Open Access

Neuromuscular training to strengthen leg muscles is an important part of the treatment of musculoskeletal disorders and chronic diseases and preventing age–related muscle loss. This study evaluates different individualization approaches and their real–time implementation for OpenSim musculoskeletal models to estimate the external knee adduction moment during a leg–press exercise. A robotic neuromuscular training platform was utilized to perform isometric and dynamic leg extension exercises. Data were collected for 13 subjects using a 3D motion capture system and force plate measurements from the robotic training platform. Functional joint parameters, determined through dynamic reference movements, were integrated into the OpenSim models, allowing a personalized representation of the hip, knee, and ankle joints. This integration was compared with a conventional scaling method. The results indicate that the incorporation of functional joint axes can significantly enhance the accuracy of biomechanical simulations. These methods provide a real–time and a more precise estimate of the external knee adduction moment compared to conventional scaling approaches and underscore the importance of individualized model parameters in biomechanical research.

This paper introduces a novel application of functional neural networks (FNNs) in the domain of electroencephalography-based (EEG-based) brain-computer interfaces (BCIs), targeting self-paced motor execution (ME) and motor imagery (MI). FNNs represent a neural network architecture tailored to smooth processes, and as such have been applied to EEG data classification recently. The paper proposes a comprehensive pipeline encompassing data acquisition, synchronization, pre-processing, training of FNNs, and real-time inference to enable the seamless integration of FNNs into real-world BCI applications. For the first time, FNNs are integrated into an end-to-end pipeline and serve for live inference outside a strict laboratory setting. In pursuit of a more accessible electroencephalography (EEG) artificial intelligence (AI) training scenario, the paper introduces a self-paced environment for auto-labeling EEG data. A customdesigned Pong game serves as the training task and enables subjects to engage in MI/ME tasks while receiving immediate visual feedback. To automate the labeling process of the recorded EEG data, the movements of both arms are captured with inertial measurement units (IMUs) and analyzed through gesture recognition. This novel training framework contributes to more natural and engaging data collection and reduces pre-processing for model training. To provide a comprehensive evaluation, the paper compares the performance of FNN and EEGNet in the self-paced MI/ME tasks. The comparative analysis addresses factors such as classification accuracy, real-time processing speed, and power consumption. Furthermore, the study explores various auto-labeling methods within the self-paced environment, analyzing their impact on the classification performances of both architectures. By evaluating these labeling methods, this work addresses the challenge of accurate and efficient EEG data labeling, crucial for training robust models for prediction of time critical events. The proposed pipeline and experimental design culminate in a full-scale evaluation of the FNN-based classification system to demonstrate its efficacy in real-time MI/ME tasks. The paper’s contributions not only establish FNNs as a potent tool for EEG classification but also provide valuable insights into enhancing the accessibility, usability, and performance of EEG-based AI systems in real-world applications.

In the common time series model Xi,n = μ(i/n) + εi,n with non-stationary errors we consider the problem of detecting a significant deviation of the mean function μ from a benchmark g(μ) (such as the initial value μ(0) or the average trend f 1 0 μ(t)dt). The problem is motivated by a more realistic modelling of change point analysis, where one is interested in identifying relevant deviations in a smoothly varying sequence of means (μ(i/n))i=1,...,n and cannot assume that the sequence is piecewise constant. A test for this type of hypotheses is developed using an appropriate estimator for the integrated squared deviation of the mean function and the threshold. By a new concept of self-normalization adapted to nonstationary processes an asymptotically pivotal test for the hypothesis of a relevant deviation is constructed. The results are illustrated by means of a simulation study and a data example.

5G cellular networks are particularly vulnerable against narrowband jammers that target specific control sub-channels in the radio signal. One mitigation approach is to detect such jamming attacks with an online observation system, based on machine learning (ML). We propose to detect jamming at the physical layer with an ML model that performs binary classification. Based on data from an experimental 5G network, we study the performance of different classification models. A convolutional neural network will be compared to support vector machines and k-nearest neighbors, where the last two methods are combined with principal component analysis. The obtained results show substantial differences in terms of classification accuracy and computation time.

Das Entdecken struktureller Veränderungen in Zeitreihen ist eine der wichtigsten Herausforderungen der modernen Statistik, da nur so eine zuverlässige statistische Inferenz gewährleistet werden kann. Nach einer kurzen Einleitung in die mathematischen Grundlagen, werden in dieser Arbeit drei Testverfahren entwickelt, mit denen Zeitreihen auf Stationarität untersucht werden können. Der erste Test basiert auf einer CUSUM-Statistik und testet die Annahme der schwachen Stationarität funktionaler Zeitreihen gegen die Alternative gradueller Veränderungen. Die anderen beiden Tests erkennen relevante Abweichungen in Erwartungswerten reeller Zeitreihen, die sich stetig über die Zeit entwickeln dürfen. Der eine dieser beiden Tests definiert eine Abweichung als relevant, falls die maximale Abweichung von einem Vergleichswert groß ist, wohingegen der andere Test den L2 Abstand betrachtet.