Showing 1–50 of 111 results for author: Maass, P

Data-driven approaches for electrical impedance tomography image segmentation from partial boundary data

Authors: Alexander Denker, Zeljko Kereta, Imraj Singh, Tom Freudenberg, Tobias Kluth, Peter Maass, Simon Arridge

Abstract: Electrical impedance tomography (EIT) plays a crucial role in non-invasive imaging, with both medical and industrial applications. In this paper, we present three data-driven reconstruction methods for EIT imaging. These three approaches were originally submitted to the Kuopio tomography challenge 2023 (KTC2023). First, we introduce a post-processing approach, which achieved first place at KTC2023… ▽ More Electrical impedance tomography (EIT) plays a crucial role in non-invasive imaging, with both medical and industrial applications. In this paper, we present three data-driven reconstruction methods for EIT imaging. These three approaches were originally submitted to the Kuopio tomography challenge 2023 (KTC2023). First, we introduce a post-processing approach, which achieved first place at KTC2023. Further, we present a fully learned and a conditional diffusion approach. All three methods are based on a similar neural network as a backbone and were trained using a synthetically generated data set, providing with an opportunity for a fair comparison of these different data-driven reconstruction methods. △ Less

Submitted 6 May, 2024; originally announced July 2024.

MSC Class: 94A08 (Primary) 68T45; 68T10 (Secondary) ACM Class: G.1.8; I.4.5

Fast Brownian cluster dynamics

Authors: Alexander P. Antonov, Sören Schweers, Artem Ryabov, Philipp Maass

Abstract: We present an efficient method to perform overdamped Brownian dynamics simulations in external force fields and for particle interactions that include a hardcore part. The method applies to particle motion in one dimension, where it is possible to update particle positions by repositioning particle clusters as a whole. These clusters consist of several particles in contact. They form because parti… ▽ More We present an efficient method to perform overdamped Brownian dynamics simulations in external force fields and for particle interactions that include a hardcore part. The method applies to particle motion in one dimension, where it is possible to update particle positions by repositioning particle clusters as a whole. These clusters consist of several particles in contact. They form because particle collisions are treated as completely inelastic rather than elastic ones. Updating of cluster positions in time steps is carried out by cluster fragmentation and merging procedures. The presented method is particularly powerful at high collision rates in densely crowded systems, where collective movements of particle assemblies is governing the dynamics. As an application, we simulate the single-file diffusion of sticky hard spheres in a periodic potential. △ Less

Submitted 7 June, 2024; originally announced June 2024.

Comments: 15 pages, 7 figures

Visualize and Paint GAN Activations

Authors: Rudolf Herdt, Peter Maass

Abstract: We investigate how generated structures of GANs correlate with their activations in hidden layers, with the purpose of better understanding the inner workings of those models and being able to paint structures with unconditionally trained GANs. This gives us more control over the generated images, allowing to generate them from a semantic segmentation map while not requiring such a segmentation in… ▽ More We investigate how generated structures of GANs correlate with their activations in hidden layers, with the purpose of better understanding the inner workings of those models and being able to paint structures with unconditionally trained GANs. This gives us more control over the generated images, allowing to generate them from a semantic segmentation map while not requiring such a segmentation in the training data. To this end we introduce the concept of tileable features, allowing us to identify activations that work well for painting. △ Less

Submitted 25 June, 2024; v1 submitted 24 May, 2024; originally announced May 2024.

Enhancing the analysis of murine neonatal ultrasonic vocalizations: Development, evaluation, and application of different mathematical models

Authors: Rudolf Herdt, Louisa Kinzel, Johann Georg Maaß, Marvin Walther, Henning Fröhlich, Tim Schubert, Peter Maass, Christian Patrick Schaaf

Abstract: Rodents employ a broad spectrum of ultrasonic vocalizations (USVs) for social communication. As these vocalizations offer valuable insights into affective states, social interactions, and developmental stages of animals, various deep learning approaches have aimed to automate both the quantitative (detection) and qualitative (classification) analysis of USVs. Here, we present the first systematic… ▽ More Rodents employ a broad spectrum of ultrasonic vocalizations (USVs) for social communication. As these vocalizations offer valuable insights into affective states, social interactions, and developmental stages of animals, various deep learning approaches have aimed to automate both the quantitative (detection) and qualitative (classification) analysis of USVs. Here, we present the first systematic evaluation of different types of neural networks for USV classification. We assessed various feedforward networks, including a custom-built, fully-connected network and convolutional neural network, different residual neural networks (ResNets), an EfficientNet, and a Vision Transformer (ViT). Paired with a refined, entropy-based detection algorithm (achieving recall of 94.9% and precision of 99.3%), the best architecture (achieving 86.79% accuracy) was integrated into a fully automated pipeline capable of analyzing extensive USV datasets with high reliability. Additionally, users can specify an individual minimum accuracy threshold based on their research needs. In this semi-automated setup, the pipeline selectively classifies calls with high pseudo-probability, leaving the rest for manual inspection. Our study focuses exclusively on neonatal USVs. As part of an ongoing phenoty** study, our pipeline has proven to be a valuable tool for identifying key differences in USVs produced by mice with autism-like behaviors. △ Less

Submitted 27 May, 2024; v1 submitted 17 May, 2024; originally announced May 2024.

Smooth Deep Saliency

Authors: Rudolf Herdt, Maximilian Schmidt, Daniel Otero Baguer, Peter Maaß

Abstract: In this work, we investigate methods to reduce the noise in deep saliency maps coming from convolutional downsampling, with the purpose of explaining how a deep learning model detects tumors in scanned histological tissue samples. Those methods make the investigated models more interpretable for gradient-based saliency maps, computed in hidden layers. We test our approach on different models train… ▽ More In this work, we investigate methods to reduce the noise in deep saliency maps coming from convolutional downsampling, with the purpose of explaining how a deep learning model detects tumors in scanned histological tissue samples. Those methods make the investigated models more interpretable for gradient-based saliency maps, computed in hidden layers. We test our approach on different models trained for image classification on ImageNet1K, and models trained for tumor detection on Camelyon16 and in-house real-world digital pathology scans of stained tissue samples. Our results show that the checkerboard noise in the gradient gets reduced, resulting in smoother and therefore easier to interpret saliency maps. △ Less

Submitted 4 April, 2024; v1 submitted 2 April, 2024; originally announced April 2024.

Solitary cluster waves in periodic potentials: Formation, propagation, and soliton-mediated particle transport

Authors: Alexander P. Antonov, Artem Ryabov, Philipp Maass

Abstract: Transport processes in crowded periodic structures are often mediated by cooperative movements of particles forming clusters. Recent theoretical and experimental studies of driven Brownian motion of hard spheres showed that cluster-mediated transport in one-dimensional periodic potentials can proceed in form of solitary waves. We here give a comprehensive description of these solitons. Fundamental… ▽ More Transport processes in crowded periodic structures are often mediated by cooperative movements of particles forming clusters. Recent theoretical and experimental studies of driven Brownian motion of hard spheres showed that cluster-mediated transport in one-dimensional periodic potentials can proceed in form of solitary waves. We here give a comprehensive description of these solitons. Fundamental for our analysis is a static presoliton state, which is formed by a periodic arrangement of basic stable clusters. Their size follows from a geometric principle of minimum free space. Adding one particle to the presoliton state gives rise to solitons. We derive the minimal number of particles needed for soliton formation, number of solitons at larger particle numbers, soliton velocities and soliton-mediated particle currents. Incomplete relaxations of the basic clusters are responsible for an effective repulsive soliton-soliton interaction seen in measurements. A dynamical phase transition is predicted to occur in current-density relations at low temperatures. Our results provide a theoretical basis for describing experiments on cluster-mediated particle transport in periodic potentials. △ Less

Submitted 7 June, 2024; v1 submitted 27 February, 2024; originally announced February 2024.

Comments: 21 pages, 12 figures

Journal ref: Chaos, Solitons & Fractals 132, 115079 (2024)

Electrical Impedance Tomography: A Fair Comparative Study on Deep Learning and Analytic-based Approaches

Authors: Derick Nganyu Tanyu, Jianfeng Ning, Andreas Hauptmann, Bangti **, Peter Maass

Abstract: Electrical Impedance Tomography (EIT) is a powerful imaging technique with diverse applications, e.g., medical diagnosis, industrial monitoring, and environmental studies. The EIT inverse problem is about inferring the internal conductivity distribution of an object from measurements taken on its boundary. It is severely ill-posed, necessitating advanced computational methods for accurate image re… ▽ More Electrical Impedance Tomography (EIT) is a powerful imaging technique with diverse applications, e.g., medical diagnosis, industrial monitoring, and environmental studies. The EIT inverse problem is about inferring the internal conductivity distribution of an object from measurements taken on its boundary. It is severely ill-posed, necessitating advanced computational methods for accurate image reconstructions. Recent years have witnessed significant progress, driven by innovations in analytic-based approaches and deep learning. This review explores techniques for solving the EIT inverse problem, focusing on the interplay between contemporary deep learning-based strategies and classical analytic-based methods. Four state-of-the-art deep learning algorithms are rigorously examined, harnessing the representational capabilities of deep neural networks to reconstruct intricate conductivity distributions. In parallel, two analytic-based methods, rooted in mathematical formulations and regularisation techniques, are dissected for their strengths and limitations. These methodologies are evaluated through various numerical experiments, encompassing diverse scenarios that reflect real-world complexities. A suite of performance metrics is employed to assess the efficacy of these methods. These metrics collectively provide a nuanced understanding of the methods' ability to capture essential features and delineate complex conductivity patterns. One novel feature of the study is the incorporation of variable conductivity scenarios, introducing a level of heterogeneity that mimics textured inclusions. This departure from uniform conductivity assumptions mimics realistic scenarios where tissues or materials exhibit spatially varying electrical properties. Exploring how each method responds to such variable conductivity scenarios opens avenues for understanding their robustness and adaptability. △ Less

Submitted 28 October, 2023; originally announced October 2023.

Weak pinning and long-range anticorrelated motion of phase boundaries in driven diffusive systems

Authors: Sören Schweers, David F. Locher, Gunter M. Schütz, Philipp Maass

Abstract: We show that domain walls separating coexisting extremal current phases in driven diffusive systems exhibit complex stochastic dynamics, with a subdiffusive temporal growth of position fluctuations due to long-range anticorrelated current fluctuations and a weak pinning at long times. This weak pinning manifests itself in a saturated width of the domain wall position fluctuations that increases su… ▽ More We show that domain walls separating coexisting extremal current phases in driven diffusive systems exhibit complex stochastic dynamics, with a subdiffusive temporal growth of position fluctuations due to long-range anticorrelated current fluctuations and a weak pinning at long times. This weak pinning manifests itself in a saturated width of the domain wall position fluctuations that increases sublinearly with the system size. As a function of time $t$ and system size $L$, the width $w(t,L)$ exhibits a scaling behavior $w(t,L)=L^{3/4}f(t/L^{9/4})$, with $f(u)$ constant for $u\gg1$ and $f(u)\sim u^{1/3}$ for $u\ll1$. An Orstein-Uhlenbeck process with long-range anticorrelated noise is shown to capture this scaling behavior. Results for the drift coefficient of the domain wall motion point to memory effects in its dynamics. △ Less

Submitted 19 October, 2023; originally announced October 2023.

Comments: 6 pages, 4 figures plus 6 pages supplemental material with 3 figures

Steerable Conditional Diffusion for Out-of-Distribution Adaptation in Imaging Inverse Problems

Authors: Riccardo Barbano, Alexander Denker, Hyung** Chung, Tae Hoon Roh, Simon Arrdige, Peter Maass, Bangti **, Jong Chul Ye

Abstract: Denoising diffusion models have emerged as the go-to framework for solving inverse problems in imaging. A critical concern regarding these models is their performance on out-of-distribution (OOD) tasks, which remains an under-explored challenge. Realistic reconstructions inconsistent with the measured data can be generated, hallucinating image features that are uniquely present in the training dat… ▽ More Denoising diffusion models have emerged as the go-to framework for solving inverse problems in imaging. A critical concern regarding these models is their performance on out-of-distribution (OOD) tasks, which remains an under-explored challenge. Realistic reconstructions inconsistent with the measured data can be generated, hallucinating image features that are uniquely present in the training dataset. To simultaneously enforce data-consistency and leverage data-driven priors, we introduce a novel sampling framework called Steerable Conditional Diffusion. This framework adapts the denoising network specifically to the available measured data. Utilising our proposed method, we achieve substantial enhancements in OOD performance across diverse imaging modalities, advancing the robust deployment of denoising diffusion models in real-world applications. △ Less

Submitted 28 August, 2023; originally announced August 2023.

Score-Based Generative Models for PET Image Reconstruction

Authors: Imraj RD Singh, Alexander Denker, Riccardo Barbano, Željko Kereta, Bangti **, Kris Thielemans, Peter Maass, Simon Arridge

Abstract: Score-based generative models have demonstrated highly promising results for medical image reconstruction tasks in magnetic resonance imaging or computed tomography. However, their application to Positron Emission Tomography (PET) is still largely unexplored. PET image reconstruction involves a variety of challenges, including Poisson noise with high variance and a wide dynamic range. To address t… ▽ More Score-based generative models have demonstrated highly promising results for medical image reconstruction tasks in magnetic resonance imaging or computed tomography. However, their application to Positron Emission Tomography (PET) is still largely unexplored. PET image reconstruction involves a variety of challenges, including Poisson noise with high variance and a wide dynamic range. To address these challenges, we propose several PET-specific adaptations of score-based generative models. The proposed framework is developed for both 2D and 3D PET. In addition, we provide an extension to guided reconstruction using magnetic resonance images. We validate the approach through extensive 2D and 3D $\textit{in-silico}$ experiments with a model trained on patient-realistic data without lesions, and evaluate on data without lesions as well as out-of-distribution data with lesions. This demonstrates the proposed method's robustness and significant potential for improved PET reconstruction. △ Less

Submitted 23 January, 2024; v1 submitted 27 August, 2023; originally announced August 2023.

Comments: Accepted for publication at the Journal of Machine Learning for Biomedical Imaging (MELBA) https://melba-journal.org/2024:001

MSC Class: 15A29; 45Q05 ACM Class: I.4.9; J.2; I.2.1

Journal ref: Machine.Learning.for.Biomedical.Imaging. 2 (2024)

Parameter Identification by Deep Learning of a Material Model for Granular Media

Authors: Derick Nganyu Tanyu, Isabel Michel, Andreas Rademacher, Jörg Kuhnert, Peter Maass

Abstract: Classical physical modelling with associated numerical simulation (model-based), and prognostic methods based on the analysis of large amounts of data (data-driven) are the two most common methods used for the map** of complex physical processes. In recent years, the efficient combination of these approaches has become increasingly important. Continuum mechanics in the core consists of conservat… ▽ More Classical physical modelling with associated numerical simulation (model-based), and prognostic methods based on the analysis of large amounts of data (data-driven) are the two most common methods used for the map** of complex physical processes. In recent years, the efficient combination of these approaches has become increasingly important. Continuum mechanics in the core consists of conservation equations that -- in addition to the always necessary specification of the process conditions -- can be supplemented by phenomenological material models. The latter are an idealized image of the specific material behavior that can be determined experimentally, empirically, and based on a wealth of expert knowledge. The more complex the material, the more difficult the calibration is. This situation forms the starting point for this work's hybrid data-driven and model-based approach for map** a complex physical process in continuum mechanics. Specifically, we use data generated from a classical physical model by the MESHFREE software to train a Principal Component Analysis-based neural network (PCA-NN) for the task of parameter identification of the material model parameters. The obtained results highlight the potential of deep-learning-based hybrid models for determining parameters, which are the key to characterizing materials occurring naturally, and their use in industrial applications (e.g. the interaction of vehicles with sand). △ Less

Submitted 9 July, 2023; originally announced July 2023.

Comments: arXiv admin note: text overlap with arXiv:2212.03130

Single-file transport of binary hard-sphere mixtures through periodic potentials

Authors: David Voráč, Philipp Maass, Artem Ryabov

Abstract: Single-file transport occurs in various scientific fields, including diffusion through nanopores, nanofluidic devices, and cellular processes. We here investigate the impact of polydispersity on particle currents for single-file Brownian motion of hard spheres, when they are driven through periodic potentials by a constant drag force. Through theoretical analysis and extensive Brownian dynamics si… ▽ More Single-file transport occurs in various scientific fields, including diffusion through nanopores, nanofluidic devices, and cellular processes. We here investigate the impact of polydispersity on particle currents for single-file Brownian motion of hard spheres, when they are driven through periodic potentials by a constant drag force. Through theoretical analysis and extensive Brownian dynamics simulations, we unveil the behavior of particle currents for random binary mixtures. The particle currents show a recurring pattern in dependence of the hard-sphere diameters and mixing ratio. We explain this recurrent behavior by showing that a basic unit cell exists in the space of the two hard-sphere diameters. Once the behavior of an observable inside the unit cell is determined, it can be inferred for any diameter. The overall variation of particle currents with the mixing ratio and hard-sphere diameters is reflected by their variation in the limit where the system is fully covered by hard spheres. In this limit, the currents can be predicted analytically. Our analysis explains the occurrence of pronounced maxima and minima of the currents by changes of an effective potential barrier for the center-of-mass motion. △ Less

Submitted 9 August, 2023; v1 submitted 22 June, 2023; originally announced June 2023.

Comments: 10 pages, 7 figures

Journal ref: J. Chem. Phys. 159, 114114 (2023)

Invertible residual networks in the context of regularization theory for linear inverse problems

Authors: Clemens Arndt, Alexander Denker, Sören Dittmer, Nick Heilenkötter, Meira Iske, Tobias Kluth, Peter Maass, Judith Nickel

Abstract: Learned inverse problem solvers exhibit remarkable performance in applications like image reconstruction tasks. These data-driven reconstruction methods often follow a two-step scheme. First, one trains the often neural network-based reconstruction scheme via a dataset. Second, one applies the scheme to new measurements to obtain reconstructions. We follow these steps but parameterize the reconstr… ▽ More Learned inverse problem solvers exhibit remarkable performance in applications like image reconstruction tasks. These data-driven reconstruction methods often follow a two-step scheme. First, one trains the often neural network-based reconstruction scheme via a dataset. Second, one applies the scheme to new measurements to obtain reconstructions. We follow these steps but parameterize the reconstruction scheme with invertible residual networks (iResNets). We demonstrate that the invertibility enables investigating the influence of the training and architecture choices on the resulting reconstruction scheme. For example, assuming local approximation properties of the network, we show that these schemes become convergent regularizations. In addition, the investigations reveal a formal link to the linear regularization theory of linear inverse problems and provide a nonlinear spectral regularization for particular architecture classes. On the numerical side, we investigate the local approximation property of selected trained architectures and present a series of experiments on the MNIST dataset that underpin and extend our theoretical findings. △ Less

Submitted 20 December, 2023; v1 submitted 2 June, 2023; originally announced June 2023.

Journal ref: Inverse Problems 39 125018 (2023)

SVD-DIP: Overcoming the Overfitting Problem in DIP-based CT Reconstruction

Authors: Marco Nittscher, Michael Lameter, Riccardo Barbano, Johannes Leuschner, Bangti **, Peter Maass

Abstract: The deep image prior (DIP) is a well-established unsupervised deep learning method for image reconstruction; yet it is far from being flawless. The DIP overfits to noise if not early stopped, or optimized via a regularized objective. We build on the regularized fine-tuning of a pretrained DIP, by adopting a novel strategy that restricts the learning to the adaptation of singular values. The propos… ▽ More The deep image prior (DIP) is a well-established unsupervised deep learning method for image reconstruction; yet it is far from being flawless. The DIP overfits to noise if not early stopped, or optimized via a regularized objective. We build on the regularized fine-tuning of a pretrained DIP, by adopting a novel strategy that restricts the learning to the adaptation of singular values. The proposed SVD-DIP uses ad hoc convolutional layers whose pretrained parameters are decomposed via the singular value decomposition. Optimizing the DIP then solely consists in the fine-tuning of the singular values, while kee** the left and right singular vectors fixed. We thoroughly validate the proposed method on real-measured $μ$CT data of a lotus root as well as two medical datasets (LoDoPaB and Mayo). We report significantly improved stability of the DIP optimization, by overcoming the overfitting to noise. △ Less

Submitted 15 May, 2023; v1 submitted 28 March, 2023; originally announced March 2023.

Model Stitching and Visualization How GAN Generators can Invert Networks in Real-Time

Authors: Rudolf Herdt, Maximilian Schmidt, Daniel Otero Baguer, Jean Le'Clerc Arrastia, Peter Maass

Abstract: In this work, we propose a fast and accurate method to reconstruct activations of classification and semantic segmentation networks by stitching them with a GAN generator utilizing a 1x1 convolution. We test our approach on images of animals from the AFHQ wild dataset, ImageNet1K, and real-world digital pathology scans of stained tissue samples. Our results show comparable performance to establish… ▽ More In this work, we propose a fast and accurate method to reconstruct activations of classification and semantic segmentation networks by stitching them with a GAN generator utilizing a 1x1 convolution. We test our approach on images of animals from the AFHQ wild dataset, ImageNet1K, and real-world digital pathology scans of stained tissue samples. Our results show comparable performance to established gradient descent methods but with a processing time that is two orders of magnitude faster, making this approach promising for practical applications. △ Less

Submitted 20 February, 2024; v1 submitted 4 February, 2023; originally announced February 2023.

Journal ref: Proceedings of Machine Learning Research 222 (2023) 422-437

Scaling laws for single-file diffusion of adhesive particles

Authors: Sören Schweers, Alexander P. Antonov, Artem Ryabov, Philipp Maass

Abstract: Single-file diffusion refers to the Brownian motion in narrow channels where particles cannot pass each other. In such processes, the diffusion of a tagged particle is typically normal at short times and becomes subdiffusive at long times. For hard-sphere interparticle interaction, the time-dependent mean squared displacement of a tracer is well understood. Here we develop a scaling theory for adh… ▽ More Single-file diffusion refers to the Brownian motion in narrow channels where particles cannot pass each other. In such processes, the diffusion of a tagged particle is typically normal at short times and becomes subdiffusive at long times. For hard-sphere interparticle interaction, the time-dependent mean squared displacement of a tracer is well understood. Here we develop a scaling theory for adhesive particles. It provides a full description of the time-dependent diffusive behavior with a scaling function that depends on an effective strength of adhesive interaction. Particle clustering induced by the adhesive interaction slows down the diffusion at short times, while it enhances subdiffusion at long times. The enhancement effect can be quantified in measurements irrespective of how tagged particles are injected into the system. Combined effects of pore structure and particle adhesiveness should speed up translocation of molecules through narrow pores. △ Less

Submitted 1 March, 2023; v1 submitted 17 January, 2023; originally announced January 2023.

Comments: 6 pages, 2 figures plus 4 pages, 3 figures in SM

Deep Learning Methods for Partial Differential Equations and Related Parameter Identification Problems

Authors: Derick Nganyu Tanyu, Jianfeng Ning, Tom Freudenberg, Nick Heilenkötter, Andreas Rademacher, Uwe Iben, Peter Maass

Abstract: Recent years have witnessed a growth in mathematics for deep learning--which seeks a deeper understanding of the concepts of deep learning with mathematics and explores how to make it more robust--and deep learning for mathematics, where deep learning algorithms are used to solve problems in mathematics. The latter has popularised the field of scientific machine learning where deep learning is app… ▽ More Recent years have witnessed a growth in mathematics for deep learning--which seeks a deeper understanding of the concepts of deep learning with mathematics and explores how to make it more robust--and deep learning for mathematics, where deep learning algorithms are used to solve problems in mathematics. The latter has popularised the field of scientific machine learning where deep learning is applied to problems in scientific computing. Specifically, more and more neural network architectures have been developed to solve specific classes of partial differential equations (PDEs). Such methods exploit properties that are inherent to PDEs and thus solve the PDEs better than standard feed-forward neural networks, recurrent neural networks, or convolutional neural networks. This has had a great impact in the area of mathematical modeling where parametric PDEs are widely used to model most natural and physical processes arising in science and engineering. In this work, we review such methods as well as their extensions for parametric studies and for solving the related inverse problems. We equally proceed to show their relevance in some industrial applications. △ Less

Submitted 16 May, 2023; v1 submitted 6 December, 2022; originally announced December 2022.

Report number: Volume 39 Number 10

Journal ref: Inverse Problems (2023)

SELTO: Sample-Efficient Learned Topology Optimization

Authors: Sören Dittmer, David Erzmann, Henrik Harms, Peter Maass

Abstract: Recent developments in Deep Learning (DL) suggest a vast potential for Topology Optimization (TO). However, while there are some promising attempts, the subfield still lacks a firm footing regarding basic methods and datasets. We aim to address both points. First, we explore physics-based preprocessing and equivariant networks to create sample-efficient components for TO DL pipelines. We evaluate… ▽ More Recent developments in Deep Learning (DL) suggest a vast potential for Topology Optimization (TO). However, while there are some promising attempts, the subfield still lacks a firm footing regarding basic methods and datasets. We aim to address both points. First, we explore physics-based preprocessing and equivariant networks to create sample-efficient components for TO DL pipelines. We evaluate them in a large-scale ablation study using end-to-end supervised training. The results demonstrate a drastic improvement in sample efficiency and the predictions' physical correctness. Second, to improve comparability and future progress, we publish the two first TO datasets containing problems and corresponding ground truth solutions. △ Less

Submitted 6 June, 2023; v1 submitted 12 September, 2022; originally announced September 2022.

Comments: 25 pages, 10 figures, submitted to the International Journal for Numerical Methods in Engineering

MSC Class: 68U05; 68U07; 65N21; 68T01

Brownian dynamics simulations of hard rods in external fields and with contact interactions

Authors: Alexander P. Antonov, Sören Schweers, Artem Ryabov, Philipp Maass

Abstract: We propose a simulation method for Brownian dynamics of hard rods in one dimension for arbitrary continuous external force fields. It is an event-driven procedure based on the fragmentation and mergers of clusters formed by particles in contact. It allows one to treat particle interactions in addition to the hard-sphere exclusion as long as the corresponding interaction forces are continuous funct… ▽ More We propose a simulation method for Brownian dynamics of hard rods in one dimension for arbitrary continuous external force fields. It is an event-driven procedure based on the fragmentation and mergers of clusters formed by particles in contact. It allows one to treat particle interactions in addition to the hard-sphere exclusion as long as the corresponding interaction forces are continuous functions of the particle coordinates. We furthermore develop a treatment of sticky hard spheres as described by Baxter's contact interaction potential. △ Less

Submitted 15 November, 2022; v1 submitted 24 August, 2022; originally announced August 2022.

Comments: 12 pages, 7 figures

Journal ref: Phys. Rev. E 106, 054606 (2022)

Hydrodynamic interactions hinder transport of flow-driven colloidal particles

Authors: Dominik Lips, Eric Cereceda-López, Antonio Ortiz-Ambriz, Pietro Tierno, Artem Ryabov, Philipp Maass

Abstract: The flow-driven transport of interacting micron-sized particles occurs in many soft matter systems spanning from the translocation of proteins to moving emulsions in microfluidic devices. Here we combine experiments and theory to investigate the collective transport properties of colloidal particles along a rotating ring of optical traps. In the co-rotating reference frame, the particles are drive… ▽ More The flow-driven transport of interacting micron-sized particles occurs in many soft matter systems spanning from the translocation of proteins to moving emulsions in microfluidic devices. Here we combine experiments and theory to investigate the collective transport properties of colloidal particles along a rotating ring of optical traps. In the co-rotating reference frame, the particles are driven by a vortex flow of the surrounding fluid. When increasing the depth of the optical potential, we observe a jamming behavior that manifests itself in a strong reduction of the current with increasing particle density. We show that this jamming is caused by hydrodynamic interactions that enhance the energetic barriers between the optical traps. This leads to a transition from an over- to an under-critical tilting of the potential in the corotating frame. Based on analytical considerations, the enhancement effect is estimated to increase with increasing particle size or decreasing radius of the ring of traps. Measurements for different ring radii and Stokesian dynamics simulations for corresponding particle sizes confirm this. The enhancement of potential barriers in the flow-driven system is contrasted to the reduction of barriers in a force-driven one. This diverse behavior demonstrates that hydrodynamic interactions can have a very different impact on the collective dynamics of many-body systems. Applications to soft matter and biological systems require careful consideration of the driving mechanism and of the role of hydrodynamic interactions. △ Less

Submitted 15 November, 2022; v1 submitted 23 August, 2022; originally announced August 2022.

Comments: 14 pages, 7 figures

Journal ref: Soft Matter, 2022

On the possibility of exploring tip-molecule interactions with STM experiments

Authors: Christoph Schiel, Philipp Rahe, Philipp Maass

Abstract: We present a theory for analyzing residence times of single molecules in a fixed detection area of a scanning tunneling microscope (STM). The approach is developed for one-dimensional molecule diffusion and can be extended to two dimensions by using the same methodology. Explicit results are derived for an harmonic attractive and repulsive tip-molecule interaction. Applications of the theory allow… ▽ More We present a theory for analyzing residence times of single molecules in a fixed detection area of a scanning tunneling microscope (STM). The approach is developed for one-dimensional molecule diffusion and can be extended to two dimensions by using the same methodology. Explicit results are derived for an harmonic attractive and repulsive tip-molecule interaction. Applications of the theory allows one to estimate the type and strength of interactions between the STM tip and the molecule. This includes the possibility of an estimation of molecule-molecule interaction when the tip is decorated by a molecule. Despite our focus on STM, this theory can analogously be applied to other experimental probes that monitor single molecules. △ Less

Submitted 21 August, 2022; originally announced August 2022.

Comments: 6 pages, 3 figures

PatchNR: Learning from Very Few Images by Patch Normalizing Flow Regularization

Authors: Fabian Altekrüger, Alexander Denker, Paul Hagemann, Johannes Hertrich, Peter Maass, Gabriele Steidl

Abstract: Learning neural networks using only few available information is an important ongoing research topic with tremendous potential for applications. In this paper, we introduce a powerful regularizer for the variational modeling of inverse problems in imaging. Our regularizer, called patch normalizing flow regularizer (patchNR), involves a normalizing flow learned on small patches of very few images.… ▽ More Learning neural networks using only few available information is an important ongoing research topic with tremendous potential for applications. In this paper, we introduce a powerful regularizer for the variational modeling of inverse problems in imaging. Our regularizer, called patch normalizing flow regularizer (patchNR), involves a normalizing flow learned on small patches of very few images. In particular, the training is independent of the considered inverse problem such that the same regularizer can be applied for different forward operators acting on the same class of images. By investigating the distribution of patches versus those of the whole image class, we prove that our model is indeed a MAP approach. Numerical examples for low-dose and limited-angle computed tomography (CT) as well as superresolution of material images demonstrate that our method provides very high quality results. The training set consists of just six images for CT and one image for superresolution. Finally, we combine our patchNR with ideas from internal learning for performing superresolution of natural images directly from the low-resolution observation without knowledge of any high-resolution image. △ Less

Submitted 21 November, 2022; v1 submitted 24 May, 2022; originally announced May 2022.

Journal ref: Inverse Problems, Volume 39, Number 6, 2023

Solitons in overdamped Brownian dynamics

Authors: Alexander P. Antonov, Artem Ryabov, Philipp Maass

Abstract: Solitons are commonly known as waves that propagate without dispersion. Here we show that they can occur for driven overdamped Brownian dynamics of hard spheres in periodic potentials at high densities. The solitons manifest themselves as periodic sequences of different assemblies of particles moving in the limit of zero noise, where transport of single particles is not possible. They give rise to… ▽ More Solitons are commonly known as waves that propagate without dispersion. Here we show that they can occur for driven overdamped Brownian dynamics of hard spheres in periodic potentials at high densities. The solitons manifest themselves as periodic sequences of different assemblies of particles moving in the limit of zero noise, where transport of single particles is not possible. They give rise to particle currents at even low temperature that appear in band-like structures around certain hard-sphere diameters. At high temperatures, the band-like structures are washed out by the noise, but the particle transport is still dominated by the solitons. All these predicted features should occur in a broad class of periodic systems and are amenable to experimental tests. △ Less

Submitted 21 July, 2022; v1 submitted 29 April, 2022; originally announced April 2022.

Comments: 6 pages, 4 figures plus supplemental material with 7 pages and 1 figure; 5 ancillary movie files

Journal ref: Phys. Rev. Lett. 129, 080601 (2022)

Signatures of geostrophic turbulence in power spectra and third-order-structure function of offshore wind speed fluctuation

Authors: So-Kumneth Sim, Joachim Peinke, Philipp Maass

Abstract: We analyze offshore wind speeds with a time resolution of one second over a long period of 20 months for different heights above the sea level. Energy spectra extending over more than seven decades give a comprehensive picture of wind fluctuations, including intermittency effects at small length scales and synoptic weather phenomena at large scales. The spectra $S(f)$ show a scaling behavior consi… ▽ More We analyze offshore wind speeds with a time resolution of one second over a long period of 20 months for different heights above the sea level. Energy spectra extending over more than seven decades give a comprehensive picture of wind fluctuations, including intermittency effects at small length scales and synoptic weather phenomena at large scales. The spectra $S(f)$ show a scaling behavior consistent with three-dimensional turbulence at high frequencies $f$, followed by a regime at lower frequencies, where $fS(f)$ varies weakly. Lowering the frequency below a crossover frequency $f_{\rm\scriptscriptstyle 2D}$, a rapid rise of $fS(f)$ occurs. An analysis of the third-order structure function $D_3(τ)$ of wind speed differences for a given time lag $τ$ shows a rapid change from negative to positive values of $D_3(τ)$ at $τ\simeq 1/f_{\rm\scriptscriptstyle 2D}$. Remarkably, after applying Taylor's hypothesis locally, we find the third-order structure function to exhibit a behavior very similar to that obtained previously from aircraft measurements at much higher altitudes in the atmosphere. In particular, the third-order structure function grows linearly with the separation distance for negative $D_3$, and with the third power for positive $D_3$. This allows us to estimate energy and enstrophy dissipation rates for offshore wind. The crossover from negative to positive values occurs at about the same separation distance of 400 km as found from the aircraft measurements, suggesting that this length is independent of the altitude in the atmosphere. △ Less

Submitted 19 October, 2023; v1 submitted 15 March, 2022; originally announced March 2022.

Comments: 11 pages, 4 figures

Journal ref: Sci. Rep. 13, 13411 (2023)

Collective excitations in jammed states: ultrafast defect propagation and finite-size scaling

Authors: Alexander P. Antonov, David Voráč, Artem Ryabov, Philipp Maass

Abstract: In crowded systems, particle currents can be mediated by propagating collective excitations which are generated as rare events, are localized and have a finite lifetime. The theoretical description of such excitations is hampered by the problem of identifying complex many-particle transition states, calculation of their free energies, and the evaluation of propagation mechanisms and velocities. He… ▽ More In crowded systems, particle currents can be mediated by propagating collective excitations which are generated as rare events, are localized and have a finite lifetime. The theoretical description of such excitations is hampered by the problem of identifying complex many-particle transition states, calculation of their free energies, and the evaluation of propagation mechanisms and velocities. Here we show that these problems can be tackled for a highly jammed system of hard spheres in a periodic potential. We derive generation rates of collective excitations, their anomalously high velocities, explain the occurrence of an apparent jamming transition and its strong dependence on the system size. The particle currents follow a scaling behavior, where for small systems the current is proportional to the generation rate and for large systems given by the geometric mean of the generation rate and velocity. Our theoretical approach is widely applicable to dense nonequilibrium systems in confined geometries. It provides new perspectives for studying dynamics of collective excitations in experiments. △ Less

Submitted 15 November, 2022; v1 submitted 12 March, 2022; originally announced March 2022.

Comments: 23 pages, 9 figures

Journal ref: New J. Phys. 24, 093020 (2022)

Regularized Orthogonal Nonnegative Matrix Factorization and $K$-means Clustering

Authors: Pascal Fernsel, Peter Maass

Abstract: In this work, we focus on connections between $K$-means clustering approaches and Orthogonal Nonnegative Matrix Factorization (ONMF) methods. We present a novel framework to extract the distance measure and the centroids of the $K$-means method based on first order conditions of the considered ONMF objective function, which exploits the classical alternating minimization schemes of Nonnegative Mat… ▽ More In this work, we focus on connections between $K$-means clustering approaches and Orthogonal Nonnegative Matrix Factorization (ONMF) methods. We present a novel framework to extract the distance measure and the centroids of the $K$-means method based on first order conditions of the considered ONMF objective function, which exploits the classical alternating minimization schemes of Nonnegative Matrix Factorization (NMF) algorithms. While this technique is characterized by a simple derivation procedure, it can also be applied to non-standard regularized ONMF models. Using this framework, we consider in this work ONMF models with $\ell_1$ and standard $\ell_2$ discrepancy terms with an additional elastic net regularization on both factorization matrices and derive the corresponding distance measures and centroids of the generalized $K$-means clustering model. Furthermore, we give an intuitive view of the obtained results, examine special cases and compare them to the findings described in the literature. △ Less

Submitted 14 December, 2021; originally announced December 2021.

MSC Class: 15A23 (Primary) 65F22; 62H30 (Secondary)

An Educated Warm Start For Deep Image Prior-Based Micro CT Reconstruction

Authors: Riccardo Barbano, Johannes Leuschner, Maximilian Schmidt, Alexander Denker, Andreas Hauptmann, Peter Maaß, Bangti **

Abstract: Deep image prior (DIP) was recently introduced as an effective unsupervised approach for image restoration tasks. DIP represents the image to be recovered as the output of a deep convolutional neural network, and learns the network's parameters such that the output matches the corrupted observation. Despite its impressive reconstructive properties, the approach is slow when compared to supervisedl… ▽ More Deep image prior (DIP) was recently introduced as an effective unsupervised approach for image restoration tasks. DIP represents the image to be recovered as the output of a deep convolutional neural network, and learns the network's parameters such that the output matches the corrupted observation. Despite its impressive reconstructive properties, the approach is slow when compared to supervisedly learned, or traditional reconstruction techniques. To address the computational challenge, we bestow DIP with a two-stage learning paradigm: (i) perform a supervised pretraining of the network on a simulated dataset; (ii) fine-tune the network's parameters to adapt to the target reconstruction task. We provide a thorough empirical analysis to shed insights into the impacts of pretraining in the context of image reconstruction. We showcase that pretraining considerably speeds up and stabilizes the subsequent reconstruction task from real-measured 2D and 3D micro computed tomography data of biological specimens. The code and additional experimental materials are available at https://educateddip.github.io/docs.educated_deep_image_prior/. △ Less

Submitted 8 February, 2023; v1 submitted 23 November, 2021; originally announced November 2021.

Journal ref: in IEEE Transactions on Computational Imaging, vol. 8, pp. 1210-1222, 2022

Predicting conductivities of alkali borophosphate glasses based on site energy distributions derived from network former unit concentrations

Authors: Marco Bosi, Philipp Maass

Abstract: For ion transport in network glasses, it is a great challenge to predict conductivities specifically based on structural properties. To this end it is necessary to gain an understanding of the energy landscape where the thermally activated hop** motion of the ions takes place. For alkali borophosphate glasses, a statistical mechanical approach was suggested to predict essential characteristics o… ▽ More For ion transport in network glasses, it is a great challenge to predict conductivities specifically based on structural properties. To this end it is necessary to gain an understanding of the energy landscape where the thermally activated hop** motion of the ions takes place. For alkali borophosphate glasses, a statistical mechanical approach was suggested to predict essential characteristics of the distribution of energies at the residence sites of the mobile alkali ions. The corresponding distribution of site energies was derived from the chemical units forming the glassy network. A hop** model based on the site energy landscape allowed to model the change of conductivity activation energies with the borate to phosphate mixing ratio. Here we refine and extend this general approach to cope with minimal local activation barriers and to calculate dc-conductivities without the need of performing extensive Monte-Carlo simulations. This calculation relies on the map** of the many-body ion dynamics onto a network of local conductances derived from the elementary jump rates of the mobile ions. Application of the theoretical modelling to three series of alkali borophosphate glasses with the compositions $0.33$Li$_2$O$-0.67$[$x$B$_2$O$_3$-$(1\!-\!x)$P$_2$O$_5$], $0.35$Na$_2$O$-0.65$[$x$B$_2$O$_3$-$(1\!-\!x)$P$_2$O$_5$] and $0.4$Na$_2$O$-0.6$[$x$B$_2$O$_3$-$(1\!-\!x)$P$_2$O$_5$] shows good agreement with experimental data. △ Less

Submitted 10 November, 2021; originally announced November 2021.

Comments: 11 pages, 10 figures

Conditional Invertible Neural Networks for Medical Imaging

Authors: Alexander Denker, Maximilian Schmidt, Johannes Leuschner, Peter Maass

Abstract: Over the last years, deep learning methods have become an increasingly popular choice to solve tasks from the field of inverse problems. Many of these new data-driven methods have produced impressive results, although most only give point estimates for the reconstruction. However, especially in the analysis of ill-posed inverse problems, the study of uncertainties is essential. In our work, we app… ▽ More Over the last years, deep learning methods have become an increasingly popular choice to solve tasks from the field of inverse problems. Many of these new data-driven methods have produced impressive results, although most only give point estimates for the reconstruction. However, especially in the analysis of ill-posed inverse problems, the study of uncertainties is essential. In our work, we apply generative flow-based models based on invertible neural networks to two challenging medical imaging tasks, i.e. low-dose computed tomography and accelerated medical resonance imaging. We test different architectures of invertible neural networks and provide extensive ablation studies. In most applications, a standard Gaussian is used as the base distribution for a flow-based model. Our results show that the choice of a radial distribution can improve the quality of reconstructions. △ Less

Submitted 26 October, 2021; originally announced October 2021.

Hydrodynamic interactions can induce jamming in flow-driven systems

Authors: Eric Cereceda-López, Dominik Lips, Antonio Ortiz-Ambriz, Artem Ryabov, Philipp Maass, Pietro Tierno

Abstract: Hydrodynamic interactions between fluid-dispersed particles are ubiquitous in soft matter and biological systems and they give rise to intriguing collective phenomena. While it was reported that these interactions can facilitate force-driven particle motion over energetic barriers, here we show the opposite effect in a flow-driven system, i.e. that hydrodynamic interactions hinder transport across… ▽ More Hydrodynamic interactions between fluid-dispersed particles are ubiquitous in soft matter and biological systems and they give rise to intriguing collective phenomena. While it was reported that these interactions can facilitate force-driven particle motion over energetic barriers, here we show the opposite effect in a flow-driven system, i.e. that hydrodynamic interactions hinder transport across barriers. We demonstrate this result by combining experiments and theory. In the experiments, we drive colloidal particles using rotating optical traps, thus creating a vortex flow in the corotating reference frame. We observe a jamming-like decrease of particle currents with density for large barriers between traps. The theoretical model shows that this jamming arises from hydrodynamic interactions between the particles. The impact of hydrodynamic interactions is reversed compared to force-driven motion, suggesting that our findings are a generic feature of flow-driven transport. △ Less

Submitted 11 October, 2021; originally announced October 2021.

Journal ref: Phys. Rev. Lett. 127, 214501 (2021)

Driven transport of soft Brownian particles through pore-like structures: Effective size method

Authors: Alexander P. Antonov, Artem Ryabov, Philipp Maass

Abstract: Single-file transport in pore-like structures constitute an important topic for both theory and experiment. For hardcore interacting particles, a good understanding of the collective dynamics has been achieved recently. Here we study how softness in the particle interaction affects the emergent transport behavior. To this end, we investigate the driven Brownian motion of particles in a periodic po… ▽ More Single-file transport in pore-like structures constitute an important topic for both theory and experiment. For hardcore interacting particles, a good understanding of the collective dynamics has been achieved recently. Here we study how softness in the particle interaction affects the emergent transport behavior. To this end, we investigate the driven Brownian motion of particles in a periodic potential. The particles interact via a repulsive softcore potential with a shape corresponding to a smoothed rectangular barrier. This shape allows us to elucidate effects of mutual particle penetration and particle crossing in a controlled manner. We find that even weak deviations from the hardcore case can have a strong impact on the particle current. Despite this fact, the knowledge about the transport in a corresponding hardcore system is shown to be useful to describe and interpret our findings for the softcore case. This is achieved by assigning a thermodynamic effective size to the particles based on the equilibrium density functional of hard spheres. △ Less

Submitted 24 December, 2021; v1 submitted 30 July, 2021; originally announced July 2021.

Comments: 8 pages, 4 figures

Journal ref: J. Chem. Phys. 155, 184102 (2021)

Generalization of the Change of Variables Formula with Applications to Residual Flows

Authors: Niklas Koenen, Marvin N. Wright, Peter Maaß, Jens Behrmann

Abstract: Normalizing flows leverage the Change of Variables Formula (CVF) to define flexible density models. Yet, the requirement of smooth transformations (diffeomorphisms) in the CVF poses a significant challenge in the construction of these models. To enlarge the design space of flows, we introduce $\mathcal{L}$-diffeomorphisms as generalized transformations which may violate these requirements on zero… ▽ More Normalizing flows leverage the Change of Variables Formula (CVF) to define flexible density models. Yet, the requirement of smooth transformations (diffeomorphisms) in the CVF poses a significant challenge in the construction of these models. To enlarge the design space of flows, we introduce $\mathcal{L}$-diffeomorphisms as generalized transformations which may violate these requirements on zero Lebesgue-measure sets. This relaxation allows e.g. the use of non-smooth activation functions such as ReLU. Finally, we apply the obtained results to planar, radial, and contractive residual flows. △ Less

Submitted 9 July, 2021; originally announced July 2021.

Deeply supervised UNet for semantic segmentation to assist dermatopathological assessment of Basal Cell Carcinoma (BCC)

Authors: Jean Le'Clerc Arrastia, Nick Heilenkötter, Daniel Otero Baguer, Lena Hauberg-Lotte, Tobias Boskamp, Sonja Hetzer, Nicole Duschner, Jörg Schaller, Peter Maaß

Abstract: Accurate and fast assessment of resection margins is an essential part of a dermatopathologist's clinical routine. In this work, we successfully develop a deep learning method to assist the pathologists by marking critical regions that have a high probability of exhibiting pathological features in Whole Slide Images (WSI). We focus on detecting Basal Cell Carcinoma (BCC) through semantic segmentat… ▽ More Accurate and fast assessment of resection margins is an essential part of a dermatopathologist's clinical routine. In this work, we successfully develop a deep learning method to assist the pathologists by marking critical regions that have a high probability of exhibiting pathological features in Whole Slide Images (WSI). We focus on detecting Basal Cell Carcinoma (BCC) through semantic segmentation using several models based on the UNet architecture. The study includes 650 WSI with 3443 tissue sections in total. Two clinical dermatopathologists annotated the data, marking tumor tissues' exact location on 100 WSI. The rest of the data, with ground-truth section-wise labels, is used to further validate and test the models. We analyze two different encoders for the first part of the UNet network and two additional training strategies: a) deep supervision, b) linear combination of decoder outputs, and obtain some interpretations about what the network's decoder does in each case. The best model achieves over 96%, accuracy, sensitivity, and specificity on the test set. △ Less

Submitted 8 March, 2021; v1 submitted 5 March, 2021; originally announced March 2021.

Network-forming units, energy landscapes and conductivity activation energies in alkali borophosphate glasses: analytical approaches

Authors: Marco Bosi, Julian Fischer, Philipp Maass

Abstract: A major challenge in the modeling of ionically conducting glasses is to understand how the large variety of possible chemical compositions and specific structural properties influence ionic transport quantities. Here we revisit and extend a theoretical approach for alkali borophosphate glasses, where changes of conductivity activation energies with the borate to phosphate mixing ratio are related… ▽ More A major challenge in the modeling of ionically conducting glasses is to understand how the large variety of possible chemical compositions and specific structural properties influence ionic transport quantities. Here we revisit and extend a theoretical approach for alkali borophosphate glasses, where changes of conductivity activation energies with the borate to phosphate mixing ratio are related to modifications of the ionic site energy landscape. The landscape modifications are caused by varying amounts of different units forming the glassy network, which lead to spatial redistributions of the counter-charges of the mobile alkali ions. Theoretical approaches are presented to calculate variations of both network former unit concentrations and activation energies with the glass composition. Applications to several alkali borophosphate glasses show good agreement with experimental data. △ Less

Submitted 30 October, 2021; v1 submitted 4 November, 2020; originally announced November 2020.

Comments: 15 pages, 7 figures

Journal ref: J. Chem. Phys. C 125 (2021) 6260-6268

Molecular self-assembly: Quantifying the balance between intermolecular attraction and repulsion from distance and length distributions

Authors: Christoph Schiel, Maximilian Vogtland, Ralf Bechstein, Angelika Kühnle, Philipp Maass

Abstract: Molecular self-assembly on surfaces constitutes a powerful method for creating tailor-made surface structures with dedicated functionalities. Varying the intermolecular interactions allows for tuning the resulting molecular structures in a rational fashion. So far, however, the discussion of the involved intermolecular interactions is often limited to attractive forces only. In real systems, the i… ▽ More Molecular self-assembly on surfaces constitutes a powerful method for creating tailor-made surface structures with dedicated functionalities. Varying the intermolecular interactions allows for tuning the resulting molecular structures in a rational fashion. So far, however, the discussion of the involved intermolecular interactions is often limited to attractive forces only. In real systems, the intermolecular interaction can be composed of both, attractive and repulsive forces. Adjusting the balance between these interactions provides a promising strategy for extending the structural variety in molecular self-assembly on surfaces. This strategy, however, relies on a method to quantify the involved interactions. Here, we investigate a molecular model system of 3-hydroxybenzoic acid molecules on calcite (10.4) in ultrahigh vacuum. This system offers both anisotropic short-range attraction and long-range repulsion between the molecules, resulting in the self-assembly of molecular stripes. We analyze the stripe-to-stripe distance distribution and the stripe length distribution and compare these distributions with analytical expressions from an anisotropic Ising model with additional repulsive interaction. We show that this approach allows to extract quantitative information about the strength of the attractive and repulsive interactions. Our work demonstrates how the detailed analysis of the self-assembled structures can be used to obtain quantitative insight into the molecule-molecule interactions. △ Less

Submitted 22 July, 2020; originally announced July 2020.

Comments: 13 pages, 6 figures

Deep image prior for 3D magnetic particle imaging: A quantitative comparison of regularization techniques on Open MPI dataset

Authors: Sören Dittmer, Tobias Kluth, Mads Thorstein Roar Henriksen, Peter Maass

Abstract: Magnetic particle imaging (MPI) is an imaging modality exploiting the nonlinear magnetization behavior of (super-)paramagnetic nanoparticles to obtain a space- and often also time-dependent concentration of a tracer consisting of these nanoparticles. MPI has a continuously increasing number of potential medical applications. One prerequisite for successful performance in these applications is a pr… ▽ More Magnetic particle imaging (MPI) is an imaging modality exploiting the nonlinear magnetization behavior of (super-)paramagnetic nanoparticles to obtain a space- and often also time-dependent concentration of a tracer consisting of these nanoparticles. MPI has a continuously increasing number of potential medical applications. One prerequisite for successful performance in these applications is a proper solution to the image reconstruction problem. More classical methods from inverse problems theory, as well as novel approaches from the field of machine learning, have the potential to deliver high-quality reconstructions in MPI. We investigate a novel reconstruction approach based on a deep image prior, which builds on representing the solution by a deep neural network. Novel approaches, as well as variational and iterative regularization techniques, are compared quantitatively in terms of peak signal-to-noise ratios and structural similarity indices on the publicly available Open MPI dataset. △ Less

Submitted 3 July, 2020; originally announced July 2020.

Ground Truth Free Denoising by Optimal Transport

Authors: Sören Dittmer, Carola-Bibiane Schönlieb, Peter Maass

Abstract: We present a learned unsupervised denoising method for arbitrary types of data, which we explore on images and one-dimensional signals. The training is solely based on samples of noisy data and examples of noise, which -- critically -- do not need to come in pairs. We only need the assumption that the noise is independent and additive (although we describe how this can be extended). The method res… ▽ More We present a learned unsupervised denoising method for arbitrary types of data, which we explore on images and one-dimensional signals. The training is solely based on samples of noisy data and examples of noise, which -- critically -- do not need to come in pairs. We only need the assumption that the noise is independent and additive (although we describe how this can be extended). The method rests on a Wasserstein Generative Adversarial Network setting, which utilizes two critics and one generator. △ Less

Submitted 3 July, 2020; originally announced July 2020.

Cycle completion times probe interactions with environment

Authors: David Voráč, Philipp Maass, Artem Ryabov

Abstract: Recent measurements of durations of non-equilibrium processes provide valuable information on microscopic mechanisms and energetics. Comprehensive theory for corresponding experiments so far is well developed for single-particle systems only. Little is known for interacting systems in non-equilibrium environments. Here, we introduce and study a basic model for cycle processes interacting with an e… ▽ More Recent measurements of durations of non-equilibrium processes provide valuable information on microscopic mechanisms and energetics. Comprehensive theory for corresponding experiments so far is well developed for single-particle systems only. Little is known for interacting systems in non-equilibrium environments. Here, we introduce and study a basic model for cycle processes interacting with an environment that can exhibit a net particle flow. We find a surprising richness of cycle time variations with environmental conditions. This manifests itself in unequal cycle times $τ^+$ and $τ^-$ in forward and backward cycle direction with both asymmetries $τ^-<τ^+$ and $τ^->τ^+$, speeding up of backward cycles by interactions, and dynamical phase transitions, where cycle times become multimodal functions of the bias. The model allows us to relate these effects to specific microscopic mechanisms, which can be helpful for interpreting experiments. △ Less

Submitted 19 June, 2020; originally announced June 2020.

Comments: 13 pages, 4 figures

Journal ref: J. Phys. Chem. Lett. 2020, 11, 16, 6887-6891 (2020)

Conditional Normalizing Flows for Low-Dose Computed Tomography Image Reconstruction

Authors: Alexander Denker, Maximilian Schmidt, Johannes Leuschner, Peter Maass, Jens Behrmann

Abstract: Image reconstruction from computed tomography (CT) measurement is a challenging statistical inverse problem since a high-dimensional conditional distribution needs to be estimated. Based on training data obtained from high-quality reconstructions, we aim to learn a conditional density of images from noisy low-dose CT measurements. To tackle this problem, we propose a hybrid conditional normalizing… ▽ More Image reconstruction from computed tomography (CT) measurement is a challenging statistical inverse problem since a high-dimensional conditional distribution needs to be estimated. Based on training data obtained from high-quality reconstructions, we aim to learn a conditional density of images from noisy low-dose CT measurements. To tackle this problem, we propose a hybrid conditional normalizing flow, which integrates the physical model by using the filtered back-projection as conditioner. We evaluate our approach on a low-dose CT benchmark and demonstrate superior performance in terms of structural similarity of our flow-based method compared to other deep learning based approaches. △ Less

Submitted 11 June, 2020; originally announced June 2020.

Comments: Submitted to the ICML Workshop on Invertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models, Vienna, Austria, 2020

Joint super-resolution image reconstruction and parameter identification in imaging operator: Analysis of bilinear operator equations, numerical solution, and application to magnetic particle imaging

Authors: Tobias Kluth, Christine Bathke, Ming Jiang, Peter Maass

Abstract: One important property of imaging modalities and related applications is the resolution of image reconstructions which relies on various factors such as instrumentation or data processing. Restrictions in resolution can have manifold origins, e.g., limited resolution of available data, noise level in the data, and/or inexact model operators. In this work we investigate a novel data processing appr… ▽ More One important property of imaging modalities and related applications is the resolution of image reconstructions which relies on various factors such as instrumentation or data processing. Restrictions in resolution can have manifold origins, e.g., limited resolution of available data, noise level in the data, and/or inexact model operators. In this work we investigate a novel data processing approach suited for inexact model operators. Here, two different information sources, high-dimensional model information and high-quality measurement on a lower resolution, are comprised in a hybrid approach. The joint reconstruction of a high resolution image and parameters of the imaging operator are obtained by minimizing a Tikhonov-type functional. The hybrid approach is analyzed for bilinear operator equations with respect to stability, convergence, and convergence rates. We further derive an algorithmic solution exploiting an algebraic reconstruction technique. The study is complemented by numerical results ranging from an academic test case to image reconstruction in magnetic particle imaging. △ Less

Submitted 27 April, 2020; originally announced April 2020.

Error analysis for filtered back projection reconstructions in Besov spaces

Authors: Matthias Beckmann, Peter Maass, Judith Nickel

Abstract: Filtered back projection (FBP) methods are the most widely used reconstruction algorithms in computerized tomography (CT). The ill-posedness of this inverse problem allows only an approximate reconstruction for given noisy data. Studying the resulting reconstruction error has been a most active field of research in the 1990s and has recently been revived in terms of optimal filter design and estim… ▽ More Filtered back projection (FBP) methods are the most widely used reconstruction algorithms in computerized tomography (CT). The ill-posedness of this inverse problem allows only an approximate reconstruction for given noisy data. Studying the resulting reconstruction error has been a most active field of research in the 1990s and has recently been revived in terms of optimal filter design and estimating the FBP approximation errors in general Sobolev spaces. However, the choice of Sobolev spaces is suboptimal for characterizing typical CT reconstructions. A widely used model are sums of characteristic functions, which are better modelled in terms of Besov spaces $\mathrm{B}^{α,p}_q(\mathbb{R}^2)$. In particular $\mathrm{B}^{α,1}_1(\mathbb{R}^2)$ with $α\approx 1$ is a preferred model in image analysis for describing natural images. In case of noisy Radon data the total FBP reconstruction error $$\|f-f_L^δ\| \le \|f-f_L\|+ \|f_L - f_L^δ\|$$ splits into an approximation error and a data error, where $L$ serves as regularization parameter. In this paper, we study the approximation error of FBP reconstructions for target functions $f \in \mathrm{L}^1(\mathbb{R}^2) \cap \mathrm{B}^{α,p}_q(\mathbb{R}^2)$ with positive $α\not\in \mathbb{N}$ and $1 \leq p,q \leq \infty$. We prove that the $\mathrm{L}^p$-norm of the inherent FBP approximation error $f-f_L$ can be bounded above by \begin{equation*} \|f - f_L\|_{\mathrm{L}^p(\mathbb{R}^2)} \leq c_{α,q,W} \, L^{-α} \, |f|_{\mathrm{B}^{α,p}_q(\mathbb{R}^2)} \end{equation*} under suitable assumptions on the utilized low-pass filter's window function $W$. This then extends by classical methods to estimates for the total reconstruction error. △ Less

Submitted 14 April, 2020; originally announced April 2020.

Comments: 32 pages, 8 figures

Journal ref: Inverse Problems 37(1), 2021, 014002

Statistics of work performed by optical tweezers with general time-variation of their stiffness

Authors: Petr Chvosta, Dominik Lips, Viktor Holubec, Artem Ryabov, Philipp Maass

Abstract: We derive an exact expression for the probability density of work done on a particle that diffuses in a parabolic potential with a stiffness varying by an arbitrary piecewise constant protocol. Based on this result, the work distribution for time-continuous protocols of the stiffness can be determined up to any degree of accuracy. This is achieved by replacing the continuous driving by a piecewise… ▽ More We derive an exact expression for the probability density of work done on a particle that diffuses in a parabolic potential with a stiffness varying by an arbitrary piecewise constant protocol. Based on this result, the work distribution for time-continuous protocols of the stiffness can be determined up to any degree of accuracy. This is achieved by replacing the continuous driving by a piecewise constant one with a number $n$ of positive or negative steps of increasing or decreasing stiffness. With increasing $n$, the work distributions for the piecewise protocols approach that for the continuous protocol. The moment generating function of the work is given by the inverse square root of a polynomial of degree $n$, whose coefficients are efficiently calculated from a recurrence relation. The roots of the polynomials are real and positive (negative) steps of the protocol are associated with negative (positive) roots. Using these properties the inverse Laplace transform of the moment generating function is carried out explicitly. Fluctuation theorems are used to derive further properties of the polynomials and their roots. △ Less

Submitted 19 May, 2020; v1 submitted 17 March, 2020; originally announced March 2020.

Comments: 21 pages, 4 figures

Journal ref: J. Phys. A: Math. Theor. 53, 275001 (2020)

Nonequilibrium transport and phase transitions in driven diffusion of interacting particles

Authors: Dominik Lips, Artem Ryabov, Philipp Maass

Abstract: Driven diffusive systems constitute paradigmatic models of nonequilibrium physics. Among them, a driven lattice gas known as the asymmetric simple exclusion process (ASEP) is the most prominent example for which many intriguing exact results have been obtained. After summarizing key findings, including the map** of the ASEP to quantum spin chains, we discuss the recently introduced Brownian asym… ▽ More Driven diffusive systems constitute paradigmatic models of nonequilibrium physics. Among them, a driven lattice gas known as the asymmetric simple exclusion process (ASEP) is the most prominent example for which many intriguing exact results have been obtained. After summarizing key findings, including the map** of the ASEP to quantum spin chains, we discuss the recently introduced Brownian asymmetric simple exclusion process (BASEP) as a related class of driven diffusive system with continuous space dynamics. In the BASEP, driven Brownian motion of hardcore-interacting particles through one-dimensional periodic potentials is considered. We study whether current-density relations of the BASEP can be considered as generic for arbitrary periodic potentials and whether repulsive particle interactions other than hardcore lead to similar results. Our findings suggest that shapes of current-density relations are generic for single-well periodic potentials and can always be attributed to the interplay of a barrier reduction, blocking and exchange symmetry effect. This implies that in general up to five different phases of nonequilibrium steady states are possible for such potentials. The phases can occur in systems coupled to particle reservoirs, where the bulk density is the order parameter. For multiple-well periodic potentials, more complex current-density relations are possible and more phases can appear. Taking a repulsive Yukawa potential as an example, we show that the effects of barrier reduction and blocking on the current are also present. The exchange symmetry effect requires hardcore interactions and we demonstrate that it can still be identified when hardcore interactions are combined with weak Yukawa interactions. △ Less

Submitted 23 February, 2020; v1 submitted 27 January, 2020; originally announced January 2020.

Comments: 14 pages, 9 figures

Deep Relevance Regularization: Interpretable and Robust Tumor Ty** of Imaging Mass Spectrometry Data

Authors: Christian Etmann, Maximilian Schmidt, Jens Behrmann, Tobias Boskamp, Lena Hauberg-Lotte, Annette Peter, Rita Casadonte, Jörg Kriegsmann, Peter Maass

Abstract: Neural networks have recently been established as a viable classification method for imaging mass spectrometry data for tumor ty**. For multi-laboratory scenarios however, certain confounding factors may strongly impede their performance. In this work, we introduce Deep Relevance Regularization, a method of restricting what the neural network can focus on during classification, in order to impro… ▽ More Neural networks have recently been established as a viable classification method for imaging mass spectrometry data for tumor ty**. For multi-laboratory scenarios however, certain confounding factors may strongly impede their performance. In this work, we introduce Deep Relevance Regularization, a method of restricting what the neural network can focus on during classification, in order to improve the classification performance. We demonstrate how Deep Relevance Regularization robustifies neural networks against confounding factors on a challenging inter-lab dataset consisting of breast and ovarian carcinoma. We further show that this makes the relevance map -- a way of visualizing the discriminative parts of the mass spectrum -- sparser, thereby making the classifier easier to interpret △ Less

Submitted 10 December, 2019; originally announced December 2019.

The LoDoPaB-CT Dataset: A Benchmark Dataset for Low-Dose CT Reconstruction Methods

Authors: Johannes Leuschner, Maximilian Schmidt, Daniel Otero Baguer, Peter Maaß

Abstract: Deep Learning approaches for solving Inverse Problems in imaging have become very effective and are demonstrated to be quite competitive in the field. Comparing these approaches is a challenging task since they highly rely on the data and the setup that is used for training. We provide a public dataset of computed tomography images and simulated low-dose measurements suitable for training this kin… ▽ More Deep Learning approaches for solving Inverse Problems in imaging have become very effective and are demonstrated to be quite competitive in the field. Comparing these approaches is a challenging task since they highly rely on the data and the setup that is used for training. We provide a public dataset of computed tomography images and simulated low-dose measurements suitable for training this kind of methods. With the LoDoPaB-CT Dataset we aim to create a benchmark that allows for a fair comparison. It contains over 40,000 scan slices from around 800 patients selected from the LIDC/IDRI Database. In this paper we describe how we processed the original slices and how we simulated the measurements. We also include first baseline results. △ Less

Submitted 3 May, 2020; v1 submitted 1 October, 2019; originally announced October 2019.

Journal ref: Scientific Data Volume 8, Article number: 109 (2021)

Heterogeneities in electricity grids strongly enhance non-Gaussian features of frequency fluctuations under stochastic power input

Authors: Matthias F. Wolff, Katrin Schmietendorf, Pedro G. Lind, Oliver Kamps, Joachim Peinke, Philipp Maass

Abstract: Stochastic feed-in of fluctuating renewable energies is steadily increasing in modern electricity grids and this becomes an important risk factor for maintaining power grid stability. Here we study the impact of wind power feed-in on the short-term frequency fluctuations in power grids based on an IEEE test grid structure, the swing equation for the dynamics of voltage phase angles, and a series o… ▽ More Stochastic feed-in of fluctuating renewable energies is steadily increasing in modern electricity grids and this becomes an important risk factor for maintaining power grid stability. Here we study the impact of wind power feed-in on the short-term frequency fluctuations in power grids based on an IEEE test grid structure, the swing equation for the dynamics of voltage phase angles, and a series of measured wind speed data. External control measures are accounted for by adjusting the grid state to the average power feed-in on a time scale of one minute. The wind power is injected at a single node by replacing one of the conventional generator nodes in the test grid by a wind farm. We determine histograms of local frequencies for a large number of one-minute wind speed sequences taken from the measured data and for different injection nodes. These histograms exhibit a common type of shape, which can be described by a Gaussian distribution for small frequencies and a nearly exponentially decaying tail part. Non-Gaussian features become particularly pronounced for wind power injection at locations, which are weakly connected to the main grid structure. This effect is only present when taking into account the heterogeneities in transmission line and node properties of the grid, while it disappears upon homogenizing of these features. The standard deviation of the frequency fluctuations increases linearly with the average injected wind power. △ Less

Submitted 1 November, 2019; v1 submitted 10 August, 2019; originally announced August 2019.

Comments: 9 pages, 7 figures

Journal ref: Chaos 29, 103149 (2019)

A Projectional Ansatz to Reconstruction

Authors: Sören Dittmer, Peter Maass

Abstract: Recently the field of inverse problems has seen a growing usage of mathematically only partially understood learned and non-learned priors. Based on first principles, we develop a projectional approach to inverse problems that addresses the incorporation of these priors, while still guaranteeing data consistency. We implement this projectional method (PM) on the one hand via very general Plug-and-… ▽ More Recently the field of inverse problems has seen a growing usage of mathematically only partially understood learned and non-learned priors. Based on first principles, we develop a projectional approach to inverse problems that addresses the incorporation of these priors, while still guaranteeing data consistency. We implement this projectional method (PM) on the one hand via very general Plug-and-Play priors and on the other hand, via an end-to-end training approach. To this end, we introduce a novel alternating neural architecture, allowing for the incorporation of highly customized priors from data in a principled manner. We also show how the recent success of Regularization by Denoising (RED) can, at least to some extent, be explained as an approximation of the PM. Furthermore, we demonstrate how the idea can be applied to stop the degradation of Deep Image Prior (DIP) reconstructions over time. △ Less

Submitted 6 August, 2019; v1 submitted 10 July, 2019; originally announced July 2019.

Single-file transport in periodic potentials: The Brownian asymmetric exclusion process

Authors: Dominik Lips, Artem Ryabov, Philipp Maass

Abstract: Single-file Brownian motion in periodic structures is an important process in nature and technology, which becomes increasingly amenable for experimental investigation under controlled conditions. To explore and understand generic features of this motion, the Brownian asymmetric simple exclusion process (BASEP) was recently introduced. The BASEP refers to diffusion models, where hard spheres are d… ▽ More Single-file Brownian motion in periodic structures is an important process in nature and technology, which becomes increasingly amenable for experimental investigation under controlled conditions. To explore and understand generic features of this motion, the Brownian asymmetric simple exclusion process (BASEP) was recently introduced. The BASEP refers to diffusion models, where hard spheres are driven by a constant drag force through a periodic potential. Here, we derive general properties of the rich collective dynamics in the BASEP. Average currents in the steady state change dramatically with the particle size and density. For an open system coupled to particle reservoirs, extremal current principles predict various nonequilibrium phases, which we verify by Brownian dynamics simulations. For general pair interactions we discuss connections to single-file transport by traveling-wave potentials and prove the impossibility of current reversals in systems driven by a constant drag and by traveling waves. △ Less

Submitted 19 November, 2019; v1 submitted 20 June, 2019; originally announced June 2019.

Comments: 15 pages, 11 figures

Journal ref: Phys. Rev. E 100, 052121 (2019)

On the Connection Between Adversarial Robustness and Saliency Map Interpretability

Authors: Christian Etmann, Sebastian Lunz, Peter Maass, Carola-Bibiane Schönlieb

Abstract: Recent studies on the adversarial vulnerability of neural networks have shown that models trained to be more robust to adversarial attacks exhibit more interpretable saliency maps than their non-robust counterparts. We aim to quantify this behavior by considering the alignment between input image and saliency map. We hypothesize that as the distance to the decision boundary grows,so does the align… ▽ More Recent studies on the adversarial vulnerability of neural networks have shown that models trained to be more robust to adversarial attacks exhibit more interpretable saliency maps than their non-robust counterparts. We aim to quantify this behavior by considering the alignment between input image and saliency map. We hypothesize that as the distance to the decision boundary grows,so does the alignment. This connection is strictly true in the case of linear models. We confirm these theoretical findings with experiments based on models trained with a local Lipschitz regularization and identify where the non-linear nature of neural networks weakens the relation. △ Less

Submitted 10 May, 2019; originally announced May 2019.

Comments: 12 pages, accepted for publication at the 36th International Conference on Machine Learning 2019

Regularization by architecture: A deep prior approach for inverse problems

Authors: Sören Dittmer, Tobias Kluth, Peter Maass, Daniel Otero Baguer

Abstract: The present paper studies so-called deep image prior (DIP) techniques in the context of ill-posed inverse problems. DIP networks have been recently introduced for applications in image processing; also first experimental results for applying DIP to inverse problems have been reported. This paper aims at discussing different interpretations of DIP and to obtain analytic results for specific network… ▽ More The present paper studies so-called deep image prior (DIP) techniques in the context of ill-posed inverse problems. DIP networks have been recently introduced for applications in image processing; also first experimental results for applying DIP to inverse problems have been reported. This paper aims at discussing different interpretations of DIP and to obtain analytic results for specific network designs and linear operators. The main contribution is to introduce the idea of viewing these approaches as the optimization of Tikhonov functionals rather than optimizing networks. Besides theoretical results, we present numerical verifications. △ Less

Submitted 18 March, 2020; v1 submitted 10 December, 2018; originally announced December 2018.

Journal ref: Journal of Mathematical Imaging and Vision (2019)