Search

Predicting future evolution based on incomplete information of the past is still a challenge even though data-driven machine learning approaches have been successfully applied to forecast complex nonlinear dynamics. The widely adopted reservoir computing (RC) can hardly deal with this since it usually requires complete observations of the past. In this paper, a scheme of RC with (D+1)-dimension input and output (I/O) vectors is proposed to solve this problem, i.e., the incomplete input time series or dynamical trajectories of a system, in which certain portion of states are randomly removed. In this scheme, the I/O vectors coupled to the reservoir are changed to (D+1)-dimension, where the first D dimensions store the state vector as in the conventional RC, and the additional dimension is the corresponding time interval. We have successfully applied this approach to predict the future evolution of the logistic map and Lorenz, Rössler, and Kuramoto-Sivashinsky systems, where the inputs are the dynamical trajectories with missing data. The dropoff rate dependence of the valid prediction time (VPT) is analyzed. The results show that it can make forecasting with much longer VPT when the dropoff rate θ is lower. The reason for the failure at high θ is analyzed. The predictability of our RC is determined by the complexity of the dynamical systems involved. The more complex they are, the more difficult they are to predict. Perfect reconstructions of chaotic attractors are observed. This scheme is a pretty good generalization to RC and can treat input time series with regular and irregular time intervals. It is easy to use since it does not change the basic architecture of conventional RC. Furthermore, it can make multistep-ahead prediction just by changing the time interval in the output vector into a desired value, which is superior to conventional RC that can only do one-step-ahead forecasting based on complete regular input data.

Current sheets are spatially localized almost-one-dimensional (1D) structures with intense plasma currents. They play a key role in storing the magnetic field energy and they separate different plasma populations in planetary magnetospheres, the solar wind, and the solar corona. Current sheets are primary regions for the magnetic field line reconnection responsible for plasma heating and charged particle acceleration. One of the most interesting and widely observed types of 1D current sheets is the rotational discontinuity, which can be force-free or include plasma compression. Theoretical models of such 1D current sheets are based on the assumption of adiabatic motion of ions, i.e., ion adiabatic invariants are conserved. We focus on three current sheet configurations, widely observed in the Earth magnetopause and magnetotail and in the near-Earth solar wind. The magnetic field in such current sheets is supported by currents carried by transient ions, which exist only when there is a sufficient number of invariants. In this paper, we apply a machine learning approach, AI Poincaré, to determine parametrical domains where adiabatic invariants are conserved. For all three current sheet configurations, these domains are quite narrow and do not cover the entire parametrical range of observed current sheets. We discuss possible interpretation of obtained results indicating that 1D current sheets are dynamical rather than static plasma equilibria.

This paper addresses the challenge of extracting meaningful information from measured bioelectric signals generated by complex, large scale physiological systems such as the brain or the heart. We focus on a combination of the well-known Laplacian eigenmaps machine learning approach with dynamical systems ideas to analyze emergent dynamic behaviors. The method reconstructs the abstract dynamical system phase-space geometry of the embedded measurements and tracks changes in physiological conditions or activities through changes in that geometry. It is geared to extract information from the joint behavior of time traces obtained from large sensor arrays, such as those used in multiple-electrode ECG and EEG, and explore the geometrical structure of the low dimensional embedding of moving time windows of those joint snapshots. Our main contribution is a method for mapping vectors from the phase space to the data domain. We present cases to evaluate the methods, including a synthetic example using the chaotic Lorenz system, several sets of cardiac measurements from both canine and human hearts, and measurements from a human brain.

In social animals, the presence of conflicts of interest or multiple leaders can promote the emergence of two or more subgroups. Such subgroups are easily recognizable by human observers, yet a quantitative and objective measure of group fragmentation is currently lacking. In this paper, we explore the feasibility of detecting group fragmentation by embedding the raw data from the individuals' motions on a low-dimensional manifold and analyzing the topological features of this manifold. To perform the embedding, we employ the ISOMAP algorithm, which is a data-driven machine learning tool extensively used in computer vision. We implement this procedure on a data set generated by a modified la à Vicsek model, where agents are partitioned into two or more subsets and an independent leader is assigned to each subset. The dimensionality of the embedding manifold is shown to be a measure of the number of emerging subgroups in the selected observation window and a cluster analysis is proposed to aid the interpretation of these findings. To explore the feasibility of using this approach to characterize group fragmentation in real time and thus reduce the computational cost in data processing and storage, we propose an interpolation method based on an inverse mapping from the embedding space to the original space. The effectiveness of the interpolation technique is illustrated on a test-bed example with potential impact on the regulation of collective behavior of animal groups using robotic stimuli. [ABSTRACT FROM AUTHOR]

Machine learning is a broad discipline that comprises a variety of techniques for extracting meaningful information and patterns from data. It draws on knowledge and "know-how" from various scientific areas such as statistics, graph theory, linear algebra, databases, mathematics, and computer science. Recently, materials scientists have begun to explore data mining ideas for discovery in materials. In this paper we explore the power of these methods for studying binary compounds that are well characterized and are often used as a test bed. By mining properties of the constituent atoms, three materials research relevant tasks, namely, separation of a number of compounds into subsets in terms of their crystal structure, grouping of an unknown compound into the most characteristically similar peers (in one instance, 100% accuracy is achieved), and specific property prediction (the melting point), are explored. [ABSTRACT FROM AUTHOR]

Single-particle trajectories measured in microscopy experiments contain important information about dynamic processes occurring in a range of materials including living cells and tissues. However, extracting that information is not a trivial task due to the stochastic nature of the particles' movement and the sampling noise. In this paper, we adopt a deep-learning method known as a convolutional neural network (CNN) to classify modes of diffusion from given trajectories. We compare this fully automated approach working with raw data to classical machine learning techniques that require data preprocessing and extraction of human-engineered features from the trajectories to feed classifiers like random forest or gradient boosting. All methods are tested using simulated trajectories for which the underlying physical model is known. From the results it follows that CNN is usually slightly better than the feature-based methods, but at the cost of much longer processing times. Moreover, there are still some borderline cases in which the classical methods perform better than CNN. [ABSTRACT FROM AUTHOR]

The precise reconstruction of jet transverse momenta in heavy-ion collisions is a challenging task. A major obstacle is the large number of (mainly) low-pT particles overlaying the jets. Strong region-to-region fluctuations of this background complicate the jet measurement and lead to significant uncertainties. In this paper, a novel approach to correct jet momenta (or energies) for the underlying background in heavy-ion collisions is introduced. The proposed method makes use of common machine learning techniques to estimate the jet transverse momentum based on several parameters, including properties of the jet constituents. Using a toy model and HIJING simulations, the performance of the new method is shown to be superior to the established standard area-based background estimator. The application of the new method to data promises the measurement of jets down to extremely low transverse momenta, unprecedented thus far in data on heavy-ion collisions. [ABSTRACT FROM AUTHOR]

Systems far from equilibrium respond to probes in a history-dependent manner. The prediction of the system response depends on either knowing the details of that history or being able to characterize all the current system properties. In crystal plasticity, various processing routes contribute to a history dependence that may manifest itself through complex microstructural deformation features with large strain gradients. However, the complete spatial strain correlations may provide further predictive information. In this paper, we demonstrate an explicit example where spatial strain correlations can be used in a statistical manner to infer and classify prior deformation history at various strain levels. The statistical inference is provided by machine-learning techniques. As source data, we consider uniaxially compressed crystalline thin films generated by two dimensional discrete dislocation plasticity simulations, after prior compression at various levels. Crystalline thin films at the nanoscale demonstrate yield-strength size effects with very noisy mechanical responses that produce a serious challenge to learning techniques. We discuss the influence of size effects and structural uncertainty to the ability of our approach to distinguish different plasticity regimes. [ABSTRACT FROM AUTHOR]

In confined systems, such as the inside of a biological cell, the outer boundary or wall can affect the dynamics of internal particles. In many cases of interest both the internal particle and outer wall are approximately spherical. Therefore, quantifying the wall effects from an outer spherical boundary on the motion of an internal eccentric sphere is very useful. However, when the two spheres are not concentric, the problem becomes nontrivial. In this paper we improve existing analytical methods to evaluate these wall effects and then train a feed-forward artificial neural network within a broader model. The final model generally performed with ~0.001% error within the training domain and ~0.05% when the outer spherical wall was extrapolated to an infinite plane. Through this model, the wall effects of an outer spherical boundary on the arbitrary motion of an internal sphere for all experimentally achievable configurations can now be conveniently and efficiently determined. [ABSTRACT FROM AUTHOR]

In this paper, we apply machine learning methods to study phase transitions in certain statistical mechanical models on the two-dimensional lattices, whose transitions involve nonlocal or topological properties, including site and bond percolations, the XY model, and the generalized XY model. We find that using just one hidden layer in a fully connected neural network, the percolation transition can be learned and the data collapse by using the average output layer gives correct estimate of the critical exponent ν. We also study the Berezinskii-Kosterlitz-Thouless transition, which involves binding and unbinding of topological defects, vortices and antivortices, in the classical XY model. The generalized XY model contains richer phases, such as the nematic phase, the paramagnetic and the quasi-long-range ferromagnetic phases, and we also apply machine learning method to it. We obtain a consistent phase diagram from the network trained with only data along the temperature axis at two particular parameter Δ values, where Δ is the relative weight of pure XY coupling. Aside from using the spin configurations (either angles or spin components) as the input information in a convolutional neural network, we devise a feature engineering approach using the histograms of the spin orientations in order to train the network to learn the three phases in the generalized XY model and demonstrate that it indeed works. The trained network by using system size L×L can be used to the phase diagram for other sizes (L′×L′, where L′≠L) without any further training. [ABSTRACT FROM AUTHOR]

Efficient and automated classification of phases from minimally processed data is one goal of machine learning in condensed-matter and statistical physics. Supervised algorithms trained on raw samples of microstates can successfully detect conventional phase transitions via learning a bulk feature such as an order parameter. In this paper, we investigate whether neural networks can learn to classify phases based on topological defects. We address this question on the two-dimensional classical XY model which exhibits a Kosterlitz-Thouless transition. We find significant feature engineering of the raw spin states is required to convincingly claim that features of the vortex configurations are responsible for learning the transition temperature. We further show a single-layer network does not correctly classify the phases of the XY model, while a convolutional network easily performs classification by learning the global magnetization. Finally, we design a deep network capable of learning vortices without feature engineering. We demonstrate the detection of vortices does not necessarily result in the best classification accuracy, especially for lattices of less than approximately 1000 spins. For larger systems, it remains a difficult task to learn vortices. [ABSTRACT FROM AUTHOR]

Machine learning is capable of discriminating phases of matter, and finding associated phase transitions, directly from large data sets of raw state configurations. In the context of condensed matter physics, most progress in the field of supervised learning has come from employing neural networks as classifiers. Although very powerful, such algorithms suffer from a lack of interpretability, which is usually desired in scientific applications in order to associate learned features with physical phenomena. In this paper, we explore support vector machines (SVMs), which are a class of supervised kernel methods that provide interpretable decision functions. We find that SVMs can learn the mathematical form of physical discriminators, such as order parameters and Hamiltonian constraints, for a set of two-dimensional spin models: the ferromagnetic Ising model, a conserved-order-parameter Ising model, and the Ising gauge theory. The ability of SVMs to provide interpretable classification highlights their potential for automating feature detection in both synthetic and experimental data sets for condensed matter and other many-body systems. [ABSTRACT FROM AUTHOR]

While high-throughput density functional theory (DFT) has become a prevalent tool for materials discovery, it is limited by the relatively large computational cost. In this paper, we explore using DFT data from high-throughput calculations to create faster, surrogate models with machine learning (ML) that can be used to guide new searches. Our method works by using decision tree models to map DFT-calculated formation enthalpies to a set of attributes consisting of two distinct types: (i) composition-dependent attributes of elemental properties (as have been used in previous ML models of DFT formation energies), combined with (ii) attributes derived from the Voronoi tessellation of the compound's crystal structure. The ML models created using this method have half the cross-validation error and similar training and evaluation speeds to models created with the Coulomb matrix and partial radial distribution function methods. For a dataset of 435 000 formation energies taken from the Open Quantum Materials Database (OQMD), our model achieves a mean absolute error of 80 meV/atom in cross validation, which is lower than the approximate error between DFT-computed and experimentally measured formation enthalpies and below 15% of the mean absolute deviation of the training set. We also demonstrate that our method can accurately estimate the formation energy of materials outside of the training set and be used to identify materials with especially large formation enthalpies. We propose that our models can be used to accelerate the discovery of new materials by identifying the most promising materials to study with DFT at little additional computational cost. [ABSTRACT FROM AUTHOR]