Abstract This document proposes a spatiotemporal deep learning model for ionospheric total electron content (TEC) modeling using global navigation satellite systems (GNSSs) observables. Data from dual-frequency GNSS receivers are used to compute the daily GNSS TEC timeseries. Usually, these timeseries are computed independently per GNSS permanent station, and the state-of-the art models proposed in literature exploit temporal characteristics of the timeseries and neglect any spatial dependencies and information from different stations. In our approach, we propose a practical solution for parallel processing of TEC timeseries and additional indicators from various adjacent stations to predict future VTEC values. We face the problem in both spatial and temporal dimensions adopting a graph neural network-based approach from the broader family of geometric deep learning. According to our proposed scheme, the different adjacent GNSS stations are structured in a graph and then, we apply the proposed temporal graph convolutional network called ION_TGNN. Our model predicts future vertical TEC (VTEC) values for all stations in a single run with mae error better than 1.0 TECU. Comparisons with state-of-the art models show the superiority of the proposed method in terms of performance but also in terms of computational cost during training and test phases.

Artificial intelligence (AI) has enabled agents to master complex video games, from first-person shooters likeCounter-Striketo real-time strategy games such asStarCraft IIand racing games likeGran Turismo. While these achievements are notable, applying these AI methods in commercial video game production remains challenging due to computational constraints. In commercial scenarios, the majority of computational resources are allocated to 3D rendering, leaving limited capacity for AI methods, which often demand high computational power, particularly those relying on pixel-based sensors. Moreover, the gaming industry prioritizes creating human-like behavior in AI agents to enhance player experience, unlike academic models that focus on maximizing game performance. This paper introduces a novel methodology for training neural networks via imitation learning to play a complex, commercial-standard, VALORANT-like 2v2 tactical shooter game, requiring only modest CPU hardware during inference. Our approach leverages an innovative, pixel-free perception architecture using a small set of ray-cast sensors, which capture essential spatial information efficiently. These sensors allow AI to perform competently without the computational overhead of traditional methods. Models are trained to mimic human behavior using supervised learning on human trajectory data, resulting in realistic and engaging AI agents. Human evaluation tests confirm that our AI agents provide human-like gameplay experiences while operating efficiently under computational constraints. This offers a significant advancement in AI model development for tactical shooter games and possibly other genres.

Affective computing faces a pressing challenge: the limited ability of affect models to generalise amidst varying contextual factors within the same task. While well recognised, this challenge persists due to the absence of suitable large-scale corpora with rich and diverse contextual information within a domain. To address this challenge, this paper introduces a GameVibe, a novel corpus explicitly tailored to confront the lack of contextual diversity. The affect corpus is sourced from 30 First Person Shooter (FPS) games, showcasing diverse game modes and designs within the same domain. The corpus comprises 2 hours of annotated gameplay videos with engagement levels annotated by a total of 20 participants in a time-continuous manner. Our preliminary analysis on this corpus sheds light on the complexity of generalising affect predictions across contextual variations in similar affective computing tasks. These initial findings serve as a catalyst for further research, inspiring deeper inquiries into this critical, yet understudied, aspect of affect modelling.

Predicting channel excess attenuation is of utmost importance for the current satellite networks design. However, the adoption of data-driven models is hampered by the absence of reliable satellite propagation measurements in real conditions. In this framework, ESA initiated in 2015 a dedicated project for a large-scale measurement campaign using the Alphasat SCIEX Ka/Q band signals (ASALASCA). Here, we propose a data-driven model to predict in excess attenuation in the next future time-steps, by exploiting the real measurements (excess attenuation and rainfall rate) from the two experimental stations in Greece by NTUA operating at Ka band (19.704GHz).

Non-intrusive load monitoring (NILM) is considered an efficient approach to infer the consumption pattern of household appliances from the aggregate consumption signal. Continual adaptability is an important aspect of practical NILM applications, as they usually require frequent post-deployment maintenance to deal with non-stationary appliances’ data distributions. However, in most approaches, the trained deep learning model weights remain static, potentially neglecting valuable information that can be used for further model training. This work alleviates the aforementioned limitation by introducing ContiNILM, a continual learning scheme for NILM to build robust models that track environmental/seasonal alterations with direct impact on several appliances’ operation. In our approach, model weights do not remain static, but utilize additional training data to further improve the disaggregation performance. A novel mechanism is proposed that determines whether new incoming samples would be beneficial for model training and alleviates the risk of "forgetting" previously learned knowledge. Experimental results demonstrate the efficiency of the proposed approach.

This work proposes an interpretable Deep Learning framework utilizing Vision Transformers (ViT) for the classification of remote sensing images into land use and land cover (LULC) classes. It uses the Shapley Additive Explanations (SHAP) values to achieve two-stage explanations: 1) bandwise feature importance per class, showing which band assists the prediction of each class and 2) spatial-wise feature understanding, explaining which embedded patches per band affected the network's performance. Experimental results on the EuroSAT dataset demonstrate the ViT's accurate classification with an overall accuracy 96.86 %, offering improved results when compared to popular CNN models. Heatmaps in each one of the dataset's existing classes highlight the effectiveness of the proposed framework in the band explanation and the feature importance.

An interpretable deep learning framework for land use and land cover classification (LULC) in remote sensing using SHAP is introduced. It utilizes a compact CNN model for the classification of satellite images and then feeds the results to a SHAP deep explainer so as to strengthen the classification results. The proposed framework is applied to Sentinel-2 satellite images containing 27000 images of pixel size 64 × 64 and operates on three-band combinations, reducing the model’s input data by 77% considering that 13 channels are available, while at the same time investigating on how different spectrum bands affect predictions on the dataset’s classes. Experimental results on the EuroSAT dataset demonstrate the CNN’s accurate classification with an overall accuracy of 94.72%, whereas the classification accuracy on three-band combinations on each of the dataset’s classes highlights its improvement when compared to standard approaches with larger number of trainable parameters. The SHAP explainable results of the proposed framework shield the network’s predictions by showing correlation values that are relevant to the predicted class, thereby improving the classifications occurring in urban and rural areas with different land uses in the same scene.

Understanding the dynamics of deforestation and land uses of neighboring areas is of vital importance for the design and development of appropriate forest conservation and management policies. In this article, we approach deforestation as a multilabel classification (MLC) problem in an endeavor to capture the various relevant land uses from satellite images. To this end, we propose a multilabel vision transformer model, ForestViT, which leverages the benefits of the self-attention mechanism, obviating any convolution operations involved in commonly used deep learning models utilized for deforestation detection. Experimental evaluation in open satellite imagery datasets yields promising results in the case of MLC, particularly for imbalanced classes, and indicates ForestViT's superiority compared with well-established convolutional structures (ResNET, VGG, DenseNet, and ModileNet neural networks). This superiority is more evident for minority classes.

Links between environmental conditions (e.g., meteo-rological factors and air quality) and COVID-19 infection/mortality have been reported worldwide. However, the existing statistical frameworks are insufficient to investigate the factors that increase the risk for COVID-19 in urban areas. In this paper, we extend the concept of machine learning-based predictive modelling for COVID-19 spread, proposing an explainable AI approach in order to i) prioritize the risk factors, ii) define the interconnections between them and iii) detect positive or negative influence of the factors with respect to COVID-19 morbidity and mortality.

Modeling ionospheric variability throughout a proper total electron content (TEC) parameter estimation is a demanding, however, crucial, process for achieving better accuracy and rapid convergence in precise point positioning (PPP). In particular, the single-frequency PPP (SF-PPP) method lacks accuracy due to the difficulty of dealing adequately with the ionospheric error sources. In order to apply ionosphere corrections in techniques, such as SF-PPP, external information of global ionosphere maps (GIMs) is crucial. In this article, we propose a deep learning model to efficiently predict TEC values and to replace the GIM-derived data that inherently have a global character, with equal or better in accuracy regional ones. The proposed model is suitable for predicting the ionosphere delay at different locations of receiver stations. The model is tested during different periods of time, under different solar and geomagnetic conditions and for stations in various latitudes, providing robust estimations of the ionospheric activity at the regional level. Our proposed model is a hybrid model comprising of a 1-D convolutional layer used for the optimal feature extraction and stacked recurrent layers used for temporal time series modeling. Thus, the model achieves good performance in TEC modeling compared to other state-of-the-art methods.

The problem of separating the household aggregated power signal into its additive sub-components, called energy (power) disaggregation or Non-Intrusive Load Monitoring (NILM) can play an instrumental role as a driver towards consumer energy consumption awareness and behavioral change. In this paper, we propose EnerGAN++, an adversarially trained model for robust energy disaggregation. We propose a unified autoencoder (AE) and GAN architecture, in which the AE achieves a non-linear power signal source separation. The discriminator performs sequence classification, using a recurrent CNN to handle the temporal dynamics of an appliance energy consumption time series. Experimental results indicate the proposed method’s superiority compared to the state of the art.

Ionospheric variability and disturbances can affect technologies in space and on Earth, disrupting satellite operations, communications networks, and navigation systems. The availability of numerous satellites deployed by GPS, GLONASS, Galileo, BeiDou navigation systems allows continuous monitoring of the Earth's ionosphere using measurements from these satellites. Here, we scrutinize the effectiveness and efficiency of a convolutional enriched recurrent neural network for spatio-temporal VTEC prediction. In our analysis, we have chosen different years under different solar and geomagnetic activity. We test our models for different days and at various latitudes to see model's response in cases of high ionosphere activity. Our experiments indicate that the proposed combined deep CNN-GRU model is capable of providing an accurate prediction of TEC values even in intense conditions.

An efficient, appliance-level approach for energy disaggregation, exploiting the benefits of Generative Adversarial Networks, is presented. The concept of adversarial training supports the creation of fine tuned dissagregators, which produce more detailed load estimations for a specific appliance, compared to state of the art deep learning models. The Generator and Discriminator of the model are appropriately adapted to fit the particularities of NILM problem, whereas a Seeder component is added to provide encoded compact input vectors to the Generator. The experimental evaluation against state of the art techniques indicates promising results.

In this paper, a Bayesian-optimized bidirectional Long Short -Term Memory (LSTM) method for energy disaggregation, is introduced. Energy disaggregation, or Non-Intrusive Load Monitoring (NILM), is a process aiming to identify the individual contribution of appliances in the aggregate electricity load. The proposed model, Bayes-BiLSTM, is structured in a modular way to address multi-dimensionality issues that arise when the number of appliances increase. In addition, a non-causal model is introduced in order to tackle with inherent structure, characterizing the operation of multi-state appliances. Furthermore, a Bayesian-optimized framework is introduced to select the best configuration of the proposed regression model, thus increasing performance. Experimental results indicate the proposed method's superiority, compared to the current state-of-the-art.