DARN: Dynamic Baselines for Real-time Network Monitoring

False arrhythmia alarms pose a major threat to the quality of care in today’s ICU. Thus, the PhysioNet/Computing in Cardiology Challenge 2015 aimed at reducing false alarms by exploiting multimodal cardiac signals recorded by a patient monitor. False alarms for asystole, extreme bradycardia, extreme tachycardia, ventricular flutter/fibrillation as well as ventricular tachycardia were to be reduced using two electrocardiogram channels, up to two cardiac signals of mechanical origin as well as a respiratory signal.In this paper, an approach combining multimodal rhythmicity estimation and machine learning is presented. Using standard short-time autocorrelation and robust beat-to-beat interval estimation, the signal’s self-similarity is analyzed. In particular, beat intervals as well as quality measures are derived which are further quantified using basic mathematical operations (min, mean, max, etc). Moreover, methods from the realm of image processing, 2D Fourier transformation combined with principal component analysis, are employed for dimensionality reduction. Several machine learning approaches are evaluated including linear discriminant analysis and random forest.Using an alarm-independent reduction strategy, an overall false alarm reduction with a score of 65.52 in terms of the real-time scoring system of the challenge is achieved on a hidden dataset. Employing an alarm-specific strategy, an overall real-time score of 78.20 at a true positive rate of 95% and a true negative rate of 78% is achieved. While the results for some categories still need improvement, false alarms for extreme tachycardia are suppressed with 100% sensitivity and specificity.

Fifty-one gynecologic specimens were collected from three women's hospitals and mailed in a prefixed status to our laboratory. The specimens were classified into a negative, a suspicious, a postradiation, and a positive group. After single cell dispersion the samples were stained for DNA and protein, analyzed, and sorted in the dual laser equipped Heidelberg flow analyzer sorter (HEIFAS). Particles with elevated DNA values (beyond 3.5 ploidy) and with intermediate protein values were sorted as the positive fraction directly on microscopic slides. After restaining according to Papanicolaou, they were re-evaluated cytologically and identified as tumor cells, dysplastic cells and false alarms. The latter consist of doublets and aggregates of more than two cells, binucleated cells, sperm aggregates and epithelial cells contaminated with bacteria. The different groups showed significant differences regarding the total rate of aggregates to single cells. In general, false alarms were very frequent in the positive region and impeded the statistical classification of the sample. The reduction of false alarms is a prerequisite for prescreening with flow instrumentation.

This study introduces a self-regulating video surveillance framework that draws upon mathematical modeling to integrate artificial intelligence, machine learning, and computer vision for enhanced automation. The proposed system is designed to operate with minimal human intervention by autonomously processing live video streams, identifying anomalous events, and dynamically adjusting monitoring parameters. A layered computational strategy is employed, where edge devices handle low-latency recognition tasks and cloud platforms manage large-scale training and optimization. This division of workload improves scalability, ensures efficient resource utilization, and maintains real-time responsiveness. The research emphasizes a formal mathematical representation of data flows, learning functions, and optimization routines, providing a structured basis for adaptive monitoring. Privacy-preserving techniques, including federated learning and blockchain integration, are incorporated to address regulatory and ethical considerations, ensuring secure and transparent data handling. Experimental evaluation highlights improvements in detection accuracy, reduction of false alarms, and efficient distribution of computational resources. By presenting a unified mathematical and algorithmic perspective, this work advances the design of intelligent surveillance systems applicable to smart cities, industrial environments, and critical infrastructure protection. The proposed approach not only enhances reliability and adaptability but also establishes a foundation for future extensions involving autonomous learning and decentralized surveillance architectures.

Older adults are more likely than younger adults to falsely recall past episodes that occurred differently or not at all. We examined whether older adults' propensity for false associative memory is related to declines in postretrieval monitoring processes and their modulation with varying memory representations. Younger (N = 20) and older adults (N = 32) studied and relearned unrelated scene-word pairs, followed by a final cued recall that was used to distribute the pairs for an associative recognition test 24 hours later. This procedure allowed individualized formation of rearranged pairs that were made up of elements of pairs that were correctly recalled in the final cued recall ("high-quality" pairs), and of pairs that were not correctly recalled ("low-quality" pairs). Both age groups falsely recognized more low-quality than high-quality rearranged pairs, with a less pronounced reduction in false alarms to high-quality pairs in older adults. In younger adults, cingulo-opercular activity was enhanced for false alarms and for low-quality correct rejections, consistent with its role in postretrieval monitoring. Older adults did not show such modulated recruitment, suggesting deficits in their selective engagement of monitoring processes given variability in the fidelity of memory representations. There were no age differences in hippocampal activity, which was higher for high-quality than low-quality correct rejections in both age groups. These results demonstrate that the engagement of cingulo-opercular monitoring mechanisms varies with memory representation quality and contributes to age-related deficits in false associative memory. (PsycINFO Database Record

A generalized likelihood ratio test (GLRT) statistic for spectral change detection based on the linear chromodynamics model is extended to accommodate unknown residual misregistration between imagery described by a prior probability density function for the spatial misregistration. Using a normal prior distribution leads to a fourth-order polynomial that can be numerically minimized over the unknown misregistration parameters. A more computationally efficient closed-form solution is developed based on a quadratic approximation and provides comparable results to the numerical minimization for the investigated test cases while running 30 times faster. The results applying the method to hyperspectral imagery indicate up to an order of magnitude reduction in false alarms at the same detection rate relative to baseline change detection methods for synthetically misregistered test data particularly in image regions containing edges and fine spatial features. Sensitivity to model parameters is assessed, and the method is compared with a previously published misregistration compensation approach yielding comparable results. Although the GLRT approach appears to exhibit comparable change detection performance, it offers the possibility of tailoring the algorithm to a priori knowledge of expected misregistration errors or to compensate structured misregistration as would occur due to parallax errors due to perspective variations (e.g., image parallax).

Autonomous planetary exploration and hazardous terrain navigation require reliable frameworks for safe landing and adaptive decision-making in uncertain environments. Conventional approaches, including rulebased planning and classical reinforcement learning, provide baseline capabilities but often lack adaptability, generalization, and precision. This research proposes a SegFormer-based Soft Actor-Critic (SAC) framework that integrates semantic segmentation with continuous-control reinforcement learning to enhance hazard perception and policy adaptation. By generating risk and suitability maps from terrain data, the system enables accurate landing site evaluation and efficient navigation. Experimental evaluation demonstrates improved navigation efficiency, energy utilization, and resilience against unknown hazards. Investigational results show a 0.5 % increase in hazard detection, a 0.3 % reduction in false alarms, and a 0.5 % improvement in mission success. The framework reduces latency, improves energy and trajectory efficiency, and enhances robustness, ensuring superior adaptability, reliability, and safety over existing methods.

MIMO radar provides improvement over traditional phased array radars for through wall imaging. By transmitting independent waveforms from a transmit array to a receive array an effective virtual array is created. This array has improved degrees of freedom over phased arrays and mono-static MIMO systems. This virtual array allows us to achieve the same effective aperture length as a phased array with less elements because the virtual array can be described as the convolution of transmit and receive array positions. In addition, data from multiple walls of the same room can be used to collect target information. If two walls are perpendicular to each other and the geometry of transmit and receive arrays is known, then data can be processed independently of each other. Since the geometry of the arrays is known, a target scene can be created where the two data sets overlap. The overlapped scene can then be processed so that image artifacts that do not correlate between the data sets can be excised. The result gives improved target detection, reduction in false alarms, robustness to noise, and robustness against errors such as improperly aligned antennas.

This paper describes the US Army's efforts under the ATR relational template matching program to develop a multi-sensor forward-looking IR (FLIR) and millimter wave (MMW) radar automatic target recognition (ATR) algorithm. The general problem consists of identifying ground targets at low depression angles of less than 6 degrees and ranges of 500 to 6000 m. However, the algorithm is directly applicable to other target sets, including air targets, and ground targets at high depression angles. Our phase I goal was to provide a proof- of-principle demonstration of this new model-based methodology. We did this on simulated FLIR imagery for a three target class problem. A comprehensive test and evaluation was performed at the Night Vision and Electronic Sensors Directorate. Our goal in phase II was to demonstrate this new algorithm approach on realistic field-collected second-generation FLIR imagery with a larger target set consisting of ten ground vehicles. With phase III, our goal was to demonstrate multisensor fusion on FLIR (first generation) fused with laser radar. This was demonstrated on a four-ground-target-class problem. The limitations to first-generation FLIR data and four targets was due purely to restrictions of the field-collected data. Our goal in phase IV is to demonstrate this methodology for FLIR fused with MMW radar data. Phase IV will be a three-year effort, with the first year concentrating on using the MMW radar data for detection, reduction in false alarms, and the provision of accurate range (i.e., scale) to the FLIR algorithm. The goal of the second and third years will be to fuse the MMW data into not only the detection phase of the algorithm but also the recognition and identification portions of the algorithm. This paper presents the general model-based methodology used for the FLIR and the neural network methodology used to provide the FLIR with the MMW detections and scale. Results are provided for the FLIR-only algorithm for short range and moderate range scenarios given an accurate estimate of target range.

<div class="section abstract"><div class="htmlview paragraph">Preventing systematic software failures is of paramount importance for any highly automatic vehicle control system, in particular for safety-critical AUTOSAR software. Among the most critical software defects are runtime errors like buffer overflows or data races. They may cause erroneous or erratic behavior, induce system failures, and constitute security vulnerabilities. Sound static analysis can be used to report all such defects in the code, or to prove their absence. It can also determine dependencies between software components and show freedom of interference without missing any data and control flow through data or function pointers. In the past, AUTOSAR projects often had to be decomposed or simplified to achieve satisfactory analysis time or memory consumption. Creating the analysis model, i.e., determining the tasks and ISRs to analyze, their priorities, synchronization, etc., required significant manual effort. In this article we present novel analysis concepts, developed in the Astrée analyzer, to support fully automatic integration analysis of AUTOSAR projects. The execution model is automatically derived from the ARXML specification. No changes to the software under analysis are required. All specified tasks and ISRs are analyzed assuming fully concurrent execution; runtime errors, data races and deadlocks are reported. Task priorities, core assignment, resources, spinlocks, and critical sections enclosed by primitives to disable/enable interrupts are automatically handled. To minimize false alarms, Astrée supports tuning the analyzer to the software under analysis by locally boosting analysis precision. We report on practical experience with real-life industry projects, giving an overview of alarm rates, analysis time and memory consumption, as well as reduction of false alarms by fine-tuning the analysis precision to the software under analysis.</div></div>

Sepsis is a life-threatening condition, caused by the body’s extreme response to an infection. In the United States, 1.7 million cases of sepsis occur annually, resulting in 265,000 deaths. Delayed diagnosis and treatment are associated with higher mortality rates. An exponential rise in the availability of medical data has allowed for the development of sophisticated machine learning algorithms to predict sepsis earlier than the onset. However, these models often underperform, as the training data are retrospective and do not fully capture the uncertain future. In this study, we develop a novel framework, which we refer to as MLePOMDP, to leverage and combine the underlying, high-level knowledge about sepsis progression and machine learning (ML) for classification. Specifically, we use a hidden Markov model to describe sepsis development at a high level, where the ML model makes the higher-order “observations” from temporal data. Consequently, a partially observable Markov decision process (POMDP) model is developed to make classification decisions. We analytically establish that the optimal policy is of threshold-type, which we exploit to efficiently optimize MLePOMDP. MLePOMDP is calibrated and tested using high-frequency physiological data collected from bedside monitors. Different from past POMDP-based frameworks, MLePOMDP is developed for a prediction task using a very small state definition, produces highly interpretable results, and accounts for a novel and clinically meaningful action space. Our results show that MLePOMDP outperforms machine learning–based benchmarks by up to 8% in precision. Importantly, MLePOMDP is able to reduce false alarms by up to 28%. An additional experiment is conducted to show the generalizability of MLePOMDP to different patient cohorts. Summary of Contribution: This study develops a novel real-time decision support framework for early sepsis prediction by integrating well-known machine learning models (random forest and neural networks) with a well-established sequential decision-making model, namely, a partially observable Markov decision process (POMDP). The structural properties of the optimal policy are further explored and a threshold-type structure is established, which is then leveraged to develop a customized algorithm to solve the problem more efficiently. The resulting framework demonstrates the benefit of applying POMDPs to augment machine learning outputs. Specifically, the framework results in the reduction of false alarms in sepsis predictions where decisions are made in real time, hence improving the overall prediction precision.

Abstract—The recent years have seen the rise of fraud in online banking systems exponentially, and the conventional rule-based mechanisms used to identify attackers are unable to cope with the new, more advanced attack patterns. This paper introduces a smart AI-based model that makes use of hybrid ensemble learning to identify fraudulent transactions with a high level of accuracy and low false-positives. The suggested framework combines machine learning models, deep learning sequence analyzers and ensemble fusion methods to overcome the key challenges such as the serious class imbalance, real-time latency needs, and model explainability demands. We test the framework on benchmark data such as BankSim and anonymised credit card transaction data and find that hybrid ensemble methods demonstrate better performance with an accuracy of over 99.7 per cent, a precision of 99.72 per cent, a recall of 99.78 per cent and an F1-score of 99.77 per cent. Compared to standalone Random Forest, XGBoost, and LSTM architectures, gradient- boosted hybrid models have been found to perform better in all evaluation metrics. The framework contains a learning feedback loop and considers deployment aspects such as latency and claims explainability modules and privacy-preserving design philosophy. We have shown that it is feasible to use intelligent hybrid AI systems to achieve a substantial reduction in false alarms, without a significant loss in detection, which can be used in real-world applications of online banking fraud detection. Keywords: Fraud Detection, Online Banking Security, Ma- chine Learning, Deep Learning, Ensemble Learning, LSTM, Gradient Boosting, Class Imbalance, Real-time Detection, Ex- plainable AI Index Terms—component, formatting, style, styling, insert

Video-based anomaly detection in urban surveillance faces a fundamental challenge: scale-projective ambiguity. This occurs when objects of different physical sizes appear identical in camera images due to perspective projection for example, a child standing 3 meters away may occupy the same number of pixels as an adult standing 5 meters away. This ambiguity causes severe failures in detecting, localizing, and tracking anomalous events. Current methods suffer from three critical limitations: (1) scale-invariant features like SURF discard absolute size information, (2) monocular depth estimation introduces unacceptable errors (>0.5 m at 10 m distance), and (3) existing tracking systems fragment when objects change scale. To address these challenges, we introduce Covariant-Scale, a unified framework that combines classical geometry with modern deep learning. Our approach makes four key contributions: First, we model the space of object transformations (including scale changes) as a curved geometric surface called a Riemannian manifold. This allows us to track how objects naturally evolve through different scales while maintaining their physical properties. Second, we develop a novel deep learning architecture a Variational Autoencoder (VAE) with geometric constraints that learns to separate an object’s appearance from its scale. This separation is enforced through a mathematical property called Ricci curvature, which ensures scale information remains consistent regardless of the object’s distance from the camera. Third, we integrate LiDAR depth sensors with cameras through physics-based principles. We mathematically prove that LiDAR reduces scale estimation errors by 10,000× compared to camera-only methods at 10-meter distances a fundamental limit we derive using information theory (Cramér-Rao bound). Fourth, we develop a tracking system that preserves the physics of motion by converting 2D pixel movements into true 3D velocities, reducing identity confusion by 41% during scale transitions. Evaluated on three benchmark datasets (UCSD, ShanghaiTech, Avenue) under extreme scale variations (>15× zoom range), Covariant-Scale achieves: 98.2% detection accuracy (22.1% improvement over state-of-the-art), 76% reduction in false alarms in crowded scenes, and real-time performance (34 ms per frame) on embedded hardware. This work establishes a new paradigm for video analytics that bridges theoretical geometry with practical computer vision, resolving a 20-year challenge in safety-critical surveillance systems.

The Optimized Deep Feature Analysis (ODFA) framework targets botnet attacks in IoT networks by combining deep learning with machine learning techniques. It uses a hybrid CNN-LSTM model to extract spatial and temporal features from network traffic, applying Recursive Feature Elimination (RFE) with Cross-Validation (CV) for refining feature selection. A Random Forest classifier is then employed to classify different types of botnet attacks. Tests conducted on the UNSW-NB15 dataset demonstrate its ability to achieve high precision, sensitivity, specificity, and accuracy. Recurrent feature optimization allows the framework to adapt to changes in attack patterns, ensuring consistent detection performance. The reduction in false alarms and improved classification of attack types highlights its contribution to securing IoT networks.

Internet videoconferencing has emerged as a viable medium for communication and entertainment. However, its widespread use is being challenged. This is because videoconference end-users frequently experience perceptual quality impairments such as video frame freezing and voice dropouts due to changes in network conditions on the Internet. These impairments cause extra end-user interaction effortand correspondingly lead to unwanted network bandwidth consumptionthat affects user Quality of Experience (QoE) and Internet congestion. Hence, it is important to measure and subsequently minimize the extra end-user interaction effort in a videoconferencing system. In this paper, we describe a novel active measurement scheme that considers end-user interaction effort and the corresponding network bandwidth consumption to provide videoconferencing interaction QoE measurements. The scheme involves a Multi-Activity Packet-Trains (MAPTs) methodology to dynamically emulate a videoconference session's participant interaction patterns and corresponding video activity levels that are affected by transient changes in network conditions. Also, we describe the implementation and validation of the Vperftool we have developed to measure the videoconferencing interaction QoE on a network path using our proposed scheme.

This paper presents an algorithm for reducing false alarms related to changes in arterial blood pressure (ABP) in intensive care unit (ICU) monitoring. The algorithm assesses the ABP signal quality, analyzes the ECG-ABP relationships using a fuzzy logic approach, and post-processes (accepts or rejects) ABP alarms produced by a commercial monitor. The algorithm was developed and evaluated using disjoint sets of data from the MIMIC database. By rejecting 96.8% of the false ABP alarms (182 of 188) produced by the monitor using the test set of data, the algorithm was able to reduce the false alarm rate from 34.9% to 1.7% of ABP alarms while accepting 99.4% of true ABP alarms (346 of 348). The results show that the algorithm is both effective and practical, and feasible for use in future monitoring systems.