Pandemic management requires reliable and efficient dynamical simulation to predict and control disease spreading. The COVID-19 (SARS-CoV-2) pandemic is mitigated by several non-pharmaceutical interventions, but it is hard to predict which of these are the most effective for a given population. We developed the computationally effective and scalable, agent-based microsimulation frameworkPanSim, allowing us to test control measures in multiple infection waves caused by the spread of a new virus variant in a city-sized societal environment using a unified framework fitted to realistic data. We show that vaccination strategies prioritising occupational risk groups minimise the number of infections but allow higher mortality while prioritising vulnerable groups minimises mortality but implies an increased infection rate. We also found that intensive vaccination along with non-pharmaceutical interventions can substantially suppress the spread of the virus, while low levels of vaccination, premature reopening may easily revert the epidemic to an uncontrolled state. Our analysis highlights that while vaccination protects the elderly from COVID-19, a large percentage of children will contract the virus, and we also show the benefits and limitations of various quarantine and testing scenarios. The uniquely detailed spatio-temporal resolution ofPanSimallows the design and testing of complex, specifically targeted interventions with a large number of agents under dynamically changing conditions.

Wastewater-based epidemiology (WBE) has emerged as a promising approach for monitoring the COVID-19 pandemic, since the measurement process is cost-effective and is exposed to fewer potential errors compared to other indicators like hospitalization data or the number of detected cases. Consequently, WBE was gradually becoming a key tool for epidemic surveillance and often the most reliable data source, as the intensity of clinical testing for COVID-19 drastically decreased by the third year of the pandemic. Recent results suggests that the model-based fusion of wastewater measurements with clinical data and other indicators is essential in future epidemic surveillance. In this work, we developed a wastewater-based compartmental epidemic model with a two-phase vaccination dynamics and immune evasion. We proposed a multi-step optimization-based data assimilation method for epidemic state reconstruction, parameter estimation, and prediction. The computations make use of the measured viral load in wastewater, the available clinical data (hospital occupancy, delivered vaccine doses, and deaths), the stringency index of the official social distancing rules, and other measures. The current state assessment and the estimation of the current transmission rate and immunity loss allow a plausible prediction of the future progression of the pandemic. Qualitative and quantitative evaluations revealed that the contribution of wastewater data in our computational epidemiological framework makes predictions more reliable. Predictions suggest that at least half of the Hungarian population has lost immunity during the epidemic outbreak caused by the BA.1 and BA.2 subvariants of Omicron in the first half of 2022. We obtained a similar result for the outbreaks caused by the subvariant BA.5 in the second half of 2022. The proposed approach has been used to support COVID management in Hungary and could be customized for other countries as well.

Full utilization of renewable energy resources is a difficult task due to the constantly changing demand-side load of the electrical grid. Demand-side management would solve this crucial problem. To enable demand-side management, knowledge about the composition of the grid load is required, as well as the ability to schedule individual loads. There are proposed Smart Plugs presented in the literature capable of such tasks. The problem, however, is that these methods lack the ability to detect if a previously unseen electrical load is connected. Misclassification of such loads presents a problem for load estimation and scheduling. Open-set recognition methods solve this problem by providing a way to detect samples not belonging to any class used during the training of the classifier. This paper evaluates the novel application of open-set recognition methods to the problem of load classification. Two approaches were examined, and both offer promising results. A Support Vector Machine based approach was first evaluated. The second, more robust method used a modified OpenMax-based algorithm to detect unseen loads.

In recent years, Artificial Intelligence for IT Operations (AIOps) has gained popularity as a solution to various challenges in IT operations, particularly in anomaly detection. Although numerous studies have focused on anomaly detection, they often overlook cloud-based systems and the vast amount of data they generate. Moreover, the practical application of these results and managing multiple IT systems within diverse computing environments present notable challenges. In this paper, we explore anomaly detection in real-world environments, including legacy systems and cloud platforms, by evaluating existing methods and explicitly introducing four novel anomaly detection algorithms: Llama Model-based Log Anomaly Detection (LMLAD), Log Clustering with Cosine Similarity for Log Anomaly Detection (LCCLAD), Convolutional Auto-Encoder based Log Anomaly Detection (CAELAD), and Statistical Clustering for Log Anomaly Detection (SCLAD). These algorithms vary in complexity and accuracy, and offer tunable performance based on the specific requirements of the environment. In experiments on four log datasets (two private and two public), our best models improved F1-scores by 6–10% over state-of-the-art baselines, demonstrating the efficacy of our approach. Through empirical analysis, we demonstrate that while advanced techniques such as CAELAD often deliver high accuracy, simpler methods such as SCLAD can be equally effective in practical settings, depending on the complexity of the problem. Our results underscore the importance of selecting the right balance between sophistication and simplicity, challenging the assumption that the most sophisticated methods are necessary for effective anomaly detection in real-world log data.

Conventional classifiers are generally unable to identify samples from classes absent during the model’s training. However, such samples frequently emerge in real-world scenarios, necessitating the extension of classifier capabilities. Open-Set Recognition (OSR) models are designed to address this challenge. Previously, we developed a robust OSR method that employs generated—“fake”—features to model the space of unknown classes encountered during deployment. Like most OSR models, this method was initially designed for image datasets. However, it is essential to extend OSR techniques to other data types, given their widespread use in practice. In this work, we adapt our model to time-series data while preserving its core efficiency advantage. Thanks to the model’s modular design, only the feature extraction component required modification. We implemented three approaches: a one-dimensional convolutional network for accurate representation, a lightweight method based on predefined statistical features, and a frequency-domain neural network. Further, we evaluated combinations of these methods. Experiments on a biometric time-series dataset, used here as a case study, demonstrate that our model achieves excellent open-set detection and closed-set accuracy. Combining feature extraction strategies yields the best performance, while individual methods offer flexibility: CNNs deliver high accuracy, whereas handcrafted features enable resource-efficient deployment. This adaptability makes the proposed framework suitable for scenarios with varying computational constraints.