Alfred Mertins’s research while affiliated with University of Lübeck and other places

... Recently, a more advanced model-based approach incorporating magnetic anisotropy in equilibrium has been proposed [23], [24]. This extended model has also been applied to the direct Chebyshev reconstruction (DCR), leading to the development of an adapted version (DCR-EQANIS) [4]. ...

Reference:

Efficient Chebyshev Reconstruction for the Anisotropic Equilibrium Model in Magnetic Particle Imaging

... Embodied communication improves the feelings of collaboration with perceived presence compared to text-based communication [56]. Furthermore, multimodal communication, such as combining visuals with verbal interactions, has been found to improve user experience [85]. For instance, the integration of voice, embodied communication, dialogue, and facial expressions provides users with a more accurate perception of AI, improving feelings of collaboration and social presence [71,87]. ...

Reference:

Improving User Experience with FAICO: Towards a Framework for AI Communication in Human-AI Co-Creativity

... This dual focus on time and frequency domains allows for a more nuanced understanding of the physiological processes underlying sleep. Studies such as those by Chen et al. [27] and Phan et al. [36] have demonstrated the effectiveness of TFRs. However, this approach is not without its challenges. ...

Reference:

Temporal focal modulation networks for sleep stage scoring

... This model can be reduced in two ways. First of all, one can assume that the Chebyshev series, which can be derived for 2D Lissajous excitation, is reduced to a single coefficient (reduced EQANIS model), see [37]. Second, one can assume isotropic particles leading to the EQ model. ...

Reference:

Equilibrium Model With Anisotropy for Model-Based Reconstruction in Magnetic Particle Imaging

... Geometric transformations are typically used for data augmentation. For instance, Parbs et al. [118] use random flipping, rotation, and cropping as data augmentation techniques. Also, Gaussian noise is used for slightly deforming the input images. ...

Reference:

Multimodal image registration techniques: a comprehensive survey

... As in the stationary case, the measurement time can be reduced by exploiting prior information, which is done in the form of sparsity in [36], [37], and in the form of a spherical harmonic representation in [38]. With a PSEQ as a transmitted signal, position tracking of the microphone using only the recorded sound can be done [39], leading to minimal cost associated with moving microphone measurements. ...

Reference:

Bayesian Sound Field Estimation Using Moving Microphones

... The authors in [4] argue that multi-label fits the reality better but often multi-class, i.e., only one active sound type per clip, seems to perform better due to learning more class interactions, resulting in a cumbersome handling for practical systems. They propose to handle the problem by creating combinatorial polyphonic classes and use a divide and conquer group splitting to keep the exploding number of classes tractable. ...

Reference:

Multi-label audio classification with a noisy zero-shot teacher

... 16 To address the demands on clinician time and the subjectivity of sleep staging in PSG, machine learning algorithms to automate sleep staging have been developed. These include algorithms developed specifically for children of varying ages, [17][18][19][20][21][22][23][24][25] due to differences in brainwave morphology at different stages of development, 26 and because adult sleep staging algorithms are known to be less performant when applied to children. 27 However, these pediatric-specific sleep staging algorithms are not freely available. ...

Reference:

Evaluation of automated pediatric sleep stage classification using U-Sleep: a convolutional neural network

... CNNs excel at capturing spatial hierarchies in data, making them particularly effective in tasks where fine-grained detail is necessary. This preference is prominently reflected in studies CNN Sequence Many-tomany Zan and Yildiz [11] CNN + RNN + FCN Sequence Many-tomany Phan et al. [38] Transformer TFR Many-tomany FCN refers to fully convolutional network such as those by Phan et al. [26], Chambon et al. [28], and many others listed in Table 1. However, the dynamic nature of sleep, characterized by transitions through various stages over time, necessitates modeling temporal dependencies alongside local features. ...

Reference:

Temporal focal modulation networks for sleep stage scoring

... The reconstruction problem is typically formulated as a system of linear equations and solved using regularized least squares techniques [1], [18], [32]. Direct reconstruction approaches, on the other hand, can be categorized into time-domain methods, commonly known as x-space MPI [8], and frequency-domain methods, such as Chebyshev reconstruction [5], [28]. For one-dimensional MPI excitation, the mathematical equivalence between x-space and Chebyshev reconstruction has been established [10]. ...

Reference:

Efficient Chebyshev Reconstruction for the Anisotropic Equilibrium Model in Magnetic Particle Imaging