Search

Machine learning (ML) methods provide advanced means for understanding inherent patterns within large and complex datasets. Here, we employ the principal component analysis (PCA) and the diffusion map (DM) techniques to evaluate the glass transition temperature ($T_\mathrm{g}$) from low-dimensional representations of all-atom molecular dynamic (MD) simulations of polylactide (PLA) and poly(3-hydroxybutyrate) (PHB). Four molecular descriptors were considered: radial distribution functions (RDFs), mean square displacements (MSDs), relative square displacements (RSDs), and dihedral angles (DAs). By applying a Gaussian Mixture Model (GMM) to analyze the PCA and DM projections, and by quantifying their log-likelihoods as a density-based metric, a distinct separation into two populations corresponding to melt and glass states was revealed. This separation enabled the $T_\mathrm{g}$ evaluation from a cooling-induced sharp increase in the overlap between log-likelihood distributions at different temperatures. $T_\mathrm{g}$ values derived from the RDF and MSD descriptors using DM closely matched the standard computer simulation-based dilatometric and dynamic $T_\mathrm{g}$ values for both PLA and PHB models. This was not the case for PCA. The DM-transformed DA and RSD data resulted in $T_\mathrm{g}$ values in agreement with experimental ones. Overall, the fusion of atomistic simulations and diffusion maps complemented with the Gaussian Mixture Models presents a promising framework for computing $T_\mathrm{g}$ and studying the glass transition in a unified way across various molecular descriptors for glass-forming materials.

We use lattice-Boltzmann molecular dynamics (LBMD) simulations to study the compression of a confined polymer immersed in a fluid and pushed by a large spherical colloid with a diameter comparable to the channel width. We examined the chain's deformation with both purely repulsive and weakly attractive Lennard-Jones (LJ) potentials applied between the monomers. The sphere's velocity was varied over 3 orders of magnitude. The chain is in a non-dense state at low pushing velocities for both repulsive and attractive monomer interactions. When the velocity of the spherical colloid exceeds a threshold $v^*$, the back end of the chain transitions to a high density state with low mean square monomer displacement (MSD) values. The front end, however, remains in a non-dense state with high MSD indicating a pseudo two-state coexistence. This crossover is also revealed through volume per monomer and MSD as a function of the sphere's velocity. We also studied polymer dynamics by investigating folding events at different times., Comment: 17 pages, 11 figures

Electromagnetically propelled helical nanoswimmers offer great potential for nanorobotic applications. Here, the effect of confinement on their propulsion is characterized using lattice-Boltzmann simulations. Two principal mechanisms give rise to their forward motion under confinement: 1) pure swimming, and 2) the thrust created by the differential pressure due to confinement. Under strong confinement, they face greater rotational drag, but display a faster propulsion for fixed driving frequency in agreement with experimental findings. This is due to the increased differential pressure created by the boundary walls when they are sufficiently close to each other and the particle. Two new analytical relations are presented: 1) for predicting the swimming speed of an unconfined particle as a function of its angular speed and geometrical properties, and 2) an empirical expression to accurately predict the propulsion speed of a confined swimmer as a function of the degree of confinement and its unconfined swimming speed. At low driving frequencies and degrees of confinement, the systems retain the expected linear behavior consistent with the predictions of the Stokes equation. However, as the driving frequency and/or the degree of confinement increase, their impact on propulsion leads to increasing deviations from the Stokesian regime and emergence of nonlinear behavior.

The molten sand, a mixture of calcia, magnesia, alumina, and silicate, known as CMAS, is characterized by its high viscosity, density, and surface tension. The unique properties of CMAS make it a challenging material to deal with in high-temperature applications, requiring innovative solutions and materials to prevent its buildup and damage to critical equipment. Here, we use multiphase many-body dissipative particle dynamics (mDPD) simulations to study the wetting dynamics of highly viscous molten CMAS droplets. The simulations are performed in three dimensions, with varying initial droplet sizes and equilibrium contact angles. We propose a coarse parametric ordinary differential equation (ODE) that captures the spreading radius behavior of the CMAS droplets. The ODE parameters are then identified based on the Physics-Informed Neural Network (PINN) framework. Subsequently, the closed form dependency of parameter values found by PINN on the initial radii and contact angles are given using symbolic regression. Finally, we employ Bayesian PINNs (B-PINNs) to assess and quantify the uncertainty associated with the discovered parameters. In brief, this study provides insight into spreading dynamics of CMAS droplets by fusing simple parametric ODE modeling and state-of-the-art machine learning techniques.

Topologically interlocking architectures can generate tough ceramics with attractive thermo-mechanical properties. This concept can make the material design pathway a challenging task, since modeling the whole design space is neither effective nor feasible. We propose an approach to design high-performance architectured ceramics using machine learning (ML) with data from finite element analysis (FEA). Convolutional neural networks (CNNs) and Multilayer Perceptrons (MLPs) are used as the deep learning approaches. A limited set of FEA simulation data containing a variety of architectural design parameters is used to train our neural networks, including learning how independent and dependent design parameters are related. A trained network is then used to predict the optimum structure from the configurations. A FEA simulation is run on the best predictions of both MLP and CNN algorithms to evaluate the performance of our networks. Although a limited amount of simulation data are available, our networks are effective in predicting the transient thermo-mechanical responses of possible panel designs. For example, the optimal design after using CNN prediction resulted in $\approx \! 30\%$ improvement in terms of edge temperature.

We propose a framework and an algorithm to uncover the unknown parts of nonlinear equations directly from data. The framework is based on eXtended Physics-Informed Neural Networks (X-PINNs), domain decomposition in space-time, but we augment the original X-PINN method by imposing flux continuity across the domain interfaces. The well-known Allen-Cahn equation is used to demonstrate the approach. The Frobenius matrix norm is used to evaluate the accuracy of the X-PINN predictions and the results show excellent performance. In addition, symbolic regression is employed to determine the closed form of the unknown part of the equation from the data, and the results confirm the accuracy of the X-PINNs based approach. To test the framework in a situation resembling real-world data, random noise is added to the datasets to mimic scenarios such as the presence of thermal noise or instrument errors. The results show that the framework is stable against significant amount of noise. As the final part, we determine the minimal amount of data required for training the neural network. The framework is able to predict the correct form and coefficients of the underlying dynamical equation when at least 50\% data is used for training.

Tissue growth kinetics and interface dynamics depend on the properties of the tissue environment and cell-cell interactions. In cellular environments, substrate heterogeneity and geometry arise from a variety factors, such as the structure of the extracellular matrix and nutrient concentration. We used the CellSim3D model, a kinetic division simulator, to investigate the growth kinetics and interface roughness dynamics of epithelial tissue growth on heterogeneous substrates with varying topologies. The results show that the presence of quenched disorder has a clear effect on the colony morphology and the roughness scaling of the interface in the moving interface regime. In a medium with quenched disorder, the tissue interface has a smaller interface roughness exponent, $\alpha$, and a larger growth exponent, $\beta$. The scaling exponents also depend on the topology of the substrate and cannot be categorized by well-known universality classes.

Radio wave propagation simulations based on the ray-optical approximation have been widely adopted in coverage analysis for a range of situations, including the outdoor-to-indoor scenario. This work presents O2I ray-tracing simulations utilizing a complete office building floor plan in the form of a laser-scanned point cloud. The simulated radio channels are compared to their measured counterparts at 4 and 14 GHz in terms of path loss and delay and angular spreads. Validation of channel simulations for the O2I case is rare, and so far non-existent for above-$6$~GHz bands. This work reveals the importance of a floor plan model in accurately simulating the channel; it is confirmed that path loss can be replicated with a simple interior path loss model in place of a detailed building interior model, but neglecting to model the interior results in high delay and angular spread errors. By modeling the interior, the ray-tracing simulations achieve relative mean error of under 10% for delay and angular spreads. Finally, effects of multi-layer insulating window on propagation simulations are reported. Noticeable variation of the penetration loss on a small change of the incident angle of a propagation path causes large changes in estimated coverage.

Laser machining is a highly flexible non-contact manufacturing technique that has been employed widely across academia and industry. Due to nonlinear interactions between light and matter, simulation methods are extremely crucial, as they help enhance the machining quality by offering comprehension of the inter-relationships between the laser processing parameters. On the other hand, experimental processing parameter optimization recommends a systematic, and consequently time-consuming, investigation over the available processing parameter space. An intelligent strategy is to employ machine learning (ML) techniques to capture the relationship between picosecond laser machining parameters for finding proper parameter combinations to create the desired cuts on industrial-grade alumina ceramic with deep, smooth and defect-free patterns. Laser parameters such as beam amplitude and frequency, scanner passing speed and the number of passes over the surface, as well as the vertical distance of the scanner from the sample surface, are used for predicting the depth, top width, and bottom width of the engraved channels using ML models. Owing to the complex correlation between laser parameters, it is shown that Neural Networks (NN) are the most efficient in predicting the outputs. Equipped with an ML model that captures the interconnection between laser parameters and the engraved channel dimensions, one can predict the required input parameters to achieve a target channel geometry. This strategy significantly reduces the cost and effort of experimental laser machining during the development phase, without compromising accuracy or performance. The developed techniques can be applied to a wide range of ceramic laser machining processes.

One of the main questions regarding complex systems at large scales concerns the effective interactions and driving forces that emerge from the detailed microscopic properties. Coarse-grained models aim to describe complex systems in terms of coarse-scale equations with a reduced number of degrees of freedom. Recent developments in machine learning (ML) algorithms have significantly empowered the discovery process of the governing equations directly from data. However, it remains difficult to discover partial differential equations (PDEs) with high-order derivatives. In this paper, we present new data-driven architectures based on multi-layer perceptron (MLP), convolutional neural network (CNN), and a combination of CNN and long short-term memory (CNN-LSTM) structures for discovering the non-linear equations of motion for phase-field models with non-conserved and conserved order parameters. The well-known Allen--Cahn, Cahn--Hilliard, and the phase-field crystal (PFC) models were used as the test cases. Two conceptually different types of implementations were used: (a) guided by physical intuition (such as local dependence of the derivatives) and (b) in the absence of any physical assumptions (black-box model). We show that not only can we effectively learn the time derivatives of the field in both scenarios, but we can also use the data-driven PDEs to propagate the field in time and achieve results in good agreement with the original PDEs.

Universality of interfacial roughness in growing epithelial tissue has remained a controversial issue. Kardar-Parisi-Zhang (KPZ) and Molecular Beam Epitaxy (MBE) universality classes have been reported among other behaviors including total lack of universality. Here, we utilize a kinetic division model for deformable cells to investigate cell-colony scaling. With seemingly minor model changes, it can reproduce both KPZ- and MBE-like scaling in configurations that mimic the respective experiments. This result neutralizes the apparent scaling controversy. It can be speculated that this diversity in growth behavior is beneficial for efficient evolution and versatile growth dynamics.

Polarization-dependent scattering anisotropy of cylindrical nanowires has numerous potential applications in, for example, nanoantennas, photothermal therapy, thermophotovoltaics, catalysis, sensing, optical filters and switches. In all these applications, temperature-dependent material properties play an important role and often adversely impact performance depending on the dominance of either radiative or dissipative damping. Here, we employ numerical modeling based on Mie scattering theory to investigate and compare the temperature and polarization-dependent optical anisotropy of metallic (gold, Au) nanowires with indirect (silicon, Si) and direct (gallium arsenide, GaAs) bandgap semiconducting nanowires. Results indicate that plasmonic scattering resonances in semiconductors, within the absorption band, deteriorate with an increase in temperature whereas those occurring away from the absorption band strengthen as a result of the increase in phononic contribution. Indirect-bandgap thin ($20 \,\mathrm{nm}$) Si nanowires present low absorption efficiencies for both the transverse electric (TE, $E_{\perp}$) and magnetic (TM, $E_{\parallel}$) modes, and high scattering efficiencies for the TM mode at shorter wavelengths making them suitable as highly efficient scatterers. Temperature-resilient higher-order anapole modes with their characteristic high absorption and low scattering efficiencies are also observed in the semiconductor nanowires ($r \! = \! 125 \! - \! 130$ nm) for the TE polarization. Herein, the GaAs nanowires present $3 \! - \! 7$ times greater absorption efficiencies compared to the Si nanowires making them especially suitable for temperature-resilient applications such as scanning near-field optical microscopy (SNOM), localized heating, non-invasive sensing or detection that require strong localization of energy in the near field., Comment: 42 pages(37 article pages + 5 pages Supplementary Information); 13 Figures (9 article figures + 4 Supplementary Information figures)

Polarizable force fields are gradually becoming a common choice for ionic soft matter, in particular for molecular dynamics (MD) simulations of ionic liquids (ILs) and deep eutectic solvents (DESs). The CL&Pol force field introduced in 2019 is the first general, transferable and polarizable force field for MD simulations of different types of DESs. The original formulation contains, however, some problems that appear in simulations of ethaline and may also have a broader impact. First, the originally proposed atomic diameter parameters are unbalanced, resulting in too weak interactions between the chlorides and the hydroxyl groups of the ethylene glycol molecules. This, in turn, causes an artificial phase separation in long simulations. Second, there is an overpolarization of chlorides due to strong induced dipoles that give rise to the presence of peaks and antipeaks at very low $q$-vector values (\SI{2.4}{\per\nano\meter}) in the partial components of the structure factors. In physical terms, this is equivalent to overestimated spatial nano-scale heterogeneity. To correct these problems, we adjusted the chloride-hydroxyl radial distribution functions against \textit{ab initio} data and then extended the use of the Tang-Toennis damping function for the chlorides' induced dipoles. These adjustments correct the problems without losing the robustness of the CL\&Pol force field. The results were also compared with the non-polarizable version, the CL&P force field. We expect that the corrections will facilitate reliable use of the CL&Pol force field for other types of DESs.

This work develops a new open source API and software package called \textit{SymPhas} for simulations of phase-field, phase-field crystal and reaction-diffusion models, supporting up to three dimensions and an arbitrary number of fields. \textit{SymPhas} delivers two novel program capabilities: 1) User specification of models from the associated dynamical equations in an unconstrained form and 2) extensive support for integrating user-developed discrete-grid-based numerical solvers into the API. The capability to specify general phase-field models is primarily achieved by developing a novel symbolic algebra functionality that can formulate mathematical expressions at compile time, is able to apply rules of symbolic algebra such as distribution, factoring and automatic simplification, and support user-driven expression tree manipulation. A modular design based on the CC++ template meta-programming paradigm is applied to the symbolic algebra library and general API implementation to minimize application runtime and increase the accessibility of the API for third party development. \textit{SymPhas} is written in C/CC++ and emphasizes high-performance capabilities via parallelization with OpenMP and the CC++ standard library. \textit{SymPhas} is equipped with a forward Euler solver and a semi-implicit Fourier spectral solver. Sample implementations and simulations of several phase-field models are presented, generated using the semi-implicit Fourier spectral solver.

Since the invention of cinema, the manipulated videos have existed. But generating manipulated videos that can fool the viewer has been a time-consuming endeavor. With the dramatic improvements in the deep generative modeling, generating believable looking fake videos has become a reality. In the present work, we concentrate on the so-called deepfake videos, where the source face is swapped with the targets. We argue that deepfake detection task should be viewed as a screening task, where the user, such as the video streaming platform, will screen a large number of videos daily. It is clear then that only a small fraction of the uploaded videos are deepfakes, so the detection performance needs to be measured in a cost-sensitive way. Preferably, the model parameters also need to be estimated in the same way. This is precisely what we propose here.

A discrete-to-continuum transformation to model 2-D discrete lattices as energetically equivalent 1-D continuum beams is developed. The study is initiated in a classical setting but results in a non-classical two-scale micropolar beam model via a novel link within a unit cell between the second-order macrorotation-gradient and the micropolar antisymmetric shear deformation. The shear deformable micropolar beam is reduced to a couple-stress and two classical lattice beam models by successive approximations. The stiffness parameters for all models are given by the micropolar constitutive matrix. The four models are compared by studying stretching- and bending-dominated lattice core sandwich beams under various loads and boundary conditions. A classical 4th-order Timoshenko beam is an apt first choice for stretching-dominated beams, whereas the 6th-order micropolar model works for bending-dominated beams as well. The 6th-order couple-stress beam is often too stiff near point loads and boundaries. It is shown that the 1-D micropolar model leads to the exact 2-D lattice response in the absence of boundary effects even when the length of the 1-D beam (macrostructure) equals that of the 2-D unit cell (microstructure), that is, when L=l., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Action grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures. The preprint has been replaced by the accepted manuscript

Compact layers containing embedded semiconductor particles consolidated using pulsed electric current sintering exhibit intense, broadband near-infrared reflectance. The composites consolidated from nano- or micro-silica powder have a different porous microstructure which causes scattering at the air-matrix interface and larger reflectance primarily in the visible region. The 3 mm thick composite compacts reflect up to 72% of the incident radiation in the near-infrared region with a semiconductor microinclusion volume fraction of 1% which closely matches predictions from multiscale Monte Carlo modeling and Kubelka-Munk theory. Further, the calculated spectra predict an improvement of the reflectance by decreasing the average particle size or broadening the standard deviation. The high reflectance is achieved with minimal dissipative losses and facile manufacturing, and the composites described herein are well-suited to control the radiative transfer of heat in devices at high temperature and under harsh conditions., Comment: 9 pages, 5 figures

Half-metallic ferromagnets show 100% spin-polarization at the Fermi level and are ideal candidates for spintronic applications. Despite the extensive research in the field, very few materials have been discovered so far. Here we present results of electronic band structure calculations based on density functional theory and extensive physical-property measurements for Fe3Se4 revealing signatures of half-metallicity. The spin-polarized electronic band structure calculations predict half-metallic ferrimagnetism for Fe3Se4. The electrical resistivity follows exponentially suppressed electron-magnon scattering mechanism in the low-temperature regime and show a magnetoresistance effect that changes the sign from negative to positive with decreasing temperature around 100 K. Other intriguing observations include the anomalous behavior of Hall resistance below 100 K and an anomalous Hall coefficient that roughly follows the \r{ho}2 behavior., Comment: Accepted

Directing the propagation of near-infrared radiation is a major concern in improving the efficiency of solar cells and thermal insulators. A facile approach to scatter light in the near-infrared region without excessive heating is to embed compact layers with semiconductor particles. The directional scattering by semiconductor@oxide (core@shell) spherical particles (containing Si, InP, TiO$_2$, SiO$_2$, or ZrO$_2$) with a total radius varying from 0.1 to 4.0 {\mu}m and in an insulating medium at low volume fraction is investigated using Lorenz-Mie theory and multiscale modelling. The optical response of each layers is calculated under irradiation by the sun or a blackbody emitter at 1180 K. Reflectance efficiency factors of up to 83.7% and 63.9% are achieved for near-infrared solar and blackbody radiation in 200 {\mu}m thick compact layers with only 1% volume fraction of bare Si particles with a radius of 0.23 {\mu}m and 0.50 {\mu}m, respectively. The maximum solar and blackbody efficiency factors of layers containing InP particles was slightly less (80.2% and 60.7% for bare particles with a radius of 0.25 {\mu}m and 0.60 {\mu}m, respectively). The addition of an oxide coating modifies the surrounding dielectric environment, which improves the solar reflectance efficiency factor to over 90% provided it matches the scattering mode energies with the incident spectral density. The layers are spectrally-sensitive and can be applied as a back or front reflector for solar devices, high temperature thermal insulators, and optical filters in Gradient Heat Flux Sensors for fire safety applications., Comment: 10 pages, 7 figures

MineRL 2019 competition challenged participants to train sample-efficient agents to play Minecraft, by using a dataset of human gameplay and a limit number of steps the environment. We approached this task with behavioural cloning by predicting what actions human players would take, and reached fifth place in the final ranking. Despite being a simple algorithm, we observed the performance of such an approach can vary significantly, based on when the training is stopped. In this paper, we detail our submission to the competition, run further experiments to study how performance varied over training and study how different engineering decisions affected these results., Comment: To appear in Post Proceedings of the Competitions & Demonstrations Track @ NeurIPS2019. Source code available at https://github.com/Miffyli/minecraft-bc

We investigate morphologies of proliferating cellular tissues using a newly developed numerical simulation model for mechanical cell division and migration in 2D. The model is applied to a bimodal mixture consisting of stiff cells with a low growth potential and soft cells with a high growth potential; cancer cells are typically considered to be softer than healthy cells. In an even mixture, the soft cells develop into a tissue matrix and the stiff cells into a dendrite-like network structure. When soft cells are placed inside a tissue consisting of stiff cells (to model cancer growth), the soft cells develop to a fast growing tumor-like structure that gradually evacuates the stiff cell matrix. The model also demonstrates 1) how soft cells orient themselves in the direction of the largest effective stiffness as predicted by the theory of Bischofs and Schwarz (Proc. Natl. Acad. Sci U.S.A., 100, 9274--9279 (2003) and 2) that the orientation and force generation continue a few cell rows behind the soft-stiff interface. With increasing inter-cell friction, tumor growth slows down and cell death occurs. The contact force distribution between cells is demonstrated to be highly sensitive to cell type mixtures and cell-cell interactions, which indicates that local mechanical forces can be useful as a regulator of tissue formation. The results shed new light on established experimental data., Comment: arXiv admin note: text overlap with arXiv:1811.07573

A displacement-based, geometrically nonlinear finite element model is developed for lattice core sandwich panels modeled as 2-D equivalent single-layer (ESL), first-order shear deformation theory (FSDT) micropolar plates. The nonlinearity is due to the moderate macrorotations of the plate which are modeled by including the von Karman strains in the micropolar strain measures. Weak form Galerkin method with linear Lagrange interpolations is used to develop the displacement-based finite element model. Selective reduced integration is used to eliminate shear locking and membrane locking. The novel finite element model is used to study the nonlinear bending and linear free vibrations of web-core and pyramid core sandwich panels. Clamped and free edge boundary conditions are considered for the first time for the 2-D micropolar ESL-FSDT plate theory. The present 2-D finite element results are in good agreement with the corresponding detailed 3-D FE results for the lattice core sandwich panels. The 2-D element provides computationally cost-effective solutions; in a nonlinear bending example, the number of elements required for the 2-D micropolar plate is of the order 10^3, whereas for the corresponding 3-D model the order is 10^5., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures. The preprint has been replaced by the accepted manuscript

We have investigated the structural properties, vibrational spectra, and electronic band structures of crystalline cellulose allomorphs and chemically modified cellulose with quantum chemical methods. The electronic band gaps of cellulose allomorphs I$_\alpha$, I$_\beta$, II, and III$_1$ lie in the range of 5.0 to 5.6 eV. We show that extra states can be created in the band gap of cellulose by chemical modification. Experimentally feasible amidation of cellulose I$_\beta$ with aniline or 4,4'- diaminoazobenzene creates narrow bands in the cellulose band gap, reducing the difference between the occupied and empty states to 4.0 or 1.8 eV, respectively. The predicted states 4,4'-diaminoazobenzene-modified cellulose I$_\beta$ fall in the visible spectrum, suggesting uses in optical applications., Comment: 15 pages + 11 pages of supporting information

'Oriented external electric fields (OEEFs)' have been shown to have great potential in being able to provide unprecedented control of chemical reactions, catalysis and selectivity with applications ranging from H2 storage to molecular machines. We report a theoretical study of the atomic origins of molecular changes due to OEEFs; understanding the characteristics of OEEF-induced couplings between atomic and molecular properties is an important step toward comprehensive understanding of the effects of strong external fields on molecular structure, stability, and reactivity. We focus on the atomic and molecular (bond) properties of a set of homo- (H2, N2, O2, F2, and Cl2) and hetero-diatomic (HF, HCl, CO, and NO) molecules under intense external electric fields in the context of quantum theory of atoms in molecules (QTAIM). It is shown that atomic properties (atomic charges and energies, and localization index) correlate linearly with the field strengths, but molecular properties (bond length, electron density at bond critical point, bond length, and electron delocalization index) exhibit non-linear responses to the imposed fields. In particular, the changes in the electron density distribution alter the shapes and locations of the zero-flux surfaces, atomic volumes, atomic electron population, and localization/delocalization indices. At the molecular level, the topography and topology of the molecular electrostatic potential undergo dramatic changes. The external fields also perturb the covalent-polar-ionic characteristic of the studied chemical bonds, hallmarking the impact of electric fields on the stability and reactivity of chemical compounds. The findings are well-rationalized within the framework of the quantum theory of atoms in molecules and form a coherent conceptual understanding of these effects in prototypical molecules such as diatomics.

Deep learning models for semantics are generally evaluated using naturalistic corpora. Adversarial methods, in which models are evaluated on new examples with known semantic properties, have begun to reveal that good performance at these naturalistic tasks can hide serious shortcomings. However, we should insist that these evaluations be fair -that the models are given data sufficient to support the requisite kinds of generalization. In this paper, we define and motivate a formal notion of fairness in this sense. We then apply these ideas to natural language inference by constructing very challenging but provably fair artificial datasets and showing that standard neural models fail to generalize in the required ways; only task-specific models that jointly compose the premise and hypothesis are able to achieve high performance, and even these models do not solve the task perfectly.

A 2-D micropolar equivalent single-layer (ESL), first-order shear deformation (FSDT) plate model for 3-D web-core sandwich panels is developed. First, a 3-D web-core unit cell is modeled by classical shell finite elements. A discrete-to-continuum transformation is applied to the microscale unit cell and its strain and kinetic energy densities are expressed in terms of the macroscale 2-D plate kinematics. The hyperelastic constitutive relations and the equations of motion (via Hamilton's principle) for the plate are derived by assuming energy equivalence between the 3-D unit cell and the 2-D plate. The Navier solution is developed for the 2-D micropolar ESL-FSDT plate model to study the bending, buckling, and free vibration of simply-supported web-core sandwich panels. In a line load bending problem, a 2-D classical ESL-FSDT plate model yields displacement errors of 34-175% for face sheet thicknesses of 2-10 mm compared to a 3-D FE solution, whereas the 2-D micropolar model gives only small errors of 2.7-3.4% as it can emulate the 3-D deformations better through non-classical antisymmetric shear behavior and local bending and twisting., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Action grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures. The preprint has been replaced by the accepted manuscript

MAX CLIQUE problem (MCP) is an NPO problem, which asks to find the largest complete sub-graph in a graph $G, G = (V, E)$ (directed or undirected). MCP is well known to be $NP-Hard$ to approximate in polynomial time with an approximation ratio of $1 + \epsilon$, for every $\epsilon > 0$ [9] (and even a polynomial time approximation algorithm with a ratio $n^{1 - \epsilon}$ has been conjectured to be non-existent [2] for MCP). Up to this date, the best known approximation ratio for MCP of a polynomial time algorithm is $O(n(log_2(log_2(n)))^2 / (log_2(n))^3)$ given by Feige [1]. In this paper, we show that MCP can be approximated with a constant factor in polynomial time through approximation ratio preserving reductions from MCP to MAX DNF and from MAX DNF to MIN SAT. A 2-approximation algorithm for MIN SAT was presented in [6]. An approximation ratio preserving reduction from MIN SAT to min vertex cover improves the approximation ratio to $2 - \Theta(1/ \sqrt{n})$ [10]. Hence we prove false the infamous conjecture, which argues that there cannot be a polynomial time algorithm for MCP with an approximation ratio of any constant factor., Comment: the reduction does not preserve the approximation ratio

Mapping states to actions in deep reinforcement learning is mainly based on visual information. The commonly used approach for dealing with visual information is to extract pixels from images and use them as state representation for reinforcement learning agent. But, any vision only agent is handicapped by not being able to sense audible cues. Using hearing, animals are able to sense targets that are outside of their visual range. In this work, we propose the use of audio as complementary information to visual only in state representation. We assess the impact of such multi-modal setup in reach-the-goal tasks in ViZDoom environment. Results show that the agent improves its behavior when visual information is accompanied with audio features.

Deep reinforcement learning has proven to be successful for learning tasks in simulated environments, but applying same techniques for robots in real-world domain is more challenging, as they require hours of training. To address this, transfer learning can be used to train the policy first in a simulated environment and then transfer it to physical agent. As the simulation never matches reality perfectly, the physics, visuals and action spaces by necessity differ between these environments to some degree. In this work, we study how general video games can be directly used instead of fine-tuned simulations for the sim-to-real transfer. Especially, we study how the agent can learn the new action space autonomously, when the game actions do not match the robot actions. Our results show that the different action space can be learned by re-training only part of neural network and we obtain above 90% mean success rate in simulation and robot experiments., Comment: Two first authors contributed equally. Accepted by ICASSP 2020

A geometrically nonlinear sandwich beam model founded on the modified couple stress Timoshenko beam theory with K\'arm\'an kinematics is derived and employed in the analysis of periodic sandwich structures. The constitutive model is based on the mechanical behavior of sandwich beams, with the bending response split into membrane-induced and local bending modes. A micromechanical approach based on the structural analysis of a unit cell is derived and utilized to obtain the stiffness properties of selected prismatic cores. The model is shown to be equivalent to the classical thick-face sandwich theory for the same basic assumptions. A two-node finite element interpolated with linear and cubic shape functions is proposed and its stiffness and geometric stiffness matrices are derived. Three examples illustrate the model capabilities in predicting deflections, stresses and critical buckling loads of elastic sandwich beams including elastic size effects. Good agreement is obtained throughout in comparisons with more involved finite element models., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Action grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures

In this work, we present the results of a wideband measurement campaign at 60 GHz conducted inside a Linkker electric city bus. Targeting prospective millimeter-wave (mmWave) public transportation wearable scenarios, we mimic a typical deployment of mobile high-end consumer devices in a dense environment. Specifically, our intra-vehicle deployment includes one receiver and multiple transmitters corresponding to a mmWave access point and passengers' wearable and hand-held devices. While the receiver is located in the front part of the bus, the transmitters repeat realistic locations of personal devices (i) at the seat level (e.g., a hand-held device) and (ii) at a height 70 cm above the seat (e.g., a wearable device: augmented reality glasses or a head-mounted display). Based on the measured received power, we construct a logarithmic model for the distance-dependent path loss. The parametrized models developed in the course of this study have the potential to become an attractive ground for the link budget estimation and interference footprint studies in crowded public transportation scenarios., Comment: 4 pages, 8 figures, 1 table, accepted to EuCAP 2019

A geometrically nonlinear finite element model is developed for the bending analysis of micropolar Timoshenko beams using the principle of virtual displacements and linear Lagrange interpolation functions. The nonlinearity enters the model via a nonlinear von K\'arm\'an strain term that allows the micropolar beam to undergo moderate rotations. The nonlinear micropolar Timoshenko beam is used as an equivalent single layer model to study four different lattice core sandwich beams. A two-scale energy method is used to derive the micropolar constitutive equations for web, hexagonal, Y-frame and corrugated core topologies. Various bending cases are studied numerically using the developed 1-D finite element model. Reduced integration techniques are used to overcome the shear and membrane locking. The present 1-D results are in good agreement with the corresponding 2-D finite element beam frame results for global bending., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Action grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures. The preprint has been replaced by the accepted manuscript

We investigate morphologies of proliferating cellular tissue using a newly developed numerical simulation model for mechanical cell division. The model reproduces structures of simple multi-cellular organisms via simple rules for selective division and division plane orientation. The model is applied to a bimodal mixture of stiff cells with a low growth potential and soft cells with a high growth potential. In an even mixture, the soft cells develop into a tissue matrix and the stiff cells into a dendrite-like network structure. For soft cell inclusion in a stiff cellular matrix, the soft cells develop to a fast growing tumour like structure that gradually evacuates the stiff cell matrix. With increasing inter-cell friction, the tumour growth slows down and parts of it is driven to self-inflicted cell death.

Standard evaluations of deep learning models for semantics using naturalistic corpora are limited in what they can tell us about the fidelity of the learned representations, because the corpora rarely come with good measures of semantic complexity. To overcome this limitation, we present a method for generating data sets of multiply-quantified natural language inference (NLI) examples in which semantic complexity can be precisely characterized, and we use this method to show that a variety of common architectures for NLI inevitably fail to encode crucial information; only a model with forced lexical alignments avoids this damaging information loss.

A new five point potential for liquid water, TIP5P/2018, is presented along with the techniques used to derive its charges from ab initio per-molecule electrostatic potentials in the liquid phase using the split charge equilibration (SQE) of Nistor et al. [J. Chem. Phys. 125, 094108 (2006)]. By taking the density and diffusion dependence on temperature as target properties, significant improvements to the behavior of isothermal compressibility were achieved along with improvements to other thermodynamic and rotational properties. While exhibiting a dipole moment close to ab initio values, TIP5P/2018 suffers from a too small quadrupole moment due to the charge assignment procedure and results in an overestimation of the dielectric constant.

Constitutive equations are derived for a 1-D micropolar Timoshenko beam made of a web-core lattice material. First, a web-core unit cell is modeled by discrete classical constituents, i.e., the Euler-Bernoulli beam finite elements (FE). A discrete-to-continuum transformation is applied to the microscale unit cell and its strain energy density is expressed in terms of the macroscale 1-D beam kinematics. Then the constitutive equations for the micropolar web-core beam are derived assuming strain energy equivalence between the microscale unit cell and the macroscale beam. A micropolar beam FE model for static and dynamic problems is developed using a general solution of the beam equilibrium equations. A localization method for the calculation of periodic classical beam responses from micropolar results is given. The 1-D beam model is used in linear bending and vibration problems of 2-D web-core sandwich panels that have flexible joints. Localized 1-D results are shown to be in good agreement with experimental and 2-D FE beam frame results., Comment: This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Action grant agreement No 745770 - SANDFECH - Micromechanics-based finite element modeling of sandwich structures. The preprint has been replaced by the accepted manuscript

A number of applications in nanoplasmonics utilize noble metals, gold (Au) and silver (Ag), as the materials of choice. However, these materials suffer from problems of poor thermal and chemical stability accompanied by significant dissipative losses under high-temperature conditions. In this regard, semiconductor nanoparticles have attracted attention with their promising characteristics of highly tunable plasmonic resonances, low ohmic losses and greater thermochemical stability. Here, we investigate the size-dependent thermoplasmonic properties of semiconducting silicon and gallium arsenide nanoparticles to compare them with metallic Au nanoparticles using Mie theory. To this end, we employ experimentally estimated models of dielectric permittivity in our computations. Among the various permittivity models for Au, we further compare the Drude-Lorentz (DL) and the Drude and critical points (DCP) models. Results show a redshift in the scattering and absorption resonances for the DL model while the DCP model presents a blueshift. The dissipative damping in the semiconductor nanoparticles is strongest for the sharp electric octupole resonances followed by the quadrupole and dipole modes. However, a reverse order with strongest values for the broad dipole resonance is observed for the Au nanoparticles. A massive Drude broadening contributes strongly to the damping of resonances in Au nanoparticles at elevated temperatures. In contrast, the semiconductor nanoparticles do not exhibit any significant deterioration in their scattering and absorption resonances at high temperatures. In combination with low dissipative damping, this makes the semiconductor nanoparticles better suited for high-temperature applications in nanoplasmonics wherein the noble metals suffer from excessive heating., Comment: 31 pages (20 Main Article + 11 Supplementary Information); 21 Figures (8 Main Article + 13 Supplementary Information)

In this paper, we parameterize an excess loss-based multipath component (MPC) cross-polarization ratio (XPR) model in indoor and outdoor environments for above-6 GHz frequency bands. The results are based on 28 measurement campaigns in several frequency bands ranging from 15 to 80 GHz. A conventional XPR model of an MPC assuming a constant mean value fits our measurements very poorly and moreover overestimates the depolarization effect. Our measurements revealed a clear trend that the MPC XPR is inversely proportional to an excess loss in reference to the free-space path loss. The model is physically sound as a higher excess loss is attributed to more lossy interactions or to a greater number of interactions with objects, leading to a greater chance of depolarization. The measurements furthermore showed that the MPC XPR is not strongly frequency or environment dependent. In our MPC XPR model, an MPC with zero-dB excess loss has a mean XPR of 28 dB. The mean XPR decreases half-a-dB as the excess loss increases by every dB and the standard deviation around the mean is 6 dB. The model is applicable to existing channel models to reproduce realistic MPC XPRs for the above 6-GHz radio links., Comment: Submitted to IEEE Transactions on Wireless Communications

In this paper, we propose a new technique for the future fifth generation cellular network wireless backhauling. We show that hundreds of bits per second per Hertz (bits per second per Hz) of spectral efficiency can be attained at a high carrier frequency (such as 26 GHz) between large antenna arrays deployed along structures (such as lamp posts) that are close and roughly parallel to each other. Hundreds of data streams are spatially multiplexed through a short range and line of sight massive multiple input massive multiple output propagation channel thanks to a new low complexity spatial multiplexing scheme, called block discrete Fourier transform based spatial multiplexing with maximum ratio transmission. Its performance in real and existing environments is assessed using accurate ray-tracing tools and antenna models. In the best simulated scenario, 1.6 kbits per second per Hz of spectral efficiency is attained, corresponding to 80% of Singular Value Decomposition performance, with a transmitter and a receiver that are 200 and 10,000 times less complex, respectively., Comment: draft version of IEEE Globecom 2016 paper