Loukas Ilias

Predicting ship fuel oil consumption is crucial for optimizing fuel efficiency, reducing operational costs, and minimizing environmental impact. Prior work uses either shallow machine learning (ML) algorithms (e.g., AdaBoost, Bagging, ExtraTrees, Random Forest) or deep neural networks that end with dense layers (e.g., BiLSTMs). However, training shallow ML algorithms yields suboptimal performance, while using dense layers into a deep neural network creates a tradeoff between capacity and practicality. To tackle these limitations, we introduce a Mixture-of-Experts (MoE) framework that combines BiLSTM, self-attention and a MoE block, and we study three variants: sparsely-gated MoE, CPμMoE, and TRμMoE. In terms of the first approach, we use a sparsely-gated MoE layer, which consists of experts and a gating network. Sparsity and noise are added to the gating network for reducing computational complexity. Next, we utilize two variants of multilinear MoE layers, which are based on factorization approaches, namely Candecomp/PARAFAC (CP) and Tensor Ring decomposition. Experiments on the DataBio fishing-ships dataset (three vessels, one month each) show that our best model (CPμMoE) consistently improves the coefficient of determination (R²) and reduces root mean squared error (RMSE), mean absolute error (MAE), and mean bias error (MBE), relative to Single-Task Learning (STL) - the same BiLSTM + self-attention backbone but with a standard dense head (no MoE) - and the tree-ensemble baselines (AdaBoost, BaggingRegressor, ExtraTrees, Gradient Boosting, Random Forest) under the same ship-stratified protocol. Ablations confirm the contribution of MoE components. These results indicate that input-conditional experts can better fit diverse operating conditions in multivariate ship sensor time series.

The advent of 6G/NextG networks offers numerous benefits, including extreme capacity, reliability, and efficiency. To mitigate emerging security threats, 6G/NextG networks incorporate advanced artificial intelligence algorithms. However, existing studies on intrusion detection predominantly rely on deep neural networks with static components that are not conditionally dependent on the input, thereby limiting their representational power and efficiency. To address these issues, we present the first study to integrate a Mixture of Experts (MoE) architecture for the identification of malicious traffic. Specifically, we use network traffic data and convert the 1D feature array into a 2D matrix. Next, we pass this matrix through a convolutional neural network (CNN) layer, followed by batch normalization and max pooling layers. Subsequently, a sparsely gated MoE layer is used. This layer consists of a set of expert networks (dense layers) and a router that assigns weights to each expert's output. Sparsity is achieved by selecting only the most relevant experts from the full set. Finally, we conduct a series of ablation experiments to demonstrate the effectiveness of our proposed model. Experiments are conducted on the 5G-NIDD dataset, a network intrusion detection dataset generated from a real 5G test network, and the NANCY dataset, which includes cyberattacks from the O-RAN 5G Testbed Dataset. The results show that our introduced approach achieves accuracies of up to 99.96% and 79.59% on the 5G-NIDD and NANCY datasets, respectively. The findings also show that our proposed model offers multiple advantages over state-of-the-art approaches.

The rapid development of 5G and emerging 6G networks introduces advanced security threats (e.g., DDoS, eavesdropping, jamming, etc.), necessitating advanced intrusion detection systems. Current studies use outdated benchmark datasets that do not capture the dynamic, AI-driven traffic of these networks. Additionally, they often rely on payload data for intrusion detection; however, in 5G/6G networks, payloads are typically encrypted, limiting the applicability and effectiveness of these approaches. While some research explores fusion methods, they typically overlook crucial inter-and intra-modal interactions. To address these limitations, this letter applies multimodal fusion methods on a publicly available dataset reflecting realistic 5G/6G conditions. Features are grouped into packet-related, time-related, and length-related modalities. Fusion methods, including bilinear pooling, gated multimodal unit, and block-term tensor decomposition, are employed. Results show that block term tensor decomposition achieves a weighted F1-score of 99.08%, outperforming the other fusion methods by 8.08-16.74%. This letter also provides a comparative discussion on the computational complexity of each fusion method. Finally, a series of ablation experiments is conducted to prove the effectiveness of the introduced approach.

The evolution of Beyond 5G (B5G) and 6G networks requires the design of powerful deep neural networks to effectively address the emerging threats and ensure predictive, adaptive security measures that can safeguard the performance and integrity of these advanced systems. Existing studies rely on fully supervised learning settings, lack mechanisms for dynamic adaptation, and perform their experiments on datasets, which do not represent the characteristics of existing 5G scenarios. To address these limitations, we present the first study incorporating hypernetworks into convolutional autoencoders for identifying attacks in 5G datasets collected under realistic conditions. Specifically, the input data is reshaped into a 2D matrix and fed into a convolutional autoencoder, consisting of an encoder and a decoder. At the same time, raw input data is passed through a hypernetwork, which is trained to generate weights for the target network. The target network receives as input the latent representation vector and gives as output the final prediction. Experiments are performed on two real 5G network datasets, namely NANCY and 5G-NIDD. Results show that the proposed deep neural network achieves an Accuracy of 98.51% and 99.91% on the NANCY and 5G-NIDD datasets respectively. Findings of an ablation study demonstrate the effectiveness of the proposed method.

Social media platforms, including X, Facebook, and Instagram, host millions of daily users, giving rise to bots automated programs disseminating misinformation and ideologies with tangible real-world consequences. While bot detection in platform X has been the area of many deep learning models with adequate results, most approaches neglect the graph structure of social media relationships and often rely on hand-engineered architectures. Our work introduces the implementation of a Neural Architecture Search (NAS) technique, namely Deep and Flexible Graph Neural Architecture Search (DFG-NAS), tailored to Relational Graph Convolutional Neural Networks (RGCNs) in the task of bot detection in platform X. Our model constructs a graph that incorporates both the user relationships and their metadata. Then, DFG-NAS is adapted to automatically search for the optimal configuration of Propagation and Transformation functions in the RGCNs. Our experiments are conducted on the TwiBot-20 dataset, constructing a graph with 229,580 nodes and 227,979 edges. We study the five architectures with the highest performance during the search and achieve an accuracy of 85.7%, surpassing state-of-the-art models. Our approach not only addresses the bot detection challenge but also advocates for the broader implementation of NAS models in neural network design automation.

Social media platforms, including X, Facebook, and Instagram, host millions of daily users, giving rise to bots-automated programs disseminating misinformation and ideologies with tangible real-world consequences. While bot detection in platform X has been the area of many deep learning models with adequate results, most approaches neglect the graph structure of social media relationships and often rely on hand-engineered architectures. Our work introduces the implementation of a Neural Architecture Search (NAS) technique, namely Deep and Flexible Graph Neural Architecture Search (DFG-NAS), tailored to Relational Graph Convolutional Neural Networks (RGCNs) in the task of bot detection in platform X. Our model constructs a graph that incorporates both the user relationships and their metadata. Then, DFG-NAS is adapted to automatically search for the optimal configuration of Propagation and Transformation functions in the RGCNs. Our experiments are conducted on the TwiBot-20 dataset, constructing a graph with 229,580 nodes and 227,979 edges. We study the five architectures with the highest performance during the search and achieve an accuracy of 85.7%, surpassing state-of-the-art models. Our approach not only addresses the bot detection challenge but also advocates for the broader implementation of NAS models in neural network design automation.

There exist many research tools for complex brain disorders both proprietary and open source. Those tools are mostly focused on specific use cases and don’t offer a holistic view of the research and diagnostic processes. The MES-CoBraD platform offers a holistic tool that participates in all the phases of the process from planning to execution. In it researchers can create workflows of the diagnostic and research procedures. The Data Analytics module provides the functionality described in these workflows, mainly the analytical functions. The tools are integrated with the rest of the platform in a seamless manner. They are capable of analyzing different categories of data including imaging data, electroencephalogram (EEG)/polysomnography (PSG), medical device data, and consumer technology data. In this paper we present a summary of how the MES-CoBraD platform operates, the role of the data analytics in it, the actual functions implemented and integrated in the Data Analytics module and a number of use cases which show how each specific function can be used during the research and clinical procedures.

The detection of depression in social media posts is crucial due to the increasing prevalence of mental health issues. Traditional machine learning algorithms often fail to capture intricate textual patterns, limiting their effectiveness in identifying depression. Existing studies have explored various approaches to this problem but often fall short in terms of accuracy and robustness. To address these limitations, this research proposes a neural network architecture leveraging transformer-based models combined with metadata and linguistic markers. The study employs DistilBERT, extracting information from the last four layers of the transformer, applying learned weights, and averaging them to create a rich representation of the input text. This representation, augmented by metadata and linguistic markers, enhances the model's comprehension of each post. Dropout layers prevent overfitting, and a Multilayer Perceptron (MLP) is used for final classification. Data augmentation techniques, inspired by the Easy Data Augmentation (EDA) methods, are also employed to improve model performance. Using BERT, random insertion and substitution of phrases generate additional training data, focusing on balancing the dataset by augmenting underrepresented classes. The proposed model achieves weighted Precision, Recall, and F1-scores of 84.26%, 84.18%, and 84.15%, respectively. The augmentation techniques significantly enhance model performance, increasing the weighted F1-score from 72.59% to 84.15%.