Search

Dataset dependence affects many real‐life applications of machine learning: the performance of a model trained on a dataset is significantly worse on samples from another dataset than on new, unseen samples from the original one. This issue is particularly acute for small and somewhat specific databases in medical applications; the automated recognition of melanoma from skin lesion images is a prime example. We document dataset dependence in dermoscopic skin lesion image classification using three publicly available medium size datasets. Standard machine learning techniques aimed at improving the predictive power of a model might enhance performance slightly, but the gain is small, the dataset dependence is not reduced, and the best combination depends on model details. We demonstrate that simple differences in image statistics account for only 5% of the dataset dependence. We suggest a solution with two essential ingredients: using an ensemble of heterogeneous models, and training on a heterogeneous dataset. Our ensemble consists of 29 convolutional networks, some of which are trained on features considered important by dermatologists; the networks' output is fused by a trained committee machine. The combined International Skin Imaging Collaboration dataset is suitable for training, as it is multi‐source, produced by a collaboration of a number of clinics over the world. Building on the strengths of the ensemble, it is applied to a related problem as well: recognizing melanoma based on clinical (non‐dermoscopic) images. This is a harder problem as both the image quality is lower than those of the dermoscopic ones and the available public datasets are smaller and scarcer. We explored various training strategies and showed that 79% balanced accuracy can be achieved for binary classification averaged over three clinical datasets. [ABSTRACT FROM AUTHOR]

Highlights • Pattern formation guided by an autocrine inhibitor through a reaction-diffusion process. • Connected molecular parameters to statistical characteristics of the vasculature. • Specific model predictions tested by experiments. Abstract Vascular patterning is a key process during development and disease. The diffusive decoy receptor sVEGFR1 (sFlt1) is a known regulator of endothelial cell behavior, yet the mechanism by which it controls vascular structure is little understood. We propose computational models to shed light on how vascular patterning is guided by self-organized gradients of the VEGF/sVEGFR1 factors. We demonstrate that a diffusive inhibitor can generate structures with a dense branching morphology in models where the activator elicits directed growth. Inadequate presence of the inhibitor leads to compact growth, while excessive production of the inhibitor blocks expansion and stabilizes existing structures. Model predictions were compared with time-resolved experimental data obtained from endothelial sprout kinetics in fibrin gels. In the presence of inhibitory antibodies against VEGFR1 vascular sprout density increases while the speed of sprout expansion remains unchanged. Thus, the rate of secretion and stability of extracellular sVEGFR1 can modulate vascular sprout density. [ABSTRACT FROM AUTHOR]

A novel approach for video instance segmentation is presented using semisupervised learning. Our Cluster2Former model leverages scribble-based annotations for training, significantly reducing the need for comprehensive pixel-level masks. We augment a video instance segmenter, for example, the Mask2Former architecture, with similarity-based constraint loss to handle partial annotations efficiently. We demonstrate that despite using lightweight annotations (using only 0.5% of the annotated pixels), Cluster2Former achieves competitive performance on standard benchmarks. The approach offers a cost-effective and computationally efficient solution for video instance segmentation, especially in scenarios with limited annotation resources. [ABSTRACT FROM AUTHOR]

Pattern formation in microbial colonies of competing strains under purely space-limited population growth has recently attracted considerable research interest. We show that the reproduction time statistics of individuals has a significant impact on the sectoring patterns. Generalizing the standard Eden growth model, we introduce a simple one-parameter family of reproduction time distributions indexed by the variation coefficient &dgr; € [0,1], which includes deterministic (&dgr; = 0) and memory-less exponential distribution (&dgr; = 1) as extreme cases. We present convincing numerical evidence and heuristic arguments that the generalized model is still in the Kardar-Parisi-Zhang (KPZ) universality class, and the changes in patterns are due to changing prefactors in the scaling relations, which we are able to predict quantitatively. With the example of Saccharomyces cerevisiae, we show that our approach using the variation coefficient also works for more realistic reproduction time distributions. [ABSTRACT FROM AUTHOR]

Abstract: The hydrodynamics of viscoelastic materials (for example, polymer melts and solutions) presents interesting and complex phenomena, for example, instabilities and turbulent flow at very low Reynolds numbers due to normal stress effects and the existence of a finite stress relaxation time. This present work is motivated by renewed interest in instabilities in polymer flow. The majority of currently used numerical methods discretize a constitutive equation on a grid with finite difference or similar methods. We present work in progress in which we simulate viscoelastic flow with dissipative particle dynamics. The advantage of this approach is that many of the numerical instabilities of conventional methods can be avoided, and that the model gives clear physical insight into the origins of many viscoelastic flow instabilities. [Copyright &y& Elsevier]

We investigate diffusion-limited aggregation (DLA) in a wedge geometry. Arneodo and collaborators have suggested that the ensemble average of DLA cluster density should be close to the noise-free selected Saffman-Taylor finger. We show that a different, but related, ensemble average, that of the conformal maps associated with random clusters, yields a nontrivial shape which is also not far from the Saffman-Taylor finger. However, we have previously demonstrated that the same average of DLA in a channel geometry is not the Saffman-Taylor finger. This casts doubt on the idea that the average of noisy diffusion-limited growth is governed by a simple transcription of noise-free results. [ABSTRACT FROM AUTHOR]

Diffusion-limited aggregation is consistent with simple scaling. However, strong subdominant terms are present, and these can account for various earlier claims of anomalous scaling. We show this in detail for the case of multiscaling. [Copyright &y& Elsevier]

Force networks form the skeleton of static granular matter. They are the key factor that determines mechanical properties such as stability, elasticity and sound transmission, which are important for civil engineering and industrial processing. Previous studies have focused on investigations of the global structure of external forces (the boundary condition) and on the probability distribution of individual contact forces. So far, however, precise knowledge of the disordered spatial structure of the force network has remained elusive. Here we report that molecular dynamics simulations of realistic granular packings reveal scale invariance of clusters of particles interacting by means of relatively strong forces. Despite visual variation, force networks for various values of the confining pressure and other parameters have identical scaling exponents and scaling function, thereby determining a universality class. Unexpectedly, the flat ensemble of force configurations (a simple generalization of equilibrium statistical mechanics) belongs to this universality class, whereas some widely studied simplified models do not. This implies that the elasticity of the grains and their geometrical disorder do not affect the universal mechanical properties. [ABSTRACT FROM AUTHOR]

The evolution of wide shear zones or shear bands was investigated experimentally and numerically for quasistatic dry granular flows in split bottom shear cells. We compare the behavior of materials consisting of beads, irregular grains, such as sand, and elongated particles. Shearing an initially random sample, the zone width was found to significantly decrease in the first stage of the process. The characteristic shear strain associated with this decrease is about unity and it is systematically increasing with shape anisotropy, i.e., when the grain shape changes from spherical to irregular (e.g., sand) and becomes elongated (pegs). The strongly decreasing tendency of the zone width is followed by a slight increase which is more pronounced for rodlike particles than for grains with smaller shape anisotropy (beads or irregular particles). The evolution of the zone width is connected to shear-induced packing density change and for nonspherical particles it also involves grain reorientation effects. The final zone width is significantly smaller for irregular grains than for spherical beads. [ABSTRACT FROM AUTHOR]

The rheology of a 3-dimensional granular system consisting of frictional elongated particles was investigated by means of discrete element model calculations. A homogenous shear flow of frictional spherocyliders was simulated, and a number of rheological quantities were calculated. In the framework of the μ(I) rheology, the effective friction was found to be a non-monotonic function of the aspect ratio for interparticle friction coefficient μp ≲ 0.4, while it was an increasing function for larger μp. We reveal the microscopic origin of this peculiar non-monotonic behaviour. We show the non-trivial dependence of the velocity fluctuations on the dissipation regime, and trace back the behaviour of the normal stress differences to particle-level quantities. [ABSTRACT FROM AUTHOR]

Many growth processes lead to intriguing stochastic patterns and complex fractal structures which exhibit local scale invariance properties. Such structures can often be described effectively by space-time trajectories of interacting particles, and their large scale behavior depends on the overall growth geometry. We establish an exact relation between statistical properties of structures in uniformly expanding and fixed geometries, which preserves the local scale invariance and is independent of other properties such as the dimensionality. This relation generalizes standard conformal transformations as the natural symmetry of self-affine growth processes. We illustrate our main result numerically for various structures of coalescing Lrvy flights and fractional Brownian motions, including also branching and finite particle sizes. One of the main benefits of this approach is a full, explicit description of the asymptotic statistics in expanding domains, which are often nontrivial and random due to amplification of initial fluctuations. [ABSTRACT FROM AUTHOR]

The alignment, ordering, and rotation of elongated granular particles was studied in shear flow. The time evolution of the orientation of a large number of particles was monitored in laboratory experiments by particle tracking using optical imaging and x-ray computed tomography. The experiments were complemented by discrete element simulations: The particles develop an orientational order. In the steady state the time- and ensemble-averaged direction of the main axis of the particles encloses a small angle with the streamlines! This shear alignment angle is independent of the applied shear rate, and it decreases with increasing grain aspect ratio. At the grain level the steady state is characterized by a net rotation of the particles, as dictated by the shear flow. The distribution of particle rotational velocities was measured both in the steady state and also during the initial transients. The average rotation speed of particles with their long axis perpendicular to the shear alignment angle is larger, while shear aligned particles rotate slower. The ratio of this fast/slow rotation increases with particle aspect ratio. During the initial transient starting from an unaligned initial condition, particles having an orientation just beyond the shear alignment angle rotate opposite to the direction dictated by the shear flow. [ABSTRACT FROM AUTHOR]

Shear induced alignment of elongated particles is studied experimentally and numerically. We show that shear alignment of ensembles of macroscopic particles is comparable even on a quantitative level to simple molecular systems, despite the completely different types of particle interactions. We demonstrate that for dry elongated grains the preferred orientation forms a small angle with the streamlines, independent of shear rate across three decades. For a given particle shape, this angle decreases with increasing aspect ratio of the particles. The shear-induced alignment results in a considerable reduction of the effective friction of the granular material. [ABSTRACT FROM AUTHOR]

The outflow of hard grains from storage containers with narrow outlets has been extensively studied in the past. Most experiments focused on discharge rates and avalanche statistics. Flow fields inside such containers have been detected optically in two-dimensional (2D) or quasi-2D geometries. Soft grains behave qualitatively different in many respects, both in their static packing properties and during silo discharge. We employ X-ray computed tomography to map the particles in a 3D container and we compare the static packing characteristics and the flow profiles of soft hydrogel spheres with those of hard spheres. The local fill fraction of the soft grains depends upon the depth below the granular surface. The outflow of the soft, low frictional hydrogel spheres involves the complete container volume, stagnant zones are absent. [ABSTRACT FROM AUTHOR]

We study the rheology of dense granular flows for frictionless spherocylinders by means of 3D numerical simulations. As in the case of spherical particles, the effective friction μ is an increasing function of the inertial number I, and we systematically investigate the dependence of μ on the particle aspect ratio Q, as well as that of the normal stress differences, the volume fraction, and the coordination number. We show in particular that the quasistatic friction coefficient is nonmonotonic with Q: from the spherical case Q=1, it first sharply increases, reaches a maximum around Q≃1.05, and then gently decreases until Q=3, passing its initial value at Q≃2. We provide a microscopic interpretation for this unexpected behavior through the analysis of the distribution of dissipative contacts around the particles: as compared to spheres, slightly elongated grains enhance contacts in their central cylindrical band, whereas at larger aspect ratios particles tend to align and dissipate by preferential contacts at their hemispherical caps. [ABSTRACT FROM AUTHOR]