Search

Artificial neural networks (ANNs) have emerged as an essential tool in machine learning, achieving remarkable success across diverse domains, including image and speech generation, game playing, and robotics. However, there exist fundamental differences between ANNs' operating mechanisms and those of the biological brain, particularly concerning learning processes. This paper presents a comprehensive review of current brain-inspired learning representations in artificial neural networks. We investigate the integration of more biologically plausible mechanisms, such as synaptic plasticity, to improve these networks' capabilities. Moreover, we delve into the potential advantages and challenges accompanying this approach. In this review, we pinpoint promising avenues for future research in this rapidly advancing field, which could bring us closer to understanding the essence of intelligence. [ABSTRACT FROM AUTHOR]

This paper reports an all-optical vector magnetometer enhanced by a machine learning model. Using a dual probing beam setup, spin projections in both probe directions are simultaneously detected. Vector information is directly obtained from the measured phases of spin projection signals. To enhance the measurement accuracy and mitigate the dead zone effect, we introduce an artificial neural network (ANN) to link the phase signals to the field direction. With the addition of amplitude information to the ANN input, the average angle error is reduced to less than 0.3 ° within a hemisphere. Furthermore, this configuration demonstrates a field angle sensitivity of better than 30 μ rad / Hz 1 / 2 . [ABSTRACT FROM AUTHOR]

The wide diversity of species and the remarkable variation in morphological features that allow plants to adapt to a wide range of terrestrial environments is a fact that highlights the fundamental and crucial role of plants in the field of biodiversity studies. Currently, research on leaf classification is limited and in its early stages. A novel classification system based on laser-induced breakdown spectroscopy (LIBS) technology was proposed in this paper, integrated with machine learning for real-time, in situ detection and analysis of leaves. Four representative leaf samples—Ilex chinensis, Camellia japonica, Cinnamomum camphora, and Osmanthus fragrans—were subjected to spectral analysis and machine learning techniques. Spectral analysis revealed distinct spectral lines corresponding to elements such as Ca, Al, Mg, Na, and Fe, alongside common elements including C, N, and O. Principal component analysis (PCA) was employed to reduce the dimensionality of the spectral data, and the first 13 principal components used in this study captured 98.76% of the total variance. Following this, support vector machine (SVM), backpropagation artificial neural network and convolutional neural network (CNN) algorithms were applied for machine learning on the principal components to develop leaf recognition classification models. Through comparison, the CNN algorithm, boasting a classification accuracy of up to 94.44%, was ultimately selected. The models established by SVM and back propagation artificial neural network achieved accuracy of only 83.33% and 90.00%, respectively. The results suggest that integrating LIBS with machine learning is an effective and precise approach for leaf classification, offering promising applications in biodiversity research. [ABSTRACT FROM AUTHOR]

Tabulation of chemical mechanisms with artificial neural networks (ANNs) offers significant speed benefits when computing the real-time integration of reaction source terms in turbulent reacting flow simulations. In such approaches, the ANNs should be physically consistent with the reaction mechanism by conserving mass and chemical elements, as well as obey the bounds of species mass fractions. In the present paper, a method is developed for satisfying these constraints to machine precision. The method can be readily applied to any reacting system and appended to the existing ANN architectures. To satisfy the conservation laws, certain species in a reaction mechanism are selected as residual species and recalculated after ANN predictions of all of the species have been made. Predicted species mass fractions are set to be bounded. While the residual species mass fractions are not guaranteed to be non-negative, it is shown that negative predictions can be avoided in almost all cases and easily rectified if necessary. The ANN method with conservation is applied to one-dimensional laminar premixed flame simulations, and comparisons are made with simulations performed with direct integration (DI) of chemical kinetics. The ANNs with conservation are shown to satisfy the conservation laws for every reacting point to machine precision and, furthermore, to provide results in better agreement with DI than ANNs without conservation. It is, thus, shown that the proposed method reduces accumulation of errors and positively impacts the overall accuracy of the ANN prediction at negligible additional computational cost. [ABSTRACT FROM AUTHOR]

Effects of the magnetic field and inertia on the onset of thermal convection in a horizontal bidispersive porous layer, rotating about a vertical axis, are analyzed. The Darcy equation with same temperature in the micro- and macrophases is used to characterize the fluid motion. The Vadasz number is taken into account in a generalized Darcy equation for the macrophase. The eigenvalue problem obtained from the linear stability analysis is solved analytically for free–free boundaries. Moving one step further from the traditional linear stability analysis, machine learning tools are introduced in this paper to include the effect of multiple parameters on the marginal state of the system. Machine learning techniques have been implemented to identify the mode of instability with respect to different parameters. In particular, classification algorithms, namely, Artificial Neural Networks (ANN) and Support vector machine, are used to examine the onset of oscillatory convection and stationary convection. The required data for training of the algorithms are generated from the results of linear stability analysis. It is found that ANN with the sufficient number of hidden layers along with good choice of training dataset can predict the mode of instability even on the small variation in a given parameter. The combined effect of rotation, magnetic field, and inertia is to reduce the oscillatory mode of instability; hence, the system exhibits the steady mode of instability for a significant region in the three dimensional space comprising the Taylor number, the Hartman number, and the Vadasz number. [ABSTRACT FROM AUTHOR]

Machine learning is a useful tool when extracting hidden information from complex measurement data obtained via surface analysis, as in secondary ion mass spectrometry. Flexible learning methods often require significant effort to adjust parameters, as these parameters may have a significant effect on results. However, machine learning methods enable the extraction of new information that cannot be found by manual analysis. This paper presents some examples of complex data analyses using conventional multivariate analysis methods based on linear combinations (principal component analysis and multivariate curve resolution), an unsupervised learning method based on artificial neural networks (sparse autoencoder), and a supervised learning method based on decision trees (random forest). To obtain reproducible and useful results from machine learning applications to surface analysis data, the preparation of data sets—including the selection of variables and the raw data conversion process—is crucial. Moreover, sufficient information representing analytical purposes, such as the chemical structures of unknown samples, material types, and physical or chemical properties of particular materials, must be contained in the data set for supervised learning. [ABSTRACT FROM AUTHOR]

Multifunctional biological neural networks exploit multistability in order to perform multiple tasks without changing any network properties. Enabling artificial neural networks (ANNs) to obtain certain multistabilities in order to perform several tasks, where each task is related to a particular attractor in the network's state space, naturally has many benefits from a machine learning perspective. Given the association to multistability, in this paper, we explore how the relationship between different attractors influences the ability of a reservoir computer (RC), which is a dynamical system in the form of an ANN, to achieve multifunctionality. We construct the "seeing double" problem in order to systematically study how a RC reconstructs a coexistence of attractors when there is an overlap between them. As the amount of overlap increases, we discover that for multifunctionality to occur, there is a critical dependence on a suitable choice of the spectral radius for the RC's internal network connections. A bifurcation analysis reveals how multifunctionality emerges and is destroyed as the RC enters a chaotic regime that can lead to chaotic itinerancy. [ABSTRACT FROM AUTHOR]

As the proportion of photovoltaic (PV) power generation rapidly increases, accurate PV output power prediction becomes more crucial to energy efficiency and renewable energy production. There are numerous approaches for PV output power prediction. Many researchers have previously summarized PV output power prediction from different angles. However, there are relatively few studies that use machine learning methods as a means to conduct a separate review of PV output power prediction. This review classifies machine learning methods from different perspectives and provides a systematic and critical review of machine learning methods for recent PV output power applications in terms of the temporal and spatial scales of prediction and finds that the artificial neural network and support vector machine are used much more frequently than other methods. In addition, this study examines the differences between the output power prediction of individual PV plants and regional PV stations and the benefits of regional PV plant prediction, while this paper presents some performance evaluation matrices commonly used for PV output power prediction. In addition, to further improve the accuracy of machine learning methods for PV output power prediction, some researchers suggest preprocessing the input data of the prediction models or considering hybrid machine learning methods. Furthermore, the potential advantages of machine model optimization for prediction performance improvement are discussed and explored in detail. [ABSTRACT FROM AUTHOR]

In this paper, we develop a multivariate regression model and a neural network model to predict the Reynolds number (Re) and Nusselt number in turbulent thermal convection. We compare their predictions with those of earlier models of convection: Grossmann–Lohse [Phys. Rev. Lett. 86, 3316 (2001)], revised Grossmann–Lohse [Phys. Fluids 33, 015113 (2021)], and Pandey–Verma [Phys. Rev. E 94, 053106 (2016)] models. We observe that although the predictions of all the models are quite close to each other, the machine-learning models developed in this work provide the best match with the experimental and numerical results. [ABSTRACT FROM AUTHOR]

Improving the fireproof performance of polymers is crucial for ensuring human safety and enabling future space colonization. However, the complexity of the mechanisms for flame retardant and the need for customized material design pose significant challenges. To address these issues, we propose a machine learning (ML) framework based on substructure fingerprinting and self-enforcing deep neural networks (SDNN) to predict the fireproof performance of flame-retardant epoxy resins. Our model is based on a comprehensive understanding of the physical mechanisms of materials and can predict fireproof performance and eliminate the needs for properties descriptors, making it more convenient than previous ML models. With a dataset of only 163 samples, our SDNN models show an average prediction error of 3% for the limited oxygen index (LOI). They also provide satisfactory predictions for the peak of heat release rate PHR and total heat release (THR), with coefficient of determination (R2) values of 0.87 and 0.85, respectively, and average prediction errors less than 17%. Our model outperforms the support vector model SVM for all three indices, making it a state-of-the-art study in the field of flame retardancy. We believe that our framework will be a valuable tool for the design and virtual screening of flame retardants and will contribute to the development of safer and more efficient polymer materials. [ABSTRACT FROM AUTHOR]

Targeted free energy perturbation uses an invertible mapping to promote configuration space overlap and the convergence of free energy estimates. However, developing suitable mappings can be challenging. Wirnsberger et al. [J. Chem. Phys. 153, 144112 (2020)] demonstrated the use of machine learning to train deep neural networks that map between Boltzmann distributions for different thermodynamic states. Here, we adapt their approach to the free energy differences of a flexible bonded molecule, deca-alanine, with harmonic biases and different spring centers. When the neural network is trained until "early stopping"—when the loss value of the test set increases—we calculate accurate free energy differences between thermodynamic states with spring centers separated by 1 Å and sometimes 2 Å. For more distant thermodynamic states, the mapping does not produce structures representative of the target state, and the method does not reproduce reference calculations. [ABSTRACT FROM AUTHOR]

Seminal gas discharge experiments of the late 19th and early 20th centuries laid the foundations of modern physics, and the influence of this "golden era" continues to resonate well into the 21st century through modern technologies, medical applications, and fundamental scientific investigations. Key to this continuing success story has been the kinetic equation formulated by Ludwig Boltzmann in 1872, which provides the theoretical foundations necessary for analyzing such highly non-equilibrium situations. However, as discussed here, the full potential of Boltzmann's equation has been realized only in the past 50 years or so, with modern computing power and analytical techniques facilitating accurate solutions for various types of charged particles (ions, electrons, positrons, and muons) in gases. Our example of thermalization of electrons in xenon gas highlights the need for such accurate methods—the traditional Lorentz approximation is shown to be hopelessly inadequate. We then discuss the emerging role of Boltzmann's equation in determining cross sections by inverting measured swarm experiment transport coefficient data using machine learning with artificial neural networks. [ABSTRACT FROM AUTHOR]

Knowledge of the physical properties of ionic liquids (ILs), such as the surface tension and speed of sound, is important for both industrial and research applications. Unfortunately, technical challenges and costs limit exhaustive experimental screening efforts of ILs for these critical properties. Previous work has demonstrated that the use of quantum-mechanics-based thermochemical property prediction tools, such as the conductor-like screening model for real solvents, when combined with machine learning (ML) approaches, may provide an alternative pathway to guide the rapid screening and design of ILs for desired physiochemical properties. However, the question of which machine-learning approaches are most appropriate remains. In the present study, we examine how different ML architectures, ranging from tree-based approaches to feed-forward artificial neural networks, perform in generating nonlinear multivariate quantitative structure–property relationship models for the prediction of the temperature- and pressure-dependent surface tension of and speed of sound in ILs over a wide range of surface tensions (16.9–76.2 mN/m) and speeds of sound (1009.7–1992 m/s). The ML models are further interrogated using the powerful interpretation method, shapley additive explanations. We find that several different ML models provide high accuracy, according to traditional statistical metrics. The decision tree-based approaches appear to be the most accurate and precise, with extreme gradient-boosting trees and gradient-boosting trees being the best performers. However, our results also indicate that the promise of using machine-learning to gain deep insights into the underlying physics driving structure–property relationships in ILs may still be somewhat premature. [ABSTRACT FROM AUTHOR]

The co-location of memory and processing is a core principle of neuromorphic computing. A local memory device for synaptic weight storage has long been recognized as an enabling element for large-scale, high-performance neuromorphic hardware. In this work, we demonstrate programmable superconducting synapses with integrated memories for use in superconducting optoelectronic neural systems. Superconducting nanowire single-photon detectors and Josephson junctions are combined into programmable synaptic circuits that exhibit single-photon sensitivity, memory cells with more than 400 internal states, leaky integration of input spike events, and 0.4 fJ programming energies (including cooling power). These results are attractive for implementing a variety of supervised and unsupervised learning algorithms and lay the foundation for a new hardware platform optimized for large-scale spiking network accelerators. [ABSTRACT FROM AUTHOR]

Despite significant advances in deep neural networks across diverse domains, challenges persist in safety-critical contexts, including domain shift sensitivity and unreliable uncertainty estimation. To address these issues, this study investigates Bayesian learning for uncertainty handling in modern neural networks. However, the high-dimensional, non-convex nature of the posterior distribution poses practical limitations for epistemic uncertainty estimation. The Laplace approximation, as a cost-efficient Bayesian method, offers a practical solution by approximating the posterior as a multivariate normal distribution but faces computational bottlenecks in precise covariance matrix computation and storage. This research employs subnetwork inference, utilizing only a subset of the parameter space for Bayesian inference. In addition, a Kronecker-factored and low-rank representation is explored to reduce space complexity and computational costs. Several corrections are introduced to converge the approximated curvature to the exact Hessian matrix. Numerical results demonstrate the effectiveness and competitiveness of this method, whereas qualitative experiments highlight the impact of Hessian approximation granularity and parameter space utilization in Bayesian inference on mitigating overconfidence in predictions and obtaining high-quality uncertainty estimates. [ABSTRACT FROM AUTHOR]

Memristor crossbar arrays are expected to achieve highly energy-efficient neuromorphic computing via implementing parallel vector–matrix multiplication (VMM) in situ. The similarities between memristors and neural synapses offer opportunities for realizing hardware-based brain-inspired computing, such as spike neural networks. However, the nonlinear I–V characteristics of the memristors limit the implementation of parallel VMM on passive memristor crossbar arrays. In our work, we propose to utilize differential conductance as a synaptic weight to implement linear VMM operations on a passive memristor array in parallel. We fabricated a TiO2/HfO2 memristor crossbar array, in which differential-conductance-based synaptic weight exhibits plasticity, nonvolatility, multi-states, and tunable ON/OFF ratio. The noise-dependent accuracy performance of VMM operations based on the proposed approach was evaluated, offering an optimization guideline. Furthermore, we demonstrated a spike neural network circuit capable of processing small spiking signals through the differential-conductance-based synapses. The experimental results showcase effective space-coded and time-coded spike pattern recognition. Importantly, our work opens up new possibilities for the development of passive memristor arrays, leading to increased energy and area efficiency in brain-inspired chips. [ABSTRACT FROM AUTHOR]

Ionic liquids (ILs) are well classified as designer solvents based on the ease of tailoring their properties through modifying the chemical structure of the cation and anion. However, while many structure–property relationships have been developed, these generally only identify the most dominant trends. Here, we have used machine learning on existing experimental data to construct robust models to produce meaningful predictions across a broad range of cation and anion chemical structures. Specifically, we used previously collated experimental data for the viscosity and conductivity of protic ILs [T. L. Greaves and C. J. Drummond, Chem. Rev. 115, 11379–11448 (2015)] as the inputs for multiple linear regression and neural network models. These were then used to predict the properties of all 1827 possible cation–anion combinations (excluding the input combinations). These models included the effect of water content of up to 5 wt. %. A selection of ten new protic ILs was then prepared, which validated the usefulness of the models. Overall, this work shows that relatively sparse data can be used productively to predict physicochemical properties of vast arrays of ILs. [ABSTRACT FROM AUTHOR]

Al-rich Al–Ce alloys have the possibility of replacing heavier steel and cast irons for use in high-temperature applications. Knowledge about the structures and properties of Al–Ce alloys at the liquid state is vital for optimizing the manufacture process to produce desired alloys. However, reliable molecular dynamics simulation of Al–Ce alloy systems remains a great challenge due to the lack of accurate Al–Ce interatomic potential. Here, an artificial neural network (ANN) deep machine learning (ML) method is used to develop a reliable interatomic potential for Al–Ce alloys. Ab initio molecular dynamics simulation data on the Al–Ce liquid with a small unit cell (∼200 atoms) and on the known Al–Ce crystalline compounds are collected to train the interatomic potential using ANN-ML. The obtained ANN-ML model reproduces well the energies, forces, and atomic structure of the Al90Ce10 liquid and crystalline phases of Al–Ce compounds in comparison with the ab initio results. The developed ANN-ML potential is applied in molecular dynamics simulations to study the structures and properties of the metallic Al90Ce10 liquid, which would provide useful insight into the guiding experimental process to produce desired Al–Ce alloys. [ABSTRACT FROM AUTHOR]

Neuromuscular disorders (NMDs) affect various parts of a motor unit, such as the motor neuron, neuromuscular junction, and muscle fibers. Abnormal spontaneous activity (SA) is detected with electromyography (EMG) as an essential hallmark in diagnosing NMD, which causes fatigue, pain, and muscle weakness. Monitoring the effects of NMD calls for new smart devices to collect and classify EMG. Delay-based Reservoir Computing (DRC) is a neuromorphic algorithm with high efficiency in classifying sequential data. This work proposes a new DRC-based algorithm that provides a reference for medical education and training and a second opinion to clinicians to verify NMD diagnoses by detecting SA in muscles. With a sampling frequency of Fs = 64 kHz, we have classified SA with EMG signals of 1 s of muscle recordings. Furthermore, the DRC model of size N = 600 nodes has successfully detected SA signals against normal muscle activity with an accuracy of up to 90.7%. The potential of using neuromorphic processing approaches in point-of-care diagnostics, alongside the supervision of a clinician, provides a more comprehensive and reliable clinical profile. Our developed model benefits from the potential to be implemented in physical hardware to provide near-sensor edge computing. [ABSTRACT FROM AUTHOR]

Developing fast and accurate computational models to simulate intricate physical phenomena has been a persistent research challenge. Recent studies have demonstrated remarkable capabilities in predicting various physical outcomes through machine learning-assisted approaches. However, it remains challenging to generalize current methods, usually crafted for a specific problem, to other more complex or broader scenarios. To address this challenge, we developed graph neural network (GNN) models with enhanced generalizability derived from the distinct GNN architecture and neural operator techniques. As a proof of concept, we employ our GNN models to predict finite element (FE) simulation results for three-dimensional solid mechanics problems with varying boundary conditions. Results show that our GNN model achieves accurate and robust performance in predicting the stress and deformation profiles of structures compared with FE simulations. Furthermore, the neural operator embedded GNN approach enables learning and predicting various solid mechanics problems in a generalizable fashion, making it a promising approach for surrogate modeling. [ABSTRACT FROM AUTHOR]

In-memory computing (IMC) using emerging nonvolatile devices has received considerable attention due to its great potential for accelerating artificial neural networks and machine learning tasks. As the basic concept and operation modes of IMC are now well established, there is growing interest in employing its wide and general application. In this perspective, the path that leads memristive IMC to general-purpose machine learning is discussed in detail. First, we reviewed the development timeline of machine learning algorithms that employ memristive devices, such as resistive random-access memory and phase-change memory. Then we summarized two typical aspects of realizing IMC-based general-purpose machine learning. One involves a heterogeneous computing system for algorithmic completeness. The other is to obtain the configurable precision techniques for the compromise of the precision-efficiency dilemma. Finally, the major directions and challenges of memristive IMC-based general-purpose machine learning are proposed from a cross-level design perspective. [ABSTRACT FROM AUTHOR]

Numerous physics theories are rooted in partial differential equations (PDEs). However, the increasingly intricate physics equations, especially those that lack analytic solutions or closed forms, have impeded the further development of physics. Computationally solving PDEs by classic numerical approaches suffers from the trade-off between accuracy and efficiency and is not applicable to the empirical data generated by unknown latent PDEs. To overcome this challenge, we present KoopmanLab, an efficient module of the Koopman neural operator (KNO) family, for learning PDEs without analytic solutions or closed forms. Our module consists of multiple variants of the KNO, a kind of mesh-independent neural-network-based PDE solvers developed following the dynamic system theory. The compact variants of KNO can accurately solve PDEs with small model sizes, while the large variants of KNO are more competitive in predicting highly complicated dynamic systems govern by unknown, high-dimensional, and non-linear PDEs. All variants are validated by mesh-independent and long-term prediction experiments implemented on representative PDEs (e.g., the Navier–Stokes equation and the Bateman–Burgers equation in fluid mechanics) and ERA5 (i.e., one of the largest high-resolution global-scale climate datasets in earth physics). These demonstrations suggest the potential of KoopmanLab to be a fundamental tool in diverse physics studies related to equations or dynamic systems. [ABSTRACT FROM AUTHOR]

Memristor devices have been investigated for their properties of resistive modulation that can be used in data storage and brain-like computation elements as artificial synapses and neurons. Memristors are characterized by an onset of high current values under applied voltage that produces a transition to a low resistance state or successively to different stable states of increasing conductivity that implement synaptic weights. Here, we develop a nonlinear model to explain the variation with time of the voltage and the resistance and compare it to experimental results on ionic–electronic halide perovskite memristors. We find separate experimental signatures of the capacitive discharge and inductive current increase. We show that the capacitor produces an increase step of the resistance due to the influence of the series resistance. In contrast, the inductor feature associated with inverted hysteresis causes a decrease of the resistance, as observed experimentally. The chemical inductor feature dominates the potentiation effect in which the conductivity increases with the voltage stimulus. Our results enable a quantitative characterization of highly nonlinear electronic devices using a combination of techniques such as time transient decays and impedance spectroscopy. [ABSTRACT FROM AUTHOR]

This study evaluates the efficacy of two machine learning (ML) techniques, namely, artificial neural networks (ANNs) and gene expression programing (GEP), that use data-driven modeling to predict wall pressure spectra (WPS) underneath turbulent boundary layers. Different datasets of WPS from experiments and high-fidelity numerical simulations covering a wide range of pressure gradients and Reynolds numbers are considered. For both ML methods, an optimal hyperparameter environment is identified that yields accurate predictions. Despite a higher memory consumption, ANN models are faster to train and are much more accurate than the GEP models, yielding an order of magnitude lower logarithmic Mean Squared Error (lMSE) than GEP. Novel training schemes are devised to address the shortcomings of GEP. These include (a) ANN-assisted GEP to reduce the noise in the training data, (b) exploiting the low- and high-frequency trends to guide the GEP search, and (c) a stepped training strategy where the chromosomes are first trained on the canonical datasets, followed by the datasets with complex features. When compared to the baseline scheme, these training strategies accelerated convergence and resulted in models with superior accuracy (≈ 30 % reduction in the median lMSE) and higher reliability (≈ 75 % reduction in the spread of lMSE in the interquartile range). The final GEP models captured the complex trends of WPS across varying flow conditions and pressure gradients, surpassing the accuracy of Goody's model. [ABSTRACT FROM AUTHOR]

Although fuel spray parameters, such as spray cone angle and penetration length, are crucial for developing high-efficiency and high-performance combustion engines, general models for predicting transient characteristics of these parameters have not been suggested. In this study, the spray characteristics of cyclopentane at sub- and transcritical conditions relevant for IC engine and rocket injections were experimentally evaluated. A single simplified model for predicting the spray cone angles and spray penetration lengths over time was developed by adopting artificial neural networks (ANN). Spray measurements were conducted by shadowgraphy and Mie scattering techniques to recognize the phase change behavior of the spray, changing the injection and chamber conditions. The ANN model was developed using a multi-layer network with six normalized inputs and four outputs. It was trained with five transient spray datasets at two subcritical and three transcritical injection conditions. It was validated with one transcritical spray dataset. The ANN prediction was assessed, and the proposed approach represents the spray characteristics of cyclopentane at sub- and transcritical conditions. According to the model results, the predicted spray parameters are in good agreement with the experimental results over a useful pressure and temperature range of 40–55 bar and 465–564 K, mean absolute percentage errors of 2.25% (shadowgraphy) and 4.92% (Mie) for the spray angles, and 1.11% (shadowgraphy) and 3.44% (Mie) for the spray penetration lengths. Moreover, the developed ANN model can predict the penetration ratio, providing information on phase changes in sprays. The developed ANN model in this study is expected to become a universal model for transient spray characteristics near the critical point. By adding the database with various fuel types and spray conditions, the universal model can be used to develop high-efficiency and high-performance combustion engines or other combustors. [ABSTRACT FROM AUTHOR]

While there are several bottlenecks in hybrid organic–inorganic perovskite (HOIP) solar cell production steps, including composition screening, fabrication, material stability, and device performance, machine learning approaches have begun to tackle each of these issues in recent years. Different algorithms have successfully been adopted to solve the unique problems at each step of HOIP development. Specifically, high-throughput experimentation produces vast amount of training data required to effectively implement machine learning methods. Here, we present an overview of machine learning models, including linear regression, neural networks, deep learning, and statistical forecasting. Experimental examples from the literature, where machine learning is applied to HOIP composition screening, thin film fabrication, thin film characterization, and full device testing, are discussed. These paradigms give insights into the future of HOIP solar cell research. As databases expand and computational power improves, increasingly accurate predictions of the HOIP behavior are becoming possible. [ABSTRACT FROM AUTHOR]

Progress in hardware and algorithms for artificial intelligence (AI) has ushered in large machine learning models and various applications impacting our everyday lives. However, today's AI, mainly artificial neural networks, still cannot compete with human brains because of two major issues: the high energy consumption of the hardware running AI models and the lack of ability to generalize knowledge and self-adapt to changes. Neuromorphic systems built upon emerging devices, for instance, memristors, provide a promising path to address these issues. Although innovative memristor devices and circuit designs have been proposed for neuromorphic computing and applied to different proof-of-concept applications, there is still a long way to go to build large-scale low-power memristor-based neuromorphic systems that can bridge the gap between AI and biological brains. This Perspective summarizes the progress and challenges from memristor devices to neuromorphic systems and proposes possible directions for neuromorphic system implementation based on memristive devices. [ABSTRACT FROM AUTHOR]

Neural networks, trained with the ADAM algorithm followed by a globally convergent modification to Newton's method, are developed to predict the threshold gain of the fundamental and first higher-order modes as functions of the refractive-index profile in a quantum cascade laser cavity. The networks are used to optimize the design of a refractive-index profile that provides essentially single-spatial-mode performance in a nominally multi-moded cavity by maximizing the threshold-gain differential between the modes. The use of neural networks allows the optimization to be performed in seconds, instead of days or weeks which would be required if Maxwell's equations were repeatedly solved to obtain the threshold gains. [ABSTRACT FROM AUTHOR]

We present multiplexed gradient descent (MGD), a gradient descent framework designed to easily train analog or digital neural networks in hardware. MGD utilizes zero-order optimization techniques for online training of hardware neural networks. We demonstrate its ability to train neural networks on modern machine learning datasets, including CIFAR-10 and Fashion-MNIST, and compare its performance to backpropagation. Assuming realistic timescales and hardware parameters, our results indicate that these optimization techniques can train a network on emerging hardware platforms orders of magnitude faster than the wall-clock time of training via backpropagation on a standard GPU, even in the presence of imperfect weight updates or device-to-device variations in the hardware. We additionally describe how it can be applied to existing hardware as part of chip-in-the-loop training or integrated directly at the hardware level. Crucially, because the MGD framework is model-free it can be applied to nearly any hardware platform with tunable parameters, and its gradient descent process can be optimized to compensate for specific hardware limitations, such as slow parameter-update speeds or limited input bandwidth. [ABSTRACT FROM AUTHOR]

In this work, we explore the application of deep neural networks to the optimization of atomic layer deposition (ALD) processes. In particular, we focus on a one-shot optimization problem, where we try to predict the optimal dose time that leads to saturation everywhere in the reactor based on thickness values measured at different points of an ALD reactor after a single trial growth. In order to tackle this problem, we introduce a dataset designed to train neural networks to predict saturation times based on these inputs for a cross-flow ALD reactor. We then explore the predictive ability of artificial neural networks of different depths and sizes using a separate testing dataset to evaluate their accuracies. The results obtained show that networks trained using stochastic gradient descent methods can accurately predict saturation times without requiring any additional information on the surface kinetics. This provides a viable approach to minimize the number of experiments required to optimize new ALD processes in a known reactor, and it highlights the way machine learning can be leveraged for thin film growth and manufacturing. While the datasets and training procedure depend on the reactor geometry, the trained neural networks provide a general surrogate model connecting thickness values and trial dose times with optimal saturation times that can be reused for different ALD processes within the same reactor. [ABSTRACT FROM AUTHOR]

The recent decade has witnessed the advent of near-field radiative heat transfer (NFRHT) in a wide range of applications, including thermal photovoltaics and thermal diodes. However, the design process for these thermal devices has remained complex, often relying on the intuition and expertise of the designer. To address these challenges, a machine learning (ML) strategy based on the combination of an artificial neural network (ANN) and a genetic algorithm (GA) is presented. The ANN is trained to model representative scenarios, viz., NFRHT between metamaterials and NFRHT and thermal rectification between nanoparticles. The influence of different problem complexities, i.e., the number of input variables of function to be fitted, on effectiveness of the trained ANN is investigated. Test results show that ANNs can obtain the radiative heat flow and rectification ratio accurately and rapidly. Subsequently, physical parameters for the largest radiative heat flow and rectification ratio are determined by the utilization of GA on the trained ANN, and underlying mechanisms of deterministic optimum are discussed. Our work shows that data-driven ML methods are a powerful tool, which offers unprecedented opportunities for future NFRHT research. [ABSTRACT FROM AUTHOR]

In inertial confinement fusion, the ignition threshold factor (ITF), defined as the ratio of the available shell kinetic energy to the minimum ignition energy, is an important metric for quantifying how far an implosion is from its performance cliff. Traditional ITF research is based on analytical theories with explicit scaling laws and parameters obtained by numerically fitting simulation data. This present study uses machine learning (ML) methods to train implicit but more reliable ITF expressions. One-dimensional numerical simulations are used to develop a dataset with 20 000 targets, in which alpha particle heating magnifies the fusion yield by a factor of 6.5. These targets are defined as marginal ignition targets whose ITF equals unity. ML models such as neural networks, support vector machines, and Gaussian processes are trained to connect the minimum ignition velocity vigt with other implosion parameters, yielding an ML-based ITF of (v imp / v igt) 7.5 , where vimp represents the implosion velocity. Then, these ML models are used to obtain curves of the ignition probability vs the ITF and improved ignition cliffs that show considerably better accuracy than traditional scaling laws, which are observed. The results demonstrate that ML methods have promising application prospects for quantifying ignition margins and can be useful in optimizing ignition target designs and practical implosion experiments. [ABSTRACT FROM AUTHOR]

We analyze and compare various empirical models of wall pressure spectra beneath turbulent boundary layers and propose an alternative machine learning approach using Artificial Neural Networks (ANNs). The analysis and the training of the ANN are performed on data from experiments and high-fidelity simulations by various authors, covering a wide range of flow conditions. We present a methodology to extract all the turbulent boundary layer parameters required by these models, also considering flows experiencing strong adverse pressure gradients. Moreover, the database is explored to unveil important dependencies within the boundary layer parameters and to propose a possible set of features from which the ANN should predict the wall pressure spectra. The results show that the ANN outperforms traditional models in adverse pressure gradients, and its predictive capabilities generalize better over the range of investigated conditions. The analysis is completed with a deep ensemble approach for quantifying the uncertainties in the model prediction and integrated gradient analysis of the model sensitivity to its inputs. Uncertainties and sensitivities allow for identifying the regions where new training data would be most beneficial to the model's accuracy, thus opening the path toward a self-calibrating modeling approach. [ABSTRACT FROM AUTHOR]