Electronic health records (EHR) are patient-level information, e.g., laboratory tests and questionnaires, stored in electronic format. Compared to physical records, the EHR alternative allows patients to access their data easily and helps staff with management procedural tasks such as information sharing across different organizations. Moreover, this type of data is commonly used by researchers for predictive and classification purposes, employing statistical and machine learning methods. However, missingness is a phenomenon that is observed very frequently for such measurements. Even though this missingness is often significant, it is usually treated poorly with either case deletion or simple methods, resulting in suboptimal and/or inaccurate predictive results. This happens because the simple methods, e.g., k-nearest neighbors (kNN) and mean/mode imputation, fail in most cases to incorporate the complex relationships that define these medical datasets. To address these limitations, in this paper we test and improve state-of-the-art missing data imputation models and practices. We propose a new missing value imputation method based on denoising autoencoders (DAE) with kNN for the pre-imputation task. We optimize the training methodology by re-applying kNN to the missing data every $N$ epochs using a different value for the variable $k$ each time to yield more accurate results. We also revise a state-of-the-art missing data imputation approach based on a generative adversarial network (GAN). Using this as a baseline, we introduce improvements regarding both the architecture and the training procedure. These models are compared with the ones usually employed within clinical research studies for both the task of imputation and post-imputation prediction. Results show that our proposed deep learning approaches outperform the standard baselines, yielding better imputation and predictive results.

The advent of computer networks and the internet has drastically altered the means by which we share information and interact with each other. However, this technological advancement has also created opportunities for malevolent behavior, with individuals exploiting vulnerabilities to gain access to confidential data, obstruct activity, etc. To this end, intrusion detection systems (IDSs) are needed to filter malicious traffic and prevent common attacks. In the past, these systems relied on a fixed set of rules or comparisons with previous attacks. However, with the increased availability of computational power and data, machine learning has emerged as a promising solution for this task. While many systems now use this methodology in real-time for a reactive approach to mitigation, we explore the potential of configuring it as a proactive time series prediction. In this work, we delve into this possibility further. More specifically, we convert a classic IDS dataset to a time series format and use predictive models to forecast forthcoming malign packets. We propose a new architecture combining convolutional neural networks, long short-term memory networks, and attention. The findings indicate that our model performs strongly, exhibiting an F1 score and AUC that are within margins of 1% and 3%, respectively, when compared to conventional real-time detection. Also, our architecture achieves an ∼8% F1 score improvement compared to an LSTM (long short-term memory) model.

Federated learning (FL) is an emerging machine learning technique where machine learning models are trained in a decentralized manner. The main advantage of this approach is the data privacy it provides because the data are not processed in a centralized device. Moreover, the local client models are aggregated on a server, resulting in a global model that has accumulated knowledge from all the different clients. This approach, however, is vulnerable to attacks because clients can be malicious or malicious actors may interfere within the network. In the first case, these types of attacks may refer to data or model poisoning attacks where the data or model parameters, respectively, may be altered. In this paper, we investigate the data poisoning attacks and, more specifically, the label-flipping case within a federated learning system. For an image classification task, we introduce two variants of data poisoning attacks, namely model degradation and targeted label attacks. These attacks are based on synthetic images generated by a generative adversarial network (GAN). This network is trained jointly by the malicious clients using a concatenated malicious dataset. Due to dataset sample limitations, the architecture and learning procedure of the GAN are adjusted accordingly. Through the experiments, we demonstrate that these types of attacks are effective in achieving their task and managing to fool common federated defenses (stealth). We also propose a mechanism to mitigate these attacks based on clean label training on the server side. In more detail, we see that the model degradation attack causes an accuracy degradation of up to 25%, while common defenses can only alleviate this for a percentage of ∼5%. Similarly, the targeted label attack results in a misclassification of 56% compared to 2.5% when no attack takes place. Moreover, our proposed defense mechanism is able to mitigate these attacks.

Neural Style Transfer (NST) is a popular technique of computer vision where the content of an image is blended with the style of another, which results in a fused image with certain properties of both original images. This approach has practical applications in various domains and has garnered significant attention in both industry and academia. An interesting application of this technique is segmented style transfer where a segmentation algorithm is used to locate objects within an image and then the style transfer method is performed locally, producing images with different styles for different objects. This approach opens up possibilities for creating visually striking compositions by seamlessly blending various artistic styles onto specific objects within an image, allowing for a new level of creative expression. This paper proposes a novel method that combines Segment Anything Model (SAM), a state-of-the-art vision transformer-based image segmentation model developed by Facebook, with style transfer. Our approach includes performing localized style transfer in selected segmentation regions of an image using classical style transfer algorithms. To ensure smooth transitions between the stylized and non-stylized border we also develop our loss function with a border smoothing technique. Experimental results demonstrate the robustness and effectiveness of the proposed methodology, including the ability to infuse multiple artistic styles into different objects within an image. The contributions of this work include integrating SAM with style transfer, proposing a novel loss function, evaluating the segmented style transfer in multiple content regions, comparing with state-of-the-art approaches, and experimenting with multiple style images for diverse stylization. Our primary focus centers on creating a model that serves as a digital painter across a wide range of image genres and artistic styles.

Vulnerability identification and assessment is a key process in risk management. While enumerations of vulnerabilities are available, it is challenging to identify vulnerability sets focused on the profiles and roles of specific organizations. To this end, we have employed systematized knowledge and relevant standards (including National Electric Sector Cybersecurity Organization Resource (NESCOR), ISO/IEC 27005:2018 and National Vulnerability Database (NVD)) to identify a set of 250 vulnerabilities for operators of energy-related critical infrastructures. We have elaborated a “double-mapping” scheme to associate (arbitrarily) categorized assets, with the pool of identified Physical, Cyber and Human/Organizational vulnerabilities. We have designed and implemented an extensible vulnerability identification and assessment framework, allowing historized assessments, based on the CVSS (Common Vulnerability Scoring System) scoring mechanism. This framework has been extended to allow modelling of the vulnerabilities and assessments using the Structured Threat Information eXpression (STIX) JSON format, as Cyber Threat Intelligence (CTI) information, to facilitate information sharing between Electrical Power and Energy Systems (EPES) and to promote collaboration and interoperability scenarios. Vulnerability assessments from the initial analysis of the project in the context of Research and Technology Development (RTD) projects have been statistically processed, offering insights in terms of the assessment’s importance and distribution. The assessments have also been transformed into a dynamic dataset processed to identify and quantify correlation and start the discussion on the interpretation of the way assessments are performed.

Federated Learning is identified as a reliable technique for distributed training of ML models. Specifically, a set of dispersed nodes may collaborate through a federation in producing a jointly trained ML model without disclosing their data to each other. Each node performs local model training and then shares its trained model weights with a server node, usually called Aggregator in federated learning, as it aggregates the trained weights and then sends them back to its clients for another round of local training. Despite the data protection and security that FL provides to each client, there are still well-studied attacks such as membership inference attacks that can detect potential vulnerabilities of the FL system and thus expose sensitive data. In this paper, in order to prevent this kind of attack and address private data leakage, we introduce FREDY, a differential private federated learning framework that enables knowledge transfer from private data. Particularly, our approach has a teachers–student scheme. Each teacher model is trained on sensitive, disjoint data in a federated manner, and the student model is trained on the most voted predictions of the teachers on public unlabeled data which are noisy aggregated in order to guarantee the privacy of each teacher’s sensitive data. Only the student model is publicly accessible as the teacher models contain sensitive information. We show that our proposed approach guarantees the privacy of sensitive data against model inference attacks while it combines the federated learning settings for the model training procedures.