrios Vamvakas, Elias Kosmatopoulos

Presence sensing systems are gaining importance and are utilized in various contexts such as smart homes, Ambient Assisted Living (AAL) and surveillance technology. Typically, these systems utilize motion sensors or cameras that have a limited field of view, leading to potential monitoring gaps within a room. However, humans release carbon dioxide (CO₂) through respiration which spreads within an enclosed space. Consequently, an observable rise in CO₂ concentration is noted when one or more individuals are present in a room. This study examines an approach to detect the presence or absence of individuals indoors by analyzing the ambient air’s CO₂ concentration using simple Markov Chain Models. The proposed scheme achieved an accuracy of up to 97% in both experimental and real data demonstrating its efficacy in practical scenarios.

Christos Karasoulas, Christoforos Keroglou, Eleftheria Katsiri, Georgios Ch. Sirakoulis

This work describes a methodology for sound event detection in domestic environments. Efficient solutions in this task can support the autonomous living of the elderly. The methodology deals with the “Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE)” 2023, and more specifically with Task 4a “Sound event detection of domestic activities”. This task involves the detection of 10 common events in domestic environments in 10 s sound clips. The events may have arbitrary duration in the 10 s clip. The main components of the methodology are data augmentation on mel-spectrograms that represent the sound clips, feature extraction by passing spectrograms through a frequency-dynamic convolution network with an extra attention module in sequence with each convolution, concatenation of these features with BEATs embeddings, and use of BiGRU for sequence modeling. Also, a mean teacher model is employed for leveraging unlabeled data. This research focuses on the effect of data augmentation techniques, of the feature extraction models, and on self-supervised learning. The main contribution is the proposed feature extraction model, which uses weighted attention on frequency in each convolution, combined in sequence with a local attention module adopted by computer vision. The proposed system features promising and robust performance.

Grigorios-Aris Cheimariotis, Nikolaos Mitianoudis

As the global population continues to age, the demand for effective and personalized care for elderly individuals becomes increasingly vital. Assisted living systems have emerged as promising technological solutions to address the challenges associated with aging populations. Numerous studies have demonstrated the positive impact of ambient assisted living (AAL) systems on seniors’ lives, including improved health outcomes, increased independence, enhanced quality of life and reduced healthcare costs. However, not all AAL systems are commercially available, primarily due to their energy-costly computational demands and expensive equipment. The paper at hand introduces an ambient low-cost solution exploiting only a single RGB sensor to monitor and recognize the activity of senior subjects. For this purpose, the utilization of bio-inspired networks, particularly the hierarchical temporal memory (HTM) model, based on Hebbian learning and implemented using the Nupic framework, as well as the spiking neural networks (SNNs) developed through the Nengo library is assessed. Both solutions are compared against the Support Vector Machine (SVM) classifier for elderly action recognition in the context of energy-efficient systems for AAL scenarios. The obtained results of the HTM showcase promising success rates and enhanced time efficiency, outperforming the rest methods in both terms. The above findings lead us to valuable insights for developing accurate and practical action recognition systems tailored to AAL scenarios

Katerina Maria Oikonomou, Ioannis Kansizoglou, Ioannis Tsampikos Papapetros, Antonios Gasteratos

The widespread proliferation of deep learning applications has triggered the need to accelerate them directly in hardware. General Matrix Multiplication (GEMM) kernels are elemental deep-learning constructs and they inherently map onto Systolic Arrays (SAs). SAs are regular structures that are well-suited for accelerating matrix multiplications. Typical SAs use a pipelined array of Processing Elements (PEs), which communicate with local connections and pre-orchestrated data movements. In this work, we show that the physical layout of SAs should be asymmetric to minimize wirelength and improve energy efficiency. The floorplan of the SA adjusts better to the asymmetric widths of the horizontal and vertical data buses and their switching activity profiles. It is demonstrated that such physically asymmetric SAs reduce interconnect power by 9.1% when executing state-of-the-art CNN layers, as compared to SAs of the same size but with a square (i.e., symmetric) layout. The savings in interconnect power translate, in turn, to 2.1% overall power savings.

Christodoulos Peltekis, Dionysios Filippas, Giorgos Dimitrakopoulos, Chrysostomos Nicopoulos

As the world’s population ages, there is an increasing demand for intelligent systems that can facilitate individuals’ independence while ensuring their safety. Assistive robotics has emerged as a viable tool in this context, providing personalized help and autonomy. This paper presents the design and implementation of the ASPiDA assisted living robot tailored to the fall detection challenge. By leveraging robust SLAM, human-aware navigation and fall detection algorithms, we aim to present a safe and effective, real-time and comprehensive robotic strategy to detect and act in fall detection events. A developed robotic platform is demonstrated and assessed in practical assisted living scenarios.

Katerina Maria Oikonomou, Ioannis Tsampikos Papapetros, Ioannis Kansizoglou, Georgios Ch. Sirakoulis, Antonios Gasteratos

Air pollution is the 4th most important factor of death globally and has a proven burden of disease while the link between covid-19 and poor air quality is proven beyond doubt. We expect sensors in buildings to grow in numbers to form Dense Indoor Sensor Networks (DISN) providing an unprecedent opportunity to analyse the performance, use and interaction with the building. However, off-the-self environmental sensors are rarely connected to a network; hence their potential to analyse the interaction with the building is not yet fully realised. This paper explores Long Range Wide Area Network (LoRaWAN), as an alternative for creating environmental WSNs and DISNs, that extends beyond its original long-distance communication purpose. Our approach is based on a real-world deployment with one gateway, one cloud-based network server and 13 heterogeneous sensor nodes measuring CO2, VOCs, PM1.0, PM2.0, PM2.5, temperature and humidity, luminocity, barometric pressure, PIR, which we evaluated empirically for a week. Moreover, we attempt a comparison with the IEEE 802.11 network protocol, which we analyse in a second, co-located development. The obtained results confirm that although LoRaWAN is feasible for minimal viable setups more research is needed to prove the feasibility of LoRaWAN-based DISNs.

Eleftheria Katsiri, Christos Karasoulas, Christoforos Keroglou

This study investigates the potential of using low-cost infrared cameras for non-contact monitoring of blood pressure (BP) in humans. Previous research has shown that robust contactless BP monitoring using RGB cameras is possible. In this study, the Eulerian Video Magnification (EVM) technique is employed to enhance minor variations in skin pixel intensity in the forehead and palm regions captured by an infrared camera. The primary focus of this study is to explore the possibility of using infrared cameras for contactless BP monitoring under low-light or night-time conditions. Results show that the proposed approach has surpassed the stringent accuracy standards set forth by the British Hypertension Society (BHS) and the Association for the Advancement of Medical Instrumentation (AAMI) protocol.

Thomas Stogiannopoulos, Nikolaos Mitianoudis

A high degree of joint synergy on the lower body is considered beneficial because it is linked to finer control of the center of mass. This is motivation to examine the stability through the prism of synergistic joint control. To identify the synergy index, the uncontrolled manifold (UCM) method can be used. This approach separates joint motion that translates to Cartesian motion from joint motion that is used for stabilization. However, research in joint synergy using the UCM staggers on the mathematical model that is used to describe the body. The human body is a non-linear multi-variable system, which makes it extremely difficult to express mathematically. As a result, certain assumptions and simplifications are needed. This has led to multiple approaches to calculate the UCM synergy index but their outcomes do not always translate to different scopes despite the theoretical background being valid. The core issue is that most of the mathematical frameworks used for joint modeling are too rigid to adapt to a different context. To address this, the product of exponentials (PoE) method to create a model of the lower body is proposed. This method allows each joint to be modeled independently as a vector and the body model can be created by simply organizing the vectors in the appropriate order. More importantly, the Jacobian matrix of the resulting model can be generated without additional differentiation. Since the UCM examines the rank and the null space of the Jacobian matrix of the system, this work is an important stepping stone towards UCM analyses for stability. Results show that the model of the right leg that was created was able to follow the hip joint center (JC) from motion capture therefore it was an accurate representation.

Dimitrios Menychtas, Athanasios Gkrekidis, Georgios Michailidis, Archontissa Kanavaki, Evgenia Kouli, Theodoros Tzelepis, Panagiotis Kasimatis, Konstantinos Astrapellos, Vassilios Gourgoulis, Ilias Smilios, Maria Michalopoulou, Eleni Douda, Georgios Ch. Syrakoulis, Nikolaos Aggelousis

The present study aimed to to map the views on the use of ambient sensors and robotic assistance for falls detection and prevention of a sample of older adults in Greece. Data were collected from 100 older adults as part of a wider observational study via an interview-administered questionnaire and were analysed using descriptive statistics and thematic analysis. Findings show that the vast majority of participants appeared to have positive views on the use of ambient sensors for falls prevention, whereas the views on robotic agents, although still positive in majority, were more equivocal. Positive views do not necessarily translate into intention to use assistive technologies. User needs assessment and tailored solutions are important steps for the adoption of these innovative technologies by older adults.

Archontissa Kanavaki, Maria Michalopoulou, Athanasios Gkrekidis, Evgenia Kouli, Dimitrios Menychtas, Helen Douda, Ilias Smilios, Vassilios Gourgoulis, Georgios Sirakoulis, Nikolaos Aggelousis

Automated monitoring systems can play a crucial role in the effective assistance of the elderly during their daily lives. Such systems aim to detect specific actions or activities of the individual which can indicate discomfort, pain or danger. Towards this end, Graph Convolutional Networks can be utilized to process skeleton data that represent the human body as a graph of interconnected notes. In this paper, a novel methodology for human action recognition is proposed that, along with the conventional full-body human graph, utilizes an arm-specific graph representation to better capture the fine-grained motion of the arms. For experimental evaluation, a new dataset is created that includes a variety of actions, performed by multiple elderly individuals, specifically for the purpose of monitoring everyday life activities. Additionally, experimentation is extended to two well-established large datasets, namely NTU RGB+D and NTU RGB+D 120, focusing on the actions that correspond to medical conditions. Experimental results demonstrate the effectiveness of our approach, compared to the conventional one, in terms of recognition accuracy.

Zois Tsakiris, Lazaros Tsochatzidis, Dimitrios Menychtas, Nikolaos Aggelousis, Ioannis Pratikakis

Older people face an increased risk of falling which can cause minor bruises, serious injury, or even death. This happens because as age progresses, the musculoskeletal system deteriorates, and maintaining balance becomes challenging. To counter that, exercise is recommended to help individuals maintain muscle mass. However, the effects of exercise have not been adequately quantified. In this work, 13 senior individuals participated in a group strengthening and balance program. They performed reactive balancing tasks before and after the training program and their stability was evaluated. The mean velocity norm of their center of mass was calculated to measure the impact of the exercise. Results show decreased mean velocity of the center of mass which implies increased stability. The effects are more pronounced during challenging tasks such as reacting to perturbations with their eyes closed. In conclusion, exercise had a quantifiable impact on the stability of the participants which should reduce their risk of experiencing falls in the future.

Dimitrios Menychtas, Evgenia Kouli, Athanasios Gkrekidis, Archontissa Kanavaki, Stefania Pavlidou, Elpiniki Nasiou, Evangelia Abrasi, Georgios Kontogiannis, Eleni Douda, Maria Michalopoulou, Ilias Smilios, Vassilios Gourgoulis, Georgios Ch. Syrakoulis, Nikolaos Aggelousis

This study presents the potential use of an alternative power source for low-power devices based on RF energy harvesting scheme, so as to solve the limitations of the conventional charging patterns. The main inspiration behind this work was to improve the technology of the obsolete healthcare monitoring and recording devices. To do so, an optimized RF energy harvesting and storage scheme is proposed (the so-called iPowerKit); its versatile design renders iPowerKit a flexible solution for the supply of a variety of low-power applications. Energy consumption analysis based on wearable devices is presented. Useful design orientations and a printed circuit board example are given.

Alexandros Boubaris, Nick Papanikolaou, Christos Christodoulou

In this paper, we propose a new technique for real-time detection of elderly people falling. The technique is a combination of two different features: the power burst curve (PBC) and the acceleration of target. A CW Radar was used to calculate the spectrum of the moving/falling elder using Short Time Fourier Transform (STFT). The long time series of raw radar is separated in to short enough time intervals so that the velocity could be considered approximately constant. This idea is similar to approximating a non-linear curve with a staircase. The window selection could be called ”time gating” or passing the long time-series through an orthogonal window.

Dimitrios Arnaoutoglou, Dimitris Dedemadis, Sotirios Katsimentes, Antigone Aikaterini Kyriakou, Athanasios Grekidis, Dimitrios Menychtas, Nikolaos Aggelousis, Georgios Ch. Sirakoulis, George Kyriacou

An estimation of the electric sources in the heart was conducted using a novel method, based on Huygens’ Principle, aiming at a direct estimation of equivalent bioelectric sources over the heart’s surface in real time. The main scope of this work was to establish a new, fast approach to the solution of the inverse electrocardiography problem. The study was based on recorded electrocardiograms (ECGs). Based on Huygens’ Principle, measurements obtained from the surface of a patient’s thorax were interpolated over the surface of the employed volume conductor model and considered as secondary Huygens’ sources. These sources, being non-zero only over the surface under study, were employed to determine the weighting factors of the eigenfunctions’ expansion, describing the generated voltage distribution over the whole conductor volume. With the availability of the potential distribution stemming from measurements, the electromagnetics reciprocity theorem is applied once again to yield the equivalent sources over the pericardium. The methodology is self-validated, since the surface potentials calculated from these equivalent sources are in very good agreement with ECG measurements. The ultimate aim of this effort is to create a tool providing the equivalent epicardial voltage or current sources in real time, i.e., during the ECG measurements with multiple electrodes.

Georgia Theodosiadou, Dimitrios G. Arnaoutoglou, Ioannis Nannis, Sotirios Katsimentes, Georgios Ch. Sirakoulis, George A. Kyriacou

Recent advances in Artificial Intelligence (AI) and Computer Vision (CV) have led to automated pose estimation algorithms using simple 2D videos. This has created the potential to perform kinematic measurements without the need for specialized, and often expensive, equipment. Even though there’s a growing body of literature on the development and validation of such algorithms for practical use, they haven’t been adopted by health professionals. As a result, manual video annotation tools remain pretty common. Part of the reason is that the pose estimation modules can be erratic, producing errors that are difficult to rectify. Because of that, health professionals prefer the use of tried and true methods despite the time and cost savings pose estimation can offer. In this work, the gait cycle of a sample of the elderly population on a split-belt treadmill is examined. The Openpose (OP) and Mediapipe (MP) AI pose estimation algorithms are compared to joint kinematics from a marker-based 3D motion capture system (Vicon), as well as from a video annotation tool designed for biomechanics (Kinovea). Bland-Altman (B-A) graphs and Statistical Parametric Mapping (SPM) are used to identify regions of statistically significant difference. Results showed that pose estimation can achieve motion tracking comparable to marker-based systems but struggle to identify joints that exhibit small, but crucial motion. Joints such as the ankle, can suffer from misidentification of their anatomical landmarks. Manual tools don’t have that problem, but the user will introduce a static offset across the measurements. It is proposed that an AI-powered video annotation tool that allows the user to correct errors would bring the benefits of pose estimation to professionals at a low cost.

Dimitrios Menychtas, Nikolaos Petrou, Ioannis Kansizoglou, Erasmia Giannakou, Athanasios Grekidis, Antonios Gasteratos, Vassilios Gourgoulis, Eleni Douda, Ilias Smilios, Maria Michalopoulou, Georgios Ch. Sirakoulis, Nikos Aggelousis

As more energy-intensive electrical appliances and active assets appear in the residential sector, electricity end-users are able to provide various demand response services, aiming to balance demand and supply profiles. This paper proposes a state-of-the-art energy management system for a microgrid. The study focuses on a four-apartment building incorporating heat pumps, an energy storage system, and photovoltaic generation. A smart control mechanism based on reinforcement learning is introduced aiming to improve thermal comfort, minimize energy consumption and ensure minimum degradation of the storage system. The reported results offer insights into the optimal residential management practices and evaluate the performance of the proposed control strategy in comparison to other alternative demand response solutions.

Christos L. Athanasiadis, Kalliopi D. Pippi, Theofilos A. Papadopoulos, Christos Korkas, Christos Tsaknakis, Vasiliki Alexopoulou, Vasileios Nikolaidis, Elias Kosmatopoulos

The increasing numbers of Electric Vehicles (EVs), require further installations of charging stations. The challenge of managing these grid-connected charging stations leads to a multi-objective optimal control problem where station profitability, user preferences, grid requirements and stability should be optimized. However, it is very challenging to determine the optimal charging/discharging EV schedule, since the controller should exploit the fluctuations in the electricity prices, the available renewable resources, the available stored energy of other vehicles and of course, cope with the uncertainty of EV arrival/departure scheduling. In addition, the growing number of connected vehicles results in complex state and action vectors, making it difficult for centralized and single-agent controllers to handle the problem. In this paper, we propose a novel multi-agent and distributed Reinforcement Learning (MARL) framework, that tackles the challenges mentioned above, producing controllers that achieve high performance levels under diverse conditions. In the proposed distributed framework, each charging spot makes its own charging/discharging decisions towards a cumulative cost reduction without sharing any type of private information addressing the problem of cost minimization and user satisfaction. The framework significantly improves the scalability and sample efficiency of the underlying Deep Deterministic Policy Gradient (DDPG) algorithm. Extensive numerical studies and simulations demonstrate the effectiveness of the proposed approach compared against Rule-Based Controllers and well-established state-of-the-art centralized RL algorithms offering performance improvement up to 25% and 20% respectively.

Christos D. Korkas, Christos Tsaknakis, Athanasios Ch. Kapoutsis, Elias Kosmatopoulos

Nowadays, the volatile economic conditions alongside the environmental concerns on using fossil fuels for electricity, necessitate the gradual transition from conventional power plants to distributed photovoltaic (PV) systems. To further promote PV installations, various support mechanisms have been introduced; net-metering (NEM) is considered as one of the most important alternating pricing policies. Nevertheless, the ever-increasing penetration of PVs and especially their intermittent nature may lead to unprecedented technical issues related to the reliable grid operation. To tackle these challenges, battery energy storage (BES) systems are used on the premises of PV prosumers, but their relatively high capital costs and limited lifetime hinder such investments. In this paper, a techno-economic model is introduced to evaluate the viability of prospective PV-BES investments of medium-voltage (MV) prosumers operating under the NEM mechanism by taking also into account the capacity degradation of BES. The focus of this paper is the analysis of the NEM program of MV prosumers in Greece. In particular, the case of the university campuses of the Democritus University of Thrace, Greece, is considered, but the methodology is applicable for evaluating other NEM programs and electricity sectors. The impact of various parameters on the viability of the NEM investments is investigated considering different scenarios. Results show that NEM is generally a profitable mechanism, especially, for prosumers with high load demand. Significant parameters that affect NEM profitability are the load demand growth, the PV and BES degradation, the inflation and the discount rate.

Kalliopi Pippi, Theofilos Papadopoulos, Georgios Kryonidis, Evangelos Kyriakopoulos

Distributed Machine learning has delivered considerable advances in training neural networks by leveraging parallel processing, scalability, and fault tolerance to accelerate the process and improve model performance. However, training of large-size models has exhibited numerous challenges, due to the gradient dependence that conventional approaches integrate. To improve the training efficiency of such models, gradient-free distributed methodologies have emerged fostering the gradient-independent parallel processing and efficient utilization of resources across multiple devices or nodes. However, such approaches, are usually restricted to specific applications, due to their conceptual limitations: computational and communicational requirements between partitions, limited partitioning solely into layers, limited sequential learning between the different layers, as well as training a potential model in solely synchronous mode. In this paper, we propose and evaluate, the Neuro-Distributed Cognitive Adaptive Optimization (ND-CAO) methodology, a novel gradient-free algorithm that enables the efficient distributed training of arbitrary types of neural networks, in both synchronous and asynchronous manner. Contrary to the majority of existing methodologies, ND-CAO is applicable to any possible splitting of a potential neural network, into blocks (partitions), with each of the blocks allowed to update its parameters fully asynchronously and independently of the rest of the blocks. Most importantly, no data exchange is required between the different blocks during training with the only information each block requires is the global performance of the model. Convergence of ND-CAO is mathematically established for generic neural network architectures, independently of the particular choices made, while four comprehensive experimental cases, considering different model architectures and image classification tasks, validate the algorithms’ robustness and effectiveness in both synchronous and asynchronous training modes. Moreover, by conducting a thorough comparison between synchronous and asynchronous ND-CAO training, the algorithm is identified as an efficient scheme to train neural networks in a novel gradient-independent, distributed, and asynchronous manner, delivering similar – or even improved results in Loss and Accuracy measures.

Panagiotis Michailidis, Iakovos T. Michailidis, Sokratis Gkelios, Georgios Karatzinis, Elias B. Kosmatopoulos

One of the most prominent examples of a complex system in nature is the nervous system, which exhibits oscillation phenomena across its structures, from single neurons to sophisticated neural networks. Memristors have been utilized in the past decade as a promising technology for building neuromorphic systems due to their intrinsic neuromorphic properties. In this paper, we introduce a novel Memristor-based Oscillator (MBO) circuit design that implements both artificial neurons and artificial synapses in the same MBO-based medium, addressing scalability issues. The circuit design is transistor-free and simple, allowing high integration density. We utilize passive unipolar memristor devices based on the JART memristor model to design MBO neurons, which have been shown to reproduce bio-plausible spiking and bursting activities. Moreover, the MBO circuit is also realized as an artificial synapse incorporating synapse and axon mechanisms. Finally, the MBO neurons are modeled as Parkinson-related neuron types and their direct bidirectional coupling demonstrates that MBO neurons are effective in driving spiking activity in non-externally stimulated MBO neurons. This qualifies them for implementation in brain-inspired networks, providing low-cost Neurological-disease-related emulators.

Ioannis Chatzipaschalis, Evangelos Tsipas, Karolos-Alexandros Tsakalos, Antonio Rubio, Georgios Sirakoulis

Approximate computation has evolved recently as a viable alternative for maximizing energy efficiency. One aspect of approximate computing involves the design of hardware units that return a sufficiently accurate result for the examined occasion, rather than computing an accurate result. As long as the hardware units are allowed to compute approximately, they can be designed with multiple new ways. In this work, we focus on the synthesis of approximate parallel-prefix adders. Instead of exploring specific architectures, as done by state-of-the-art approaches, the introduced synthesizer can produce every solution that meets the designer’s criteria, resulting in adders with various delay, area, and error tradeoffs. This automatic design space exploration allows approaching, in several cases, optimal solutions that could have not been designed with any other known parallel-prefix architecture. The synthesized adders, when compared with state-of-the-art adders, achieve 27%–36% better error frequency (EF) on average for random inputs and improve image quality metrics by 8%–42% for image filtering. These results are achieved with the proposed adders requiring the same or marginally more hardware area or energy. On the contrary, in split-accuracy configurations, more than 30% of hardware area/energy can be saved for the same classification accuracy for a neural network application.

Apostolos Stefanidis, Ioanna Zoumpoulidou, Dionysios Filippas, Giorgos Dimitrakopoulos, Georgios Ch. Sirakoulis

Exercise in older adults has important benefits, including maintenance of physical function and falls prevention, and is recommended at least twice per week. Older adults in rural areas in Greece may be unfamiliar with exercise regimes and are unlikely to have access to relevant information and services. As part of the ASPIDA project for falls prevention, a brief exercise and educational intervention took place in a village of Rodopi, Greece. The program consisted of 6-week x 1hour sessions of supervised group exercise. Participants also received brief education on the importance of exercise for falls prevention and were encouraged to set personal goals and perform the exercises at home. Older women were invited to participate via local social services and word of mouth. Dynamic balance (Timed-Up-and-Go) and lower limb strength (30sec chair-stand) were assessed before and after the intervention. Participants also completed a questionnaire on current physical activity (PA) before the program. A total of 16 older women ≥65 years, mean age 71.81±7, 87.5% married, 87.5% cohabiting, were assessed before and after the intervention. Based on self-reported PA, before prior to the program only 14.2% engaged in ≥1 hours/week of structured physical exercise, whereas 57.1% did not engage in any. Also, 35.7% reported walking ≥3 hours/week and another 35,7% <1hour/week. Lower limb strength showed a significant improvement, before (M=13.25)-after (M=14.56), t(15)=-2,23, p=.042. There was no statistically significant difference in dynamic balance. However, there were improvements (-0.53sec) which have clinical significance based on the normative test values, i.e. +0.3 to +0.4sec average change per 5 years in these age groups. Most participants reported incorporating some exercises in their daily routine. All were satisfied with the program and would be keen to continue if they had the opportunity. Findings show that even a low-resource intervention of short duration and frequency can have positive effects on older women’s strength and balance, raise awareness and motivate them to adopt exercise routines. Such programs are feasible with the collaboration of local services and have the potential to decrease risk of falls. Follow-up assessments will determine the extent of maintenance of exercise behaviour.

Kanavaki, A.M., Michalopoulou, M., Iliopoulos, S., Irakleous, E., Nasiou, E., Menychtas, D., Kouli, E., Gkrekidis, A., Douda, H.T., Smilios, I., Gourgoulis, V., Syrakoulis, G., Aggelousis, N.

WHO’s Global Action Plan on Physical Activity policy recommendations include incorporating physical activity (PA) promotion in health and social services and prioritising programs for the least active, such as older adults. Easy-to-use and valid PA assessment tools are needed in this process. The General Practice Physical Activity Questionnaire (GPPAQ) is used in primary care in the UK to identify insufficiently active individuals 16-74 years, who may benefit from a brief PA promotion intervention. The GPPAQ classifies individuals as Active based on their occupational activity, physical exercise and cycling, but does not include walking in score calculation. It has not been validated for adults over 75 years. The present study aimed to assess the validity of GPPAQ and a modified version, GPPAQ-Old, which includes walking in score calculation, in a Greek sample of adults 65-90 years. The tool was translated in Greek using a standard procedure. GPPAQ and GPPAQ-Old Active/ non-Active classification was compared against meeting/not meeting the guidelines of ≥150min/week of moderate-to-vigorous PA (MVPA), accelerometer-measured. Sensitivity and specificity were calculated using a 2×2 table with accelerometer-measured MVPA as criterion. A total of 97 retired adults, mean age 73.67 ± 5.90, 82.7% women, from urban and rural areas in municipality of Rodopi, Greece, wore Actigraph accelerometers for a week and subsequently completed the GPPAQ. Based on accelerometry, 88.7% participants met the MVPA guidelines. 4.1% were classified as Active by GPPAQ and 62.9% by GPPAQ-Old. Sensitivity was 4.5% and specificity 100% for GPPAQ, 62.5% and 33.3% for GPPAQ-Old. GPPAQ showed very poor sensitivity and excellent specificity, whereas GPPAQ-Old showed good sensitivity, but poor specificity. Similar results are common when self-report PA and accelerometry are compared. GPPAQ-Old can correctly identify 6 in 10 adults as sufficiently Active, but may miss 2 in 3 insufficiently Active. It also offers insight on PA types older adults engage in their daily life. Results may be influenced by the highly active sample (only 11.3% din not meet MVPA guidelines).

A.M. Kanavaki, M. Michalopoulou, V. Gourgoulis, Ε. Kouli, Α. Gkrekidis, Η.Τ. Douda, Ι. Smilios, G. Syrakoulis, Ν., Aggelousis

The elderly population is prone to falls, which can have severe consequences, including hospitalization or mortality. Numerous factors can contribute to falls in older adults, including health conditions, medication use, environmental hazards, poor nutrition, and insufficient physical activity. Specifically, insufficient exercise and physical activity can lead to reduced flexibility, weakness, and poor balance, which are significant risk factors for falls. To prevent falls in older adults, it is crucial to recognize and manage these factors. The aim of this study was to compare the functional fitness levels of older adults who were at risk of falling with those who were not. The study recruited 190 independent adults aged 65 years or older and determined their fall risk status using a fall risk assessment scale. Thirty-three of them were identified as being at risk of falling. The Senior Fitness Test, which included six tests for strength, endurance, flexibility, and balance, was used to assess the participants’ functional fitness. The results revealed significant differences in functional fitness between the two groups, with those at risk of falling scoring lower, particularly in agility, dynamic balance, and aerobic endurance tests. The researchers identified the Timed up and go (TUG) test and the 2-minute step test as the most effective predictors of fall risk. The study concludes that targeted interventions to enhance agility and endurance may benefit older adults who are at risk of falling due to their lower functional fitness levels.

F. Papanikolaou, E. Kouli, A. Gkrekidis, A. Kanavaki, P. Manaveli, D. Menychtas, E. Douda, I. Smilios, Μ. Michalopoulou, V. Gourgoulis, G. Sirakoulis, N. Aggelousis

Falls can have serious consequences for elderly and stroke patients, and understanding the factors that contribute to falls is crucial for developing effective prevention strategies. Visual conflict is one such factor that may cause disruptions in static balance. The analysis of the center of pressure (COP) trajectory can help evaluate this factor. This study aimed to examine the differences in static balance metrics concerning visual conflict among fall-prone older adults residing in the community and chronic stroke survivors. The study included a total sample of 16 elderly fallers and 16 chronic stroke survivors. All subjects performed a static balance test which involved maintaining stability for 30 seconds while wearing a helmet in the shape of a dome. The dome-shaped helmet provided conflicting feedback to the brain as the eyes detected a constant distance from the surface of the dome, indicating a stable head position in relation to the environment, while the vestibular system accurately perceived any head movements. Data on Ground Reaction Force (GRF) were collected from both groups, and an open-access code was used to compute the Center of Pressure (COP) features. The COP trajectory was analyzed during quiet stance with feedback conflict, utilizing over 50 different equations. Independent t-tests were performed, revealing that six parameters exhibited significant differences between the elderly fallers and the stroke survivors. The findings of this research are consistent with prior studies that emphasized the significance of identifying criteria for categorizing people who are vulnerable and tracking their development over time. With the latest progress in data analysis, the integration of COP trajectory examination into standard medical examinations is becoming increasingly viable, presenting the opportunity for customized treatment plans and better rehabilitation consequences.

M. Karageorgopoulou, D. Menychtas, A. Grekidis, P. Manaveli, E. Kouli, A. Kanavaki, E. Douda, C. Kokkotis, M. Michalopoulou, V. Gourgoulis, I. Smilios, G. Sirakoulis, N. Aggelousis

Falls are a common problem among elderly and can lead to serious consequences. The consequences of falls can be severe, especially in older adults, as they can lead to fractures, head injuries, and other complications that can result in disability or even death. Balance during daily activities is a complex process that requires coordination between the limbs and trunk. Vector coding is an important technique for examining coordination in balance, as it allows for the measurement of the magnitude and direction of movement in different body segments. The aim of this study is to examine differences in femur-thorax coordination patterns in the 3-axis on tandem static balance in elderly fallers and non-fallers. 11 elderly fallers, participants who had at least one fall this year, and 11 elderly non-fallers aged 65-90y were employed in this study. Kinematic data were collected from fallers and non-fallers at the Biomechanics Lab – DUTH in Greece. Vector coding technique was used for calculations of coupling angles and mean coupling angles for femur and thorax segment angles in sagittal, frontal and transverse planes. Then, the relative frequencies for four coordination patterns were calculated. Independent t-test was used for our analysis in order to investigate the relative frequencies in the four coordination patterns between the fallers and the non-fallers. According to the results, individuals who did not experience falls tended to use their femur more than their thorax in anti-phase movement within the sagittal plane. In contrast, those who did fall used their thorax more than their femur in anti-phase movement within the frontal plane. These findings highlight the usefulness of vector coding as a quantitative tool for analyzing movement patterns. By identifying specific patterns of coordination associated with falls, this technique can contribute to the development of effective strategies for fall prevention.

C. Kokkotis, D. Menychtas, P. Manaveli, A. Gkrekidis, E. Kouli, A. Kanavaki, E. Douda, V. Gourgoulis, I. Smilios, M. Michalopoulou, G. Sirakoulis, N. Aggelousis

Falls are the leading cause of injury-related deaths and hospitalizations among the elderly, highlighting the need for effective falls prevention strategies. These consequences can significantly impact the well-being of the elderly and their families. The certain study aims to provide an explainable approach that could identify high-risk individuals and risk factors related to falls. Data were recorded from adults 55-90 years of any gender in the province of East Macedonia and Thrace in Greece. This study considered multidisciplinary data gathered from interview-administered questionnaires and physical performance tests among the elderly. A total of 208 variables were considered, with the main indicator for assessing falling status being a combination of the John Hopkins Fall Risk Assessment tool score and the number of falls during an exercise program. Using binary classification, the study aimed to predict high-risk fallers and non-fallers. Specifically, the first class consists of 29 subjects who were classified as high-risk fallers, while the second class consists of 41 subjects who were classified as non-fallers. The proposed approach combines a wrapper feature selection algorithm with three well-known machine learning (ML) models. Through this approach, a concise subset of only 8 risk factors was identified, achieving an accuracy rate of 95.24%. To further analyze the significance of these risk factors, the study used SHapley additive explanations (SHAP). Overall, the findings of this study have the potential to assist in the development of effective risk stratification strategies and the identification of risk profiles of each individual who falls, ultimately enabling appropriate interventions to be implemented.

C. Kokkotis, A. Kanavaki, E. Kouli, A. Gkrekidis, D. Menychtas, P. Manaveli, M. Michalopoulou, E. Douda, V. Gourgoulis, I. Smilios, G. Sirakoulis, N. Aggelousis

Biomechanical motion analysis approaches, such as OpenSim, are useful tools for simulating movement and estimating muscle forces and activations.In order to develop prevention strategies for elderly falls, it is important to study the mechanisms behind them, which can be achieved through motion analysis.. While OpenSim has been validated for young people, it is important to check the validity of the calculated muscle activations for elderly subjects. Therefore, this study aimed to compare the muscle activations evaluated through OpenSim’s Static Optimization tool with their EMG recordings obtained during the gait of elderly individuals, using the Cosine Similarity approach. Two healthy male elderly subjects were asked to walk at their normal pace for a minute on a split belt treadmill, while data was collected using a 3D motion capture system, two force platforms and four surface EMG sensors securely placed at the lateral gastrocnemius (LG), soleus (S), tibialis anterior (TA), and peroneus longus (PL) muscles of the left leg. Motion data collected, for ten gait cycles of each subject’s left limb, was used for OpenSim calculations with a generic full body musculoskeletal model. Finally, OpenSim Static Optimization analyses were performed to determine the muscle activations of the left limb for each gait cycle. The results showed that the calculated muscle activations had low similarity with the recorded activations, which could be due to the time shift difference between the recorded and calculated data. Therefore, further research with a larger group of participants and a wider range of muscles is needed to validate the use of OpenSim for elderly subjects.

A. Gkrekidis, G. Giarmatzis, D. Menychtas, V. Karakasis, A. Kanavaki, E. Kouli, I. Smilios, V. Gourgoulis, M. Michalopoulou, E. Douda, G. Sirakoulis, N. Aggelousis

Falls are a major cause of injury in elderly individuals, with some cases resulting in long hospital stays or even fatalities. Muscle weakness is a significant risk factor for falls, and Musculoskeletal (MSK) modeling, such as OpenSim, is a valuable tool that can provide data on muscle activations and forces during gait through movement simulations. The aim of this study was to compare ankle joint muscle force generation patterns during gait, as calculated by OpenSim, between elderly fallers and non-fallers. The study involved six elderly participants, divided into fallers (n=3) and non-fallers (n=3) groups based on their score on the John Hopkins fall risk assessment tool. The participants were asked to walk at their normal pace for one minute on a split-belt treadmill, and their gait cycles were recorded using a 3D motion capture system. The recorded data was then used to determine muscle activations and forces of the left limb for each gait cycle using a generic full body MSK model in OpenSim. Statistical Parametric Mapping (SPM) was used to compare the time curves of the ankle joint plantar and dorsi flexors normalized to weight forces between the two groups. The results showed significant differences in muscle force production during gait between fallers and non-fallers, with non-fallers demonstrating higher muscle force values. Fallers showed lower muscle force values in the plantar flexors, which suggests weaker plantar flexors, a known risk factor for falls. This finding highlights the importance of identifying plantar flexor weakness in the elderly as a preventive measure for falls.

A. Gkrekidis, G. Giarmatzis, D. Menychtas, P. Manaveli, E. Kouli, A. M. Kanavaki, V. Gourgoulis, E. Douda, I. Smilios, M. Michalopoulou, G. Sirakoulis, N. Aggelousis

According to research, well-being is a key factor in determining one’s mental health and tends to decrease gradually over time. As the prevalence of depression increases among the elderly population, this metric becomes increasingly significant. Although short-term physical activity and exercise initiatives have been shown to enhance well-being, there is limited research on whether there is a distinction in well-being outcomes between those vulnerable to falls and those who are not. This study aims to examine the impact of a single exercise session on the well-being of older adults and to explore any differences between fallers and non-fallers. The study involved 41 participants from open protection centers for the elderly who anonymously completed the Profile of Mood States (POMS) questionnaire before and after one training session. The results showed a statistically significant interaction between measurement and group regarding anxiety, confusion, and overall well-being, with non-fallers experiencing positive changes following their involvement in the exercise program. Moreover, statistically significant differences were observed between measurements in the energy, aggressiveness, and depression measures, indicating improvements for all participants post-program. However, no significant differences were detected regarding fatigue levels. These findings suggest that a single exercise session can enhance the mental well-being of elderly individuals, including those at risk of falls and those who are not.

E. Kouli, F. Papanikolaou, K. Anagnostopoulos, D. Menychtas, A. Kanavaki, A. Grekidis, M. Michalopoulou, I. Smilios, E. Douda, V. Gourgoulis, G. Sirakoulis, N. Aggelousis

Stroke is one of the most ordinary medical conditions that affect the function of the brain and occurs when the supply of blood delivered to the brain is significantly altered. This results in a wide range of motor limitations which is also associated with aging. The purpose of the present study was to compare the physical and functional capacity between chronic stroke survivors and elderly fallers. The study had 24 participants (N=24), who were divided into two groups: a group of chronic stroke survivors (n=12) and a group of elderly fallers (n=12). The average age of the stroke survivors’ group was 63.8 (± 8.4), while the average age of the elderly fallers group was 75.2 (±8.4). Standardized functional fitness tests assessing balance, strength, endurance, agility and flexibility were performed using Senior Fitness Test. The results of the data analyses revealed that there was no statistically significant effect of the “group” factor on the Chair Stand Step, Back Scratch Test with left hand on the top, Back Scratch Test with the right hand on the top, Chair Sit and Reach Test with left foot extended, Chair Sit and Reach Test with right foot extended, Arm Curl Test with the left hand, 2.45-m Up-and-Go Test and, 2-minute Step Test. However, a significant difference was found in the Arm Curl Test with the right hand, which was the paretic arm for stroke survivors. According to the findings mentioned above, physical fitness does not distinguish between chronic stroke patients and aged matched elderly fallers. Overall, physical fitness is influenced by various factors, and stroke survivors and elderly fallers may share some common factors that affect their physical fitness levels.

K. Anagnostopoulos, E. Kouli, A. Kanavaki, S. Iliopoulos, A. Gkrekidis, D. Menychtas, I. Smilios, E. Douda, M. Michalopoulou, V. Gourgoulis, G. Sirakoulis, N. Aggelousis

The possibility of falling increases dramatically as a person ages. In the elderly population, falling is prevalent causing serious injuries, psychological stress, loss of independence, and even death. However, the risk of falling does not increase equally for all people as they age. Intrinsic factors such as the ability to process visual information affect the balancing strategy. In general, individuals with a high risk of falls will perform differently when performing balancing tasks under different visual conditions. To understand the impact of visual feedback, two groups of elders, one without risk of fall and another with high risk were recorded using three different visual conditions, eyes open, closed eyes, and wearing a dome-shaped helmet. A marker-based system was used to capture the movements of the performers, and their performance was recorded with the help of ten infrared cameras. In addition, two force plates were utilized to record their ground reaction forces. The mean velocity and the sway area of their center of pressure (CoP) were evaluated and compared to quantify their posture. The mean velocity appears to be a more sensitive metric than the sway area. This makes the mean velocity of the CoP a better metric to be used for balance assessment. However, it is not a good metric to distinguish between individuals with different risks of falling.

D. Menychtas, N. Petrou, A. Gkrekidis, E. Kouli, A. Kanavaki, E. Giannakou, V. Gourgoulis, M. Michalopoulou, I. Smilios, E. Douda, G. Sirakoulis, N. Aggelousis

Convolutional Neural Networks (CNNs) are the state-of-the-art solution for many deep learning applications. For maximum scalability, their computation should combine high performance and energy efficiency. In practice, the convolutions of each CNN layer are mapped to a matrix multiplication that includes all input features and kernels of each layer and is computed using a systolic array. In this work, we focus on the design of a systolic array with configurable pipeline with the goal to select an optimal pipeline configuration for each CNN layer. The proposed systolic array, called ArrayFlex, can operate in normal, or in shallow pipeline mode, thus balancing the execution time in cycles and the operating clock frequency. By selecting the appropriate pipeline configuration per CNN layer, ArrayFlex reduces the inference latency of state-of-the-art CNNs by 11 %, on average, as compared to a traditional fixed-pipeline systolic array. Most importantly, this result is achieved while using 13 %-23 % less power, for the same applications, thus offering a combined energy-delay-product efficiency between 1.4× and 1.8× .

C. Peltekis, D. Filippas, G. Dimitrakopoulos, C. Nicopoulos, D. Pnevmatikatos

Physical activity guidelines for older adults recommend a minimum of 150minutes of moderate-to-vigorous physical activity (MVPA)/week. The study examined well-established psychological exercise determinants (i.e. motivation, self-efficacy, outcome expectations) as potential predictors of accelerometer-assessed total MVPA. One hundred and twenty retired older adults were recruited from community centers in urban and rural areas of Rodopi, Greece. Participants completed questionnaires on psychosocial PA determinants (Behavioural Regulation for Exercise Scale, Barriers-to-exercise Efficacy Scale, Multidimensional Outcome Expectations for Exercise Scale). Multiple regression analysis tested in separate models if Autonomous Motivation (AM), Controlled Motivation (CM), Self-efficacy (SE) or Outcome Expectations (OE) predicted MVPA, controlling for age and sex. Participants spent on average 63.5min/day in MVPA and 91.2% met the MVPA guidelines. Only the AM model was significant. Age, sex and AM explained 19.1% of MVPA variance (R2=.44, F(3,116)=9.11, p<.001). AM significantly predicted MVPA (β=.20, p=.021), as did age (β=-.30, p<.001) and sex (β=-.16, p=.035). Study participants met the MVPA guidelines in vast majority and engaged in a high amount of MVPA daily. AM was a significant MVPA predictor, whereas CM, SE and OE were not. Further inspection of participant responses revealed that a) although participants mostly reported walking as the «exercise» they engaged in, some responded to the questionnaires having group-exercise in mind; b) in SE there was contrasting directionality in the association of individual scale items with MVPA, c) there was a ceiling effect in OE. MVPA seems to be embedded in housework that most participants engaged in, which is commonly considered light PA.

Kanavaki, A.M., Michalopoulou, M., Apostolidou, M., Gkrekidis, A., Kouli, E., Douda, H., Smillios, I., Gourgoulis, V., Syrakoulis, G., Aggelousis, N.

This work explores the use of infrared low-cost cameras for monitoring peripheral oxygen saturation (SpO₂), a vital sign that is particularly important for individuals with fragile health, such as the elderly. The development of contactless SpO₂ monitoring utilizing RGB cameras has already proven successful. This study utilizes the Eulerian Video Magnification (EVM) technique to enhance minor variations in skin pixel intensity in particular facial regions. More specifically, the emphasis in this study is in the utilization of infrared cameras, in order to explore the possibility of contactless SpO₂ monitoring under low-light or night-time conditions. Many different methods were employed for regression. A study of machine learning regression methods was performed, including a Generalized Additive Model (GAM) and an Extra Trees Regressor, based on 12 novel features extracted from the extracted amplified photoplethysmography (PPG) signal. Deep learning methods were also explored, including a 3D Convolution Neural Network (CNN) and a Video Vision Transformer (ViViT) architecture on the amplified forehead/cheeks video. The estimated SpO₂ values of the best performing method reach a low root mean squared error of 1.331 and an R^₂ score of 0.465 that fall within the acceptable range for these applications.

Thomas Stogiannopoulos, Grigorios-Aris Cheimariotis, Nikolaos Mitianoudis

The increase in the ageing of the population calls for advanced monitoring and actuation systems for the smart control and management of elderly households that should manage to optimise multi-objective control problems. This study proposes a state-of-art approach, based on Approximate Dynamic Programming, that tackles multiple challenges, aiming to maximise the energy efficiency, the electricity cost reduction, the user comfort, and the load monitoring of a typical Greek residency (simulated in EnergyPlus), including both controllable and uncontrollable loads, an energy storage system, and power generation units. The article presents the various simulated loads, pricing schemes under Greek regulations and variety of control strategies that are co-ordinately used to minimise the defined objective functions in contrast to the current state-of-the-art approaches. The latter either focus on single objective problems or use simpler iterative and often limited to specific variables, approaches for multi-objective control. Finally, the presented results provide insights for the optimal and safe residential management and validate the performance of the proposed control strategy in comparison with alternative demand response strategies, offering improved performance by 20% in terms of energy efficiency, cost reduction, and thermal comfort.

Christos Tsaknakis, Christos Korkas, Kalliopi Pippi, Christos Athanasiadis, Vasiliki Alexopoulou, Elias Kosmatopoulos, Vassilis C. Nikolaidis, Theofilos Papadopoulos

Navigating in a perpetually changing world can provide the basis for numerous challenging autonomous robotic applications. With a view to long-term autonomy, visual place recognition (vPR) systems should be able to robustly operate under extreme appearance changes in their environment. Typically, the utilized data representations are heavily influenced by those changes, negatively affecting the vPR performance. In this article, we propose a sequence-based technique that decouples such changes from the similarity estimation procedure. This is achieved by remapping the sequential representation data into the distance-space domain, i.e., a domain in which we solely consider the distances between image instances, and subsequently normalize them. In such a way, perturbations related to different environmental conditions and embedded into the original representation vectors are avoided, therefore the scene recognition efficacy is enhanced. We evaluate our framework under multiple different instances, with results indicating a significant performance improvement over other approaches.

Ioannis Tsampikos Papapetros, Ioannis Kansizoglou, Loukas Bampis, Antonios Gasteratos

Physical activity (PA) in older adults is essential for healthy aging. A minimum of 150 minutes of moderate-to-vigorous PA (MVPA) and at least twice per week PA for strength, balance and flexibility is recommended for benefits in health and well-being. Given the limited available data, the present study reports on PA levels in Greek older adults, assessed by accelerometers in a sample of adults 60-90 years. This was part of ASPiDA, a multidisciplinary project on physical activity, falls and quality of life in older adults. Participants were invited to participate via KAPI, social services and word of mouth. Actigraph accelerometers were used to assess PA for a week. Participants self-reported participation in structured exercise. Time spent in PA of various intensities (light, MVPA, sedentary behaviours (SB)) was calculated by applying age-specific cut-points to accelerometer data. Descriptive statistics and ANOVA were applied for statistical analysis. A total of 134 retired individuals, mean age 72.0±6.01 had valid accelerometer data, 83.9% women. Participants spent on average 64±33min per day in MVPA, 134±37min in light PA and 640±14min in SB. 91.4% participants met aerobic PA guidelines. 71.6% reported doing no physical exercise and only 14.8% >1hour per week. There was a significant effect of age group (61-70, 71-80, 81-90) on MVPA, but not on light PA and SB. Although MVPA levels decrease by 47% from the youngest to oldest age group, the vast majority of older adults were sufficiently active. Exercise participation in older adults should be promoted as part of public health policies.

Kanavaki A.M., Michalopoulou M., Pavlidou S., Kouli E., Gkrekidis A., Menychtas D., Smilios I., Douda H., Gourgoulis V., Aggelousis N.

Peripheral oxygen saturation (SpO₂) is one important vital sign to be monitored in individuals, whose health is fragile, such as the elderly. Contactless SpO₂ monitoring using RGB cameras has been already developed with satisfactory results. This work explores the case of achieving an acceptable level of performance, when the lightning conditions are not optimal, particularly during night time, by processing solely infrared low-cost camera recordings. The Eulerian Video Magnification (EVM) technique was used to enhance the subtle differences in skin pixel intensity in the facial area. Two approaches were explored for performing regression: one using 12 novel features extracted from the amplified photoplethysmography (PPG) signal and Generalized Additive Models and a second using a 3D Convolution Neural Network (CNN) architecture on the raw amplified forehead video. The root mean square error in the estimated SpO₂ levels using both methods is minimal and in the accepted range for these applications.

Thomas Stogiannopoulos, Grigorios-Aris Cheimariotis, Nikolaos Mitianoudis

The human brain can be considered as one of the most complex systems found in nature and its underlying physical and biological mechanisms have attracted an increasing research interest especially in reference to modeling of pathogenic neurological conditions. In particular, one of the most well-known paradigms of such pathological neurological conditions that occasionally appear to stress hard the mental and physical condition of elderly people is the Parkinson’s disease. Despite the recent progress of the aforementioned disease modeling as well as its tentative treatment by usage of software techniques like the Deep Brain Stimulation (DBS) approach, there is a even growing demand for real-time emulations of such neuromorphic conditions to handle more efficiently such situations. This work proposes a hardware accelerator employing Asynchronous Cellular Neural Networks (ACNNs) in Field Programmable Gate Arrays (FPGAs) to boost further Parkinson’s disease studies. Our results suggest that the proposed hardware design provides sufficient neuromorphic plausibility by integrating Synapse and Axon models. As a proof-of-concept, an implementation of the proposed model on Virtex-7 XC7VX330T FPGA has successfully simulated up to 50 Izhikevich neurons in real time with low-power consumption aiming to Parkinson’s Disease Treatment emulation by employing the DBS approach.

Ioannis Chatzipaschalis, Karolos-Alexandros Tsakalos, Georgios Sirakoulis, Antonio Rubio

Εισαγωγή. Οι πτώσεις στην τρίτη ηλικία (ΑΤΗ) είναι ένας σημαντικός αιτιολογικός παράγοντας οστεοπορωτικών καταγμάτων. Για την πρόληψη τους, συστήνονται η τακτική αξιολόγηση κινδύνου στον πληθυσμό και οι συστάσεις ή παρεμβάσεις, που περιλαμβάνουν άσκηση, από επαγγελματίες υγείας ως σημαντικό στοιχείων των κοινωνικών πολιτικών του συστήματος.  Η παρούσα μελέτη περιγράφει πτώσεις και φυσική δραστηριότητα (ΦΔ) ΑΤΗ στο Ν. Ροδόπης. Μέθοδος. ΑΤΗ που διαμένουν στο Ν.Ροδόπης, μέλη ΚΑΠΗ στην πλειοψηφία, εντάχθηκαν στη μελέτη από 10/2021 έως 11/2022 στα πλαίσια του διεπιστημονικού προγράμματος ΑΣΠΙΔΑ. Χρησιμοποιήθηκαν ερωτηματολόγια ΦΔ και κινδύνου πτώσεων, ενώ η ΦΔ μετρήθηκε και με επιταχυνσιόμετρα Actigraph για μια εβδομάδα. Αποτελέσματα. Συνολικά εντάχθηκαν στην μελέτη 192 άτομα, μ.ο. ηλικίας 72.8 έτη, στην πλειονότητα γυναίκες(83.9%) και διαμένοντες σε πόλη(74.5%). Το 85.5% είχαν τουλάχιστον μία πάθηση που σχετίζεται με αυξημένο κίνδυνο πτώσεων. 33.5% ανέφεραν τουλάχιστον μία πτώση τους τελευταίους 12 μήνες, 9.5% με τραυματισμό. Από όσους είχαν πτώση, μόνο το 11.2% ανέφερε ότι είχε συζητήσει με το γιατρό του για πρόληψη. Σύμφωνα με τις ερωτήσεις αυτό-αναφοράς 50% περπατούσαν >3ώρες/εβδομάδα και 13% έκαναν τουλάχιστον 1ώρα/εβδομάδα άλλη σωματική άσκηση(10.9% αυτών με ιστορικό πτώσεων). Με βάση τα επιταχυνσιόμετρα 91.9% έκαναν τουλάχιστον 150λεπτά/εβδομάδα μέτρια-προς-έντονη ΦΔ. Συμπεράσματα. Η πρόληψη πτώσεων είναι σημαντική για την πρόληψη των οστεοπορωτικών καταγμάτων. Δυστυχώς σύμφωνα με την παρούσα μελέτη, 1/3 συμμετέχοντες ανέφεραν πτώση το τελευταίο 12μηνο, και 1/10 δέκα με τραυματισμό. Στην πλειοψηφία τους οι συμμετέχοντες έκαναν επαρκή (συστάσεις Π.Ο.Υ.) αεροβική ΦΔ, αλλά όχι άσκηση ενδυνάμωσης/ισορροπίας/λειτουργική, που είναι καίρια στην πρόληψη πτώσεων. Η ενσωμάτωση σχετικών παρεμβάσεων στην κλινική πράξη μπορεί να συμβάλει σε αποτελεσματικότερη πρόληψη πτώσεων.

Α. Καναβάκη, Μ. Μιχαλοπούλου, Ε. Ηλιόπουλος, Α. Γκρεκίδης, Ε. Κούλη, Ν. Αγγελούσης

The increasing demand for applications in competitive fields, such as assisted living and aerial robots, drives contemporary research into the development, implementation and integration of power-constrained solutions. Although, deep neural networks (DNNs) have achieved remarkable performances in many robotics applications, energy consumption remains a major limitation. The paper at hand proposes a hybrid variation of the well-established deep deterministic policy gradient (DDPG) reinforcement learning approach to train a 6 degree of freedom robotic arm in the target-reach task. In particular, we introduce a spiking neural network (SNN) for the actor model and a DNN for the critic one, aiming to find an optimal set of actions for the robot. The deep critic network is employed only during training and discarded afterwards, allowing the deployment of the SNN in neuromorphic hardware for inference. The agent is supported by a combination of RGB and laser scan data exploited for collision avoidance and object detection. We compare the hybrid-DDPG model against a classic DDPG one, demonstrating the superiority of our approach.

Katerina Maria Oikonomou, Ioannis Kansizoglou, Antonios Gasteratos

Discrete quantum walks are a universal model of quantum computation equivalent to the quantum circuit model and can be mapped onto quantum circuits and executed using quantum computers. Quantum walks can model and simulate many physical systems and several quantum algorithms are based on them. Discrete quantum walks have been extensively studied, but quantum walks that evolve in spaces in which potentials are applied received little or no attention. Here, we formulate the discrete quantum walk model in one and two-dimensional spaces in which potentials are applied. In this formulation the quantum walker carries a “charge” affected by the potentials and the walk evolution is driven by both constant and time-varying potentials. We reproduce the tunneling through a barrier phenomenon and study the quantum walk evolution in one and two-dimensional spaces with various potential distributions. We demonstrate that our formulation can serve as a basis for applied quantum computing by studying maze running and the motion of vehicles in urban spaces. In these spaces curbs and buildings are modeled as impenetrable potential barriers and traffic lights as time-varying potential barriers. Quantum walks in spaces with applied potentials may open the way for the development of novel quantum algorithms in which inputs are introduced as potential profiles.

Georgios D. Varsamis, Ioannis G. Karafyllidis, Georgios Ch. Sirakoulis

In recent decades, modern developed societies have experienced a significant increase in life expectancy, resulting in a significant degree of aging in the population. This demographic shift has had a significant impact on the daily needs and habits of the population. According to recent demographic projections, the elderly population is expected to triple and reach 2 billion people worldwide by 2050. However, future societies around the world are not adequately prepared to face the potential challenges arising from population aging. Elderly people tend to spend more time at home, but the functionalities and ergonomics of current home equipment and appliances do not fully satisfy their needs. Moreover, the absence of familiarity with new, smarter, and automated domestic technologies creates a significant gap that affects the acceptance and adoption of these technologies by the elderly. In this work, we present the most important technical challenges for the integration of assisted living technologies into a Mobile Robotic Platform (MRP). Additionally, we present the ASPiDA concept, a project proposing a holistic system to support elderly people in their domicile environment.

Christoforos Keroglou, Ioannis Kansizoglou, Panagiotis Michailidis, Katerina Maria Oikonomou, Ioannis Tsampikos Papapetros, Paraskevi Dragkola, Iakovos T. Michailidis, Antonios Gasteratos, Elias B. Kosmatopoulos, Georgios Ch. Sirakoulis

Due to the wide spread of robotics technologies in everyday activities, from industrial automation to domestic assisted living applications, cutting-edge techniques such as deep reinforcement learning are intensively investigated with the aim to advance the technological robotics front. The mandatory limitation of power consumption remains an open challenge in contemporary robotics, especially in real-case applications. Spiking neural networks (SNN) constitute an ideal compromise as a strong computational tool with low-power capacities. This paper introduces a spiking neural network actor for a baseline robotic manipulation task using a dual-finger gripper. To achieve that, we used a hybrid deep deterministic policy gradient (DDPG) algorithm designed with a spiking actor and a deep critic network to train the robotic agent. Thus, the agent learns to obtain the optimal policies for the three main tasks of the robotic manipulation approach: target-object reach, grasp, and transfer. The proposed method has one of the main advantages that an SNN possesses, namely, its neuromorphic hardware implementation capacity that results in energy-efficient implementations. The latter accomplishment is highly demonstrated in the evaluation results of the SNN actor since the deep critic network was exploited only during training. Aiming to further display the capabilities of the introduced approach, we compare our model with the well-established DDPG algorithm.

Katerina Maria Oikonomou, Ioannis Kansizoglou, Antonios Gasteratos

Acetone vapor sensors find great use in many areas as they are being used for non-invasive diabetes detection and monitoring, fat burning rate monitoring, general industrial applications or even detection of explosive environments. Even though several sensors have been proposed, there is a continuous need for low power and low-cost ones, ideally implemented on flexible substrates. Thus, 2D nanomaterials such as graphene, due to their large specific surface area and exceptional electrical properties, have attracted broad attention for their gas sensing potential even at low temperature range of operation. As an example, recently, an acetone sensor was shown using graphene nanoplatelets combined with Zinc ferrite (ZnFe₂O₄) and achieved a sensitivity of 7% at 200 ppm acetone vapor while operating at 275°C. In this work, we compare various graphene-based gas sensors, capable of detecting acetone vapors, that are implemented on glass substrates and operate at room temperature. We demonstrate that such sensors exhibit good repeatability and sensitivity at 200 ppm of acetone vapor. All acetone sensing measurements were conducted by exposing the sensors in repeating cycles of several acetone concentrations diluted in nitrogen gas. To test the repeatability of the sensors, they were exposed to three cycles of 200 ppm acetone vapors followed by nitrogen gas during recovery. The sensors were tested in room temperature.

Michael Georgas, George Zardalidis, Filippos Farmakis

Net-metering (NEM) is one of the most widely known support mechanisms aiming to promote the installation of distributed photovoltaic (PV) systems. However, due to the increasing penetration of PVs, challenges related to the secure operation of the power system are emerged. For this reason, battery energy storage (BES) systems are installed alongside PVs to tackle these technical problems. In this paper, a systematic assessment analysis of NEM policy in medium-voltage (MV) prosumers with PV-BES systems, e.g., university campuses, hospitals, etc., is conducted in both technical and economic terms. In the analysis, annual generation and consumption timeseries of university campuses of the Democritus University of Thrace, Greece, are indicatively employed and various scenarios of PV-BES systems are investigated, to evaluate the profitability of NEM policy in MV PV-BES prosumers and consequently determine the optimal investment plan in monetary terms.

Kalliopi Pippi, Evangelos Kyriakopoulos, Theofilos Papadopoulos, Georgios Kryonidis

This paper discusses the incorporation of battery units in modern Zero Energy Buildings (ZEBs), on the path toward a more energy-efficient building sector in the EU. In more detail, the ZEB concept and its benefits in terms of environmental footprint are discussed, along with its impact on electrical distribution grids. In this context, the incorporation of battery units is regarded as a tool to tackle power quality and protection issues raised due to the mass ZEB transformation. In addition, the efficacy of the most popular design tools for ZEB energy production is assessed, in light of batteries installations. In this initial work, a typical case study is considered, highlighting the benefits (and the technical limitations) of their incorporation.

Giorgos Diamantidis, Faidra Kotarela, Nick Rigogiannis, Anastasios Kyritsis, Nick Papanikolaou

During the past decade, microelectronics incorporated inkjet-printing technology as a versatile tool for industrial applications, as it combines high printing quality and resolution along with low cost compared to conventional microelectronics techniques that require high cost and complex equipment. In addition, inherently, inkjet printing requires no photolithography steps. In this review paper we present temperature sensors that have been manufactured by inkjet printing. Based on the main active sensing material, the research studies are classified into four different types, i.e. metal and metal oxide-based, carbon-based, polymer-based and composite that combine properties of multiple active materials. It is demonstrated that silver ink has been, by far, the most popular material for metal based temperature sensors with a temperature coefficient of resistance (TCR) of around 2 × 10 ^-3 ◦C ^-1 . Silver sensors as well as almost all metal, carbon and polymer based sensors were either resistance temperature detectors (RTDs) or thermistors. Regarding polymer and carbon-based sensors, it was found that, in some cases, they outperformed metal-based sensors in terms of TCR, although fabrication using such materials had a less predictable result with a TCR ranging from -16 × 10 ^-3 ◦C ^-1 to 2 × 10 ^-3 ◦C ^-1 . Finally, in the case of composite materials as temperature sensors, several combinations of active materials exhibited interesting results and yielded a variety of sensing technologies such as thermocouples and radio frequency based.

Michael Georgas, Petros Selinis, George Zardalidis, Filippos Farmakis

The rapid growth of aging population dictates the necessity of sophisticated monitoring and actuation systems for the smart control and management of elderly households. This paper proposes a state-of-the-art energy management system aiming at increased energy efficiency, lower electricity cost and improved user comfort. The study focuses on a Greek residency with elderly people incorporating controllable and uncontrollable loads, an energy storage system, and photovoltaic generation. A smart home energy management system under the net-metering policy is proposed consisting of three control mechanisms. The reported results offer insights into the optimal residential management practices and evaluate the performance of the proposed control strategy in comparison to other alternative demand response solutions.

Christos L. Athanasiadis, Kalliopi D. Pippi, Theofilos A. Papadopoulos, Christos Korkas, Christos Tsaknakis, Vasiliki Alexopoulou, Vasileios Nikolaidis, Elias Kosmatopoulos

Introduction. Falls in older adults are a public health problem and physical activity (PA) is an important behavioural factor in falls prevention. Following the reopening of community centers post Covid-19 restrictions, the study aimed to compare physical activity (PA) and sedentary behaviour (SB) of older adults at different levels of fall risk (FR) and identify socio-cognitive PA determinants. Methods. This is baseline data from ASPIDA, an on-going, longitudinal study, with 96 current recruits from community centers for older adults in Municipality of Rodopi, Northern Greece. Measures included an adjusted version of John Hopkins Fall Risk Assessment Tool, the Geriatric Depression Scale, Falls Efficacy Scale-International, Multidimensional Outcome Expectations for Exercise Scale, Self-efficacy for Exercise Scale, administered during site-visits, and triaxial accelerometers (Actigraph GT3X, GT9X), hip-worn for a week. Results. Mean age was 73.6 years (SD=6.2), 77.7% women. The proportion of people at low, medium and high FR was 22%, 60% and 16% respectively, with 69% reporting no fall history in the last 12 months. On average participants wore the accelerometers for 6.4 days and spent 58.3 minutes (SD=28.9) in moderate-to-vigorous PA (MVPA), 126.2 minutes (SD=36) in light PA (LPA) and 9.9 hours (SD=2) in SB daily. The guidelines of 150 minutes of weekly MVPA were met by 91% of participants. 67% reported doing no exercise, whilst 13% reported doing ≥1hour/week. There was no effect of sex, BMI and depression on FR. There was a significant effect of MVPA [F(2,34.35)=22.08, p<.001, η²= .15], the proportion of time spent in SB [F(2,30.39)=11.21, p<.001, η²= .12] and fear of falls [F(2,22.74)=10.60, p<.001, η²= .36] on FR. Those at high FR spent significantly less time in MVPA, more time in SB and had greater fear of falls than those at medium and low FR groups. Participants’ PA/exercise outcome expectations (OE) were overall positive [M=4.3 (SD=0.5) for physical OE, M=4.2 (SD=0.6) for self-evaluative, M=3.2 (SD=0.7) for social]; self-efficacy was average [M=5.33 (SD=1.9)]. Only self-evaluative outcome expectations were associated with MVPA [r(73)= 0.26, p<.05]. Discussion. The majority of participants met MVPA, but not exercise guidelines. Those at high risk of falls had a distinct profile with regard to MVPA and SB. In terms of falls prevention, which is the overarching aim of the ASPIDA project, these preliminary findings have implications for behaviour change targets: (a) increase exercise time and (b) boost PA associations with well-being and self-efficacy for all, (c) tailor interventions for individuals at high FR, e.g. by focusing on SB reduction and addressing fear of falls and safety concerns.

A. Kanavaki, M. Michalopoulou, N. Aggelousis and the ASPIDA Group

A physically active lifestyle is important for healthy ageing. Yet with regard to falls, beneficial and harmful effects of specific physical activity (PA) and sedentary behaviour (SB) patterns, like bouted and total PA of various intensities, are unclear. The study aims to explore the relationships of objectively measured PA and SB patterns with physical function, falls and quality of life in community-dwelling older adults at high/low risk of falls; also, to identify psychosocial determinants of PA and SB patterns, namely motivation, self-efficacy, outcome expectations, fear of falls, depression. Methods. This is an observational study with 12-month follow-up, part of the wider multidisciplinary ASPIDA project, is set up at community centers in municipality of Rodopi, Northern Greece. Baseline recruitment will place from October 2021 to October 2022 with the aim to recruit 300 adults over 65 years. Measures include accelometry (waist-worn Actigraph GT3X, GT9X), questionnaires and physical performance tests repeated at baseline and follow-up visits to recruiting sites, with quarterly recording of falls via telephone contact. Regression models will examine if changes in hypothesized determinants predict changes in outcome variables for each aim and path analysis will examine multivariate relationships and hypothesized process models. Discussion. Examining theory-informed underlying mechanisms of PA/ SB behaviours and implications of PA/ SB for function, falls and well-being, the study will build a comprehensive evidence base for understanding this modifiable lifestyle factor in community-dwelling older adults. Findings will guide the development of lifestyle interventions, such as community- and home-based PA interventions, to improve the selected outcomes.

A. Kanavaki, N. Aggelousis, M.Michalopoulou and the ASPIDA Group

The spatial search problem is an interesting and important problem in computer science and especially the area of algorithms. The objective is a marked site to be found in a finite physical space, that can be modeled as a finite lattice or a graph. Many approaches have been developed to address this problem. Classical random walks and quantum walks are efficient models that address the spatial search problem. Quantum walks is a universal model of quantum computation and can be mapped directly to quantum circuits and consequently executed on quantum computers. Quantum walks utilized for quantum search proved to achieve significantly lower hitting times than their classical counterpart, classical random walks. The evolution space for the quantum walks as well as the classical random walks is up until now a free space. In our approach, we introduce external electrical potentials to the evolution space. We study the evolution of discrete time quantum and classical random walks in such potential spaces and the probability — hitting time on finding marked sites. We considered the differences in applied potential among neighboring sites as weights for the lattice — graph. We introduce these weights to the evolution space as an operator for the discrete time quantum walk and as coin probabilities for the classical random walk. Our results show that quantum walks again, evolve faster in the evolution space with the applied potential. Quantum walks also achieve better probability — hitting time on finding the marked site in the potential space. With the introduction of electrical potentials, quantum walks evolving in potential spaces, can lead to the development of novel quantum algorithms, where input parameters can be introduced as external potentials.

Georgios D. Varsamis, Ioannis G. Karafyllidis, Georgios Ch. Sirakoulis

Robust and energy-efficient robot planning is of utmost importance for mobile robots since the dynamic changes of the environment entail robotic agents with high adaptation capacities, so as to excel in their tasks. In this work, we introduce a hybrid spiking and deep neural network architecture for actor-critic control of a 6-DOF robot arm. Our method firstly involves autonomous object detection via active vision exploration and thereafter, the entire hybrid architecture is described. In specific, the actor utilises an integrated-and-fire model for action generation, while the critic a deep neural one for action evaluation. Lastly, the benefits of this approach in terms of energy efficiency are extensively discussed.

Katerina Maria Oikonomou, Ioannis Kansizoglou, Antonios Gasteratos

Unconventional circuits with built-in memory and computing functionalities are becoming the cornerstones of artificial intelligence (AI) at the edge. In the currently deployed systems, sensing and computing occur in separate physical locations, imposing a vast amount of data shuttling between the sensor module and the cloud-computing platforms. Regarding the acceleration of image processing at the edge, in this work, a memristive computing circuit has been designed. By exploiting the non-linear behavior and memory capabilities of memristor devices, a memristive circuit, capable of tracking the shifting of an image is proposed. The presented circuit design can be also combined with an array of sensors, aiming to implement a discrete image tracking module.

Theodoros Panagiotis Chatzinikolaou, Iosif-Angelos Fyrigos, Georgios Ch. Sirakoulis

As the aged population is rapidly increased, the need for efficient and low-cost ambient systems becomes vital. The effectiveness of such systems lies upon the accurate and fast motion analysis in order to predict the elderly’s action and develop systems to act in need. To achieve that, the precise estimation of the entire human body pose is often exploited, providing the required motion-related information. Yet, the exploitation of the entire human pose can present several limitations. The paper at hand exploits state-of-the-art data-driven classifiers and compares their efficiency in action recognition based on a specific set of joints or coordinates, i.e., the x, y and z-axis. The above rests upon the notion that each action in real life can be effectively perceived by observing only a specific set of joints. Considering that, we aim to investigate the capacity of such a joint analysis and its ability to deliver an enhanced pose-based action recognition system. To that end, we correlate specific joints with each action, indicating the joints that contribute the most. We evaluate our findings on two different senior subjects using two different classifiers, viz., support vector machine (SVM) and convolutional neural network (CNN), showing that the above strategy can improve recognition rates.

Katerina Maria Oikonomou, Ioannis Kansizoglou, Pelagia Manaveli, Athanasios Grekidis, Dimitrios Menychtas, Nikolaos Aggelousis, Georgios Ch. Sirakoulis, Antonios Gasteratos

As loop-closure detection plays a fundamental role in any simultaneous localization and mapping (SLAM) system, through its ability to recognize previously visited locations, one of its main objectives is to permit consistent map generation for an extended period. Within large-scale SLAM autonomy, the scalability in terms of timing needed for database search and the storage requirements has to be addressed. In this paper, a low-storage visual loop-closure detection technique is proposed. Our system is based on the incremental bag-of-tracked-words scheme for the trajectory mapping still, the generated visual representations are reduced to lower dimensions through a resampling process. This way, we achieve to shorten the overall database size and searching time, while at the same time preserving the high performance. The evaluation, which took place on different well-known datasets, exhibits the system’s low-storage requirements and high recall scores compared to the baseline version and other state-of-the-art approaches.

Konstantinos A. Tsintotas, Shan An, Ioannis Tsampikos Papapetros, Fotios K. Konstantinidis, Georgios Ch. Sirakoulis, Antonios Gasteratos

One’s internal state is mainly communicated through nonverbal cues, such as facial expressions, gestures and tone of voice, which in turn shape the corresponding emotional state. Hence, emotions can be effectively used, in the long term, to form an opinion of an individual’s overall personality. The latter can be capitalized on in many human–robot interaction (HRI) scenarios, such as in the case of an assisted-living robotic platform, where a human’s mood may entail the adaptation of a robot’s actions. To that end, we introduce a novel approach that gradually maps and learns the personality of a human, by conceiving and tracking the individual’s emotional variations throughout their interaction. The proposed system extracts the facial landmarks of the subject, which are used to train a suitably designed deep recurrent neural network architecture. The above architecture is responsible for estimating the two continuous coefficients of emotion, i.e., arousal and valence, following the broadly known Russell’s model. Finally, a user-friendly dashboard is created, presenting both the momentary and the long-term fluctuations of a subject’s emotional state. Therefore, we propose a handy tool for HRI scenarios, where robot’s activity adaptation is needed for enhanced interaction performance and safety.

Ioannis Kansizoglou, Evangelos Misirlis, Konstantinos Tsintotas, Antonios Gasteratos

Finding or estimating the lowest eigenstate of quantum system Hamiltonians is an important problem for quantum computing, quantum physics, quantum chemistry, and material science. Several quantum computing approaches have been developed to address this problem. The most frequently used method is variational quantum eigensolver (VQE). Many quantum systems, and especially nanomaterials, are described using tight-binding Hamiltonians, but until now no quantum computation method has been developed to find the lowest eigenvalue of these specific, but very important, Hamiltonians. We address the problem of finding the lowest eigenstate of tight-binding Hamiltonians using quantum walks. Quantum walks is a universal model of quantum computation equivalent to the quantum gate model. Furthermore, quantum walks can be mapped to quantum circuits comprising qubits, quantum registers, and quantum gates and, consequently, executed on quantum computers. In our approach, probability distributions, derived from wave function probability amplitudes, enter our quantum algorithm as potential distributions in the space where the quantum walk evolves. Our results showed the quantum walker localization in the case of the lowest eigenvalue is distinctive and characteristic of this state. Our approach will be a valuable computation tool for studying quantum systems described by tight-binding Hamiltonians.

Georgios D. Varsamis, Ioannis G. Karafyllidis

Loop-closure detection (LCD) has become an essential part of any simultaneous localization and mapping (SLAM) framework. It provides a means to rectify the drift error, which is typically accumulated along a robot’s trajectory. In this article we propose an LCD method based on tracked visual features, combined with a signal peak-trace filtering approach for loop-closure identification. In particular, local binary features are firstly extracted and tracked through consecutive frames. This way online visual words are generated, which in turn form an incremental bag of visual words (BoVW) vocabulary. Loop-closures (LCs) result from a classification method, which considers current and past state peaks on the similarity matrix. The system discerns the movement of the peaks to identify whether they come about to be true-positive detections or background noise. The suggested peak-trace filtering technique provides exceeding robustness to noisy signals, enabling the usage of only a handful of visual local features per image; thus resulting into a considerably downsized visual vocabulary.

Ioannis Tsampikos Papapetros, Vasiliki Balaska, Antonios Gasteratos

The exploitation of deep neural networks (DNNs) as descriptors in feature learning challenges enjoys apparent popularity over the past few years. The above tendency focuses on the development of effective loss functions that ensure both high feature discrimination among different classes, as well as low geodesic distance between the feature vectors of a given class. The vast majority of the contemporary works rely their formulation on an empirical assumption about the feature space of a network’s last hidden layer, claiming that the weight vector of a class accounts for its geometrical center in the studied space. This article at hand follows a theoretical approach and indicates that the aforementioned hypothesis is not exclusively met. This fact raises stability issues regarding the training procedure of a DNN, as shown in our experimental study. Consequently, a specific symmetry is proposed and studied both analytically and empirically that satisfies the above assumption, addressing the established convergence issues. More specifically, the aforementioned symmetry suggests that all weight vectors are unit, coplanar, and their vector summation equals zero. Such a layout is proven to ensure a more stable learning curve compared against the corresponding ones succeeded by popular models in the field of feature learning.

Ioannis Kansizoglou, Loukas Bampis, Antonios Gasteratos