Layer Neural Network Research Articles

Anomalous hall and skyrmion topological hall resistivity in magnetic heterostructures for the neuromorphic computing applications

Topologically protected spin textures, such as magnetic skyrmions, have shown the potential for high-density data storage and energy-efficient computing applications owing to their particle-like behavior, small size, and low driving current requirements. Evaluating the writing and reading of the skyrmion’s magnetic and electrical characteristics is crucial to implementing these devices. In this paper, we present the magnetic heterostructure Hall bar device and study the anomalous Hall and topological Hall signals in these devices. Using different measurement techniques, we investigate the magnetic and electrical characteristics of the magnetic structure. We measure the skyrmion topological resistivity and the magnetic field at different temperatures. MFM imaging and micromagnetic simulations further explain the anomalous Hall and topological Hall resistivity characteristics at various magnetic fields and temperatures. The study is extended to propose a skyrmion-based synaptic device showing spin-orbit torque-controlled plasticity. The resistance states are read using the anomalous Hall measurement technique. The device integration in a neuromorphic circuit is simulated in a 3-layer feedforward artificial neural network ANN. Based on the proposed synapses, the neural network is trained and tested on the MNIST data set, where a recognition accuracy performance of about 90% is achieved. Considering the nanosecond reading/writing time scale and a good system level performance, these devices exhibit a substantial prospect for energy-efficient neuromorphic computing.

Performance Prediction and Optimization of Single-Piston Free Piston Expander-Linear Generator Based on Machine Learning and Genetic Algorithm

This paper proposed a single-piston free piston expander-linear generator (SFPE-LG) prototype applied to organic Rankine cycle systems. Two valve timing control strategies, namely, time control strategy (TCS) and position control strategy (PCS), were developed. Based on the experimental data, a back propagation neural network (BPNN) prediction model was established. The effects of structural parameters such as neural network layers, transfer function, training function, hidden layer nodes, and learning rate on the prediction accuracy of this BPNN model were discussed. The training and prediction accuracy of the BPNN model was verified using 5-fold cross-validation and Wilcoxon signed-rank test. Moreover, the BPNN model was integrated with a genetic algorithm to predict and optimize the maximum output power of the SFPE-LG. The results showed that the BPNN model used to predict the motion characteristics and output performance of the SFPE-LG exhibits strong learning ability and high prediction accuracy. Notably, the prediction accuracy of the BPNN model is significantly higher under the PCS compared to TCS. The effect of hidden layer nodes on mean square error (MSE) and correlation coefficient (R) is greater than that of the learning rate. When the number of hidden layer nodes exceeds 30, the BPNN model consistently achieves low MSE and high R. The optimization results showed that the SFPE-LG can obtain a maximum output power of 141.69 W under the TCS, when the working parameters are inlet pressure of 0.7 MPa, intake duration of 35 ms, load resistance of 67 Ω, and expansion duration of 104 ms, respectively.

The scope of artificial intelligence in retinopathy of prematurity (ROP) management.

Artificial Intelligence (AI) is a revolutionary technology that has the potential to develop into a widely implemented system that could reduce the dependence on qualified professionals/experts for screening the large at-risk population, especially in the Indian scenario. Deep learning involves learning without being explicitly told what to focus on and utilizes several layers of artificial neural networks (ANNs) to create a robust algorithm that is capable of high-complexity tasks. Convolutional neural networks (CNNs) are a subset of ANNs that are particularly useful for image processing as well as cognitive tasks. Training of these algorithms involves inputting raw human-labeled data, which are then processed through the algorithm's multiple layers and allow CNN to develop their own learning of image features. AI systems must be validated using different population datasets since the performance of the AI system would vary according to the population. Indian datasets have been used in AI-based risk model that could predict whether an infant would develop treatment-requiring retinopathy of prematurity (ROP). AI also served as an epidemiological tool by objectively showing that a higher ROP severity was in Neonatal intensive care units (NICUs) that did not have the resources to monitor and titrate oxygen. There are rising concerns about the medicolegal aspect of AI implementation as well as discussion on the possibilities of catastrophic life-threatening diseases like retinoblastoma and lipemia retinalis being missed by AI. Computer-based systems have the advantage over humans in not being susceptible to biases or fatigue. This is especially relevant in a country like India with an increased rate of ROP and a preexisting strained doctor-to-preterm child ratio. Many AI algorithms can perform in a way comparable to or exceeding human experts, and this opens possibilities for future large-scale prospective studies.

State of charge estimation for Li-ion battery during dynamic driving process based on dual-channel deep learning methods and conditional judgement

The accurate estimation of the state of charge (SOC) is crucial for the safe and reliable operation of Li-ion batteries. In this paper, novel dual-channel deep learning methods were proposed for SOC estimation during dynamic driving cycles. For dual-channel deep learning methods, each channel of the deep learning methods could be a single convolutional neural network (CNN) layer, single long short-term memory (LSTM) layer or sequentially combined structure of CNN and LSTM. After evaluation, it could be found that dual-channel deep learning methods could achieve higher accuracy than conventional single-channel method, however, the model size and floating-point operations (FLOPs) could be either reduced or increased. Among all the dual-channel deep learning methods, CNN + LSTM-CNN + LSTM method could achieve the highest accuracy and the shortest training time with sacrifices in computational complexity. CNN–CNN + LSTM method could achieve higher accuracy than the single-channel deep learning method with reduced computational complexity and model size. As the error of SOC estimation is highly correlated with voltage signals, conditional judgement was then integrated with dual-channel deep learning method to obtain higher accuracy. After integrated with conditional judgement, the averaged error of SOC estimation could be reduced by more than 45% without a substantial sacrifice in computational time.

Hierarchical bottleneck for heterogeneous graph representation

Heterogeneous graphs (HGs) contain many nodes and their interaction relationships, which can model complex systems and provide rich semantic and structural information for task execution. Among these, HG representation stands as the fundamental and pivotal component. Existing HG representation methods primarily employ graph neural networks to acquire the semantics of nodes along various meta-paths and fuse them to represent the nodes. The most prevalent HG representation methods encompass two steps: semantic information extraction within meta-paths and semantic fusion between meta-paths. However, these methods overlooked the consideration of node heterogeneity within meta-paths and the simultaneous semantic correlation between meta-paths. Specifically, node heterogeneity within meta-paths signifies that the meta-path-based neighbors do not consistently contain information that positively influences the target node, and the semantic correlation between meta-paths indicates that different meta-path spaces are not entirely independent. Disregarding either of these issues leads to the propagation of irrelevant or redundant information and potential disruption of HG embedding. Consequently, in this study, we propose the HBHG, which is a hierarchical bottleneck for heterogeneous graph representation. HBHG primarily employs the information bottleneck (IB) as a guiding principle, constraining the propagation of irrelevant information within and between meta-paths while preserving relevant information. The central concept of the IB revolves around viewing model learning as the preservation of relevant information and compression of irrelevant information, accomplished by minimizing the dependency between input and hidden features through mutual information (MI) and maximizing the dependency between hidden features and ground-truth. Considering the complexity associated with MI estimation, this paper introduces a novel dependency index, namely the Hilbert-Schmidt independence criterion (HSIC), which offers ease of calculation. Specifically, HBHG comprises two primary components: a semantic bottleneck within meta-paths and a semantic bottleneck between meta-paths. The semantic bottleneck within meta-paths relies primarily on the HSIC-based limitations of dependencies at different layers of the graph neural network on various meta-paths, thereby maximizing the extraction of information relevant to the target node from neighboring nodes. The semantic bottleneck between meta-paths enables flexible extraction and fusion of semantic information based on downstream tasks, achieved by managing the trade-off of dependencies with HSIC between different meta-path semantic spaces. In summary, the proposed HBHG integrates hierarchical bottleneck constraints within and between meta-paths. This integration serves to maximize the aggregation of relevant information while effectively compressing irrelevant information, thereby enhancing the quality of heterogeneous graph embedding. The effectiveness of HBHG was validated through performance and ablation experiments conducted on multiple datasets.

The significance of radiative heat and mass transfer through a vertical sheet with chemical reaction: Designing by artificial approach Levenberg-Marquardt

The present study examines the impact of incorporating soft computing algorithms during neural network training on the evaluation and prediction performance of artificial neural networks. The research centers on a magneto hydrodynamic flow model (RHMT-VSCR) that depicts the movement of rotating fluid along a vertical sheet in a permeable medium. Furthermore, the model considers the impact of heat source-sink interactions, thermal radiation, and reactive species in conjunction with the effects of Hall current on the enhancement of energy and solute profiles. Because the induced magnetic field has a negligible magnetic Reynolds number, it is disregarded. A set of governing nonlinear PDEs is converted to a system of ODEs by applying an appropriate postulate of similarity variables in order to analyze the system. The multilayer perceptron network utilizes a supervised Levenberg–Marquardt Backpropagation algorithm (LMLA-BPNN) to ascertain the ideal quantity of neurons to be included in the hidden layer of artificial neural networks intended for modeling purposes. The bvp4c numerical approach is utilized to establish a continuous neural network mapping, which produces datasets that are utilized for the purposes of authentication, training, and testing. The range i.e., 10−2−10−8 of absolute error of the reference and target data demonstrates the optimal accuracy performance of LMLA-BPNN networks. After acquiring knowledge of neural network mapping, it is applied to approximate solutions for a wide range of scenarios through the manipulation of physical constraints, including suction/injection quantities, magnetic parameter, and porosity parameter, which influence the characteristics of flow, energy, and concentration.To assess the precision of the proposed method, statistical graphs based on regression, mean squared error analysis graphs, and error histogram graphs are employed. The results of the research demonstrate that the Levenberg-Marquardt Backpropagation neural network mappings' derivation, convergence, authentication, and stability were effectively validated.

High-resolution short-term prediction of the COVID-19 epidemic based on spatial-temporal model modified by historical meteorological data

In the global challenge of Coronavirus disease 2019 (COVID-19) pandemic, accurate prediction of daily new cases is crucial for epidemic prevention and socioeconomic planning. In contrast to traditional local, one-dimensional time-series data-based infection models, the study introduces an innovative approach by formulating the short-term prediction problem of new cases in a region as multidimensional, gridded time series for both input and prediction targets. A spatial-temporal depth prediction model for COVID-19 (ConvLSTM) is presented, and further ConvLSTM by integrating historical meteorological factors (Meteor-ConvLSTM) is refined, considering the influence of meteorological factors on the propagation of COVID-19. The correlation between 10 meteorological factors and the dynamic progression of COVID-19 was evaluated, employing spatial analysis techniques (spatial autocorrelation analysis, trend surface analysis, etc.) to describe the spatial and temporal characteristics of the epidemic. Leveraging the original ConvLSTM, an artificial neural network layer is introduced to learn how meteorological factors impact the infection spread, providing a 5-day forecast at a 0.01° × 0.01° pixel resolution. Simulation results using real dataset from the 3.15 outbreak in Shanghai demonstrate the efficacy of Meteor-ConvLSTM, with reduced RMSE of 0.110 and increased R2 of 0.125 (original ConvLSTM: RMSE = 0.702, R2 = 0.567; Meteor-ConvLSTM: RMSE = 0.592, R2 = 0.692), showcasing its utility for investigating the epidemiological characteristics, transmission dynamics, and epidemic development.

Optical fiber vibration signal recognition based on an efficient multidimensional feature extraction network.

In the field of optical fiber vibration signal recognition, one-dimensional signals have few features. People often used the shallow layer of a one-dimensional convolutional neural network (1D-CNN), which results in fewer features being learned by the network, leading to a poor recognition rate. There are also many complex algorithms and data processing methods, which make the whole signal recognition process more complicated. Therefore, an optical vibration signal recognition method based on an efficient multidimensional feature extraction network was proposed. Based on ResNet-50, efficient channel attention (ECA) was used to improve image features extraction ability, and a long short-term memory (LSTM) network was used to enhance the extraction of temporal features. Three different vibration signals were collected using a phase-sensitive optical time-domain reflectometry (Φ-OTDR) optical fiber sensing system. Vibration signals were converted into 128×128 grayscale images, which have more effective vibration information. The experimental results show that the three types of signals can be recognized and classified effectively by the network, and the average recognition rate is 98.67%.

Architecture of a three-dimensional convolutional neural network for detecting the fact of falsification of a video sequence

The article reflects the use of neural network technologies to determine the facts of falsification of the contents of video sequences. In the modern world, new technologies have become an integral part of the multimedia environment, but their proliferation has also created a new threat - the possibility of misuse to falsify the contents of video sequences. This leads to serious problems, such as the spread of fake news and misinformation of society. The scientific article examines this problem and determines the need to use neural networks to solve it. In comparison with other existing models and approaches, neural networks have high efficiency and accuracy in detecting video data falsification due to their ability to extract complex features and learn from large amounts of source data, which is especially important when reducing the resolution of the analyzed video sequence. Within the framework of this work, a mathematical model for identifying the falsification of audio and video sequences in video recordings is presented, as well as a model based on a three-dimensional convolutional neural network to determine the fact of falsification of a video sequence by analyzing the contents of individual frames. Within the framework of this work, it was proposed to consider the problem of identifying falsifications in video recordings as a joint solution to two problems: identification of falsification of audio and video sequences, and the resulting problem itself was transformed into a classical classification problem. Any video recording can be assigned to one of the four groups described in the work. Only the videos belonging to the first group are considered authentic, and all the others are fabricated. To increase the flexibility of the model, probabilistic classifiers have been added, which allows to take into account the degree of confidence in the predictions. The peculiarity of the resulting solution is the ability to adjust the threshold values, which allows to adapt the model to different levels of rigor depending on the task. The architecture of a three-dimensional convolutional neural network, including a preprocessing layer and a neural network layer, is proposed to determine fabricated photoreceads. The resulting model has a sufficient degree of accuracy in determining falsified video sequences, taking into account a significant decrease in frame resolution. Testing of the model on a training dataset showed the proportion of correct detection of video sequence falsification above 70%, which is noticeably better than guessing. Despite the sufficient accuracy, the model can be refined to more significantly increase the proportion of correct predictions.

What does artificial intelligence mean in rheumatology?

Intelligence is the ability of humans to learn from experiences to ascribe conscious weights and unconscious biases to modulate their outputs from given inputs. Transferring this ability to computers is artificial intelligence (AI). The ability of computers to understand data in an intelligent manner is machine learning. When such learning is with images and videos, which involves deeper layers of artificial neural networks, it is described as deep learning. Large language models are the latest development in AI which incorporate self-learning into deep learning through transformers. AI in Rheumatology has immense potential to revolutionize healthcare and research. Machine learning could aid clinical diagnosis and decision-making, and deep learning could extend this to analyze images of radiology or positron emission tomography scans or histopathology images to aid a clinician's diagnosis. Analysis of routinely obtained patient data or continuously collected information from wearables could predict disease flares. Analysis of high-volume genomics, transcriptomics, proteomics, or metabolomics data from patients could help identify novel markers of disease prognosis. AI might identify newer therapeutic targets based on in-silico modelling of omics data. AI could help automate medical administrative work such as inputting information into electronic health records or transcribing clinic notes. AI could help automate patient education and counselling. Beyond the clinic, AI has the potential to aid medical education. The ever-expanding capabilities of AI models bring along with them considerable ethical challenges, particularly related to risks of misuse. Nevertheless, the widespread use of AI in Rheumatology is inevitable and a progress with great potential.

Physics-informed neural networks for transonic flow around a cylinder with high Reynolds number

The physics-informed neural network (PINN) method is extended to learn and predict compressible steady-state aerodynamic flows with a high Reynolds number. To better learn the thin boundary layer, the sampling distance function and hard boundary condition are explicitly introduced into the input and output layers of the deep neural network, respectively. A gradient weight factor is considered in the loss function to implement the PINN methods based on the Reynolds averaged Navier–Stokes (RANS) and Euler equations, respectively, denoted as PINN–RANS and PINN–Euler. Taking a transonic flow around a cylinder as an example, these PINN methods are first verified for the ability to learn complex flows and then are applied to predict the global flow based on a part of physical data. When predicting the global flow based on velocity data in local key regions, the PINN–RANS method can always accurately predict the global flow field including the boundary layer and wake, while the PINN–Euler method can accurately predict the inviscid region. When predicting the subsonic and transonic flows under different freestream Mach numbers (Ma∞= 0.3–0.7), the flow fields predicted by both methods avoid the inconsistency with the real physical phenomena of the pure data-driven method. The PINN–RANS method is insufficient in shock identification capabilities. Since the PINN–Euler method does not need the second derivative, the training time of PINN–Euler is only 1/3 times that of PINN–RANS at the same sampling point and deep neural network.

Neural Network Methods Based on Efficient Optimization Algorithms for Solving Impulsive Differential Equations

In view of the fact that impulsive differential equations have the discreteness due to the impulse phenomenon, this paper proposes a single hidden layer neural network method based extreme learning machine and a physicsinformed neural network method combined with learning rate attenuation strategy to solve linear impulsive differential equations and nonlinear impulsive differential equations respectively. For the linear impulsive differential equations, firstly, the interval is segmented according to the impulse points, and a single hidden layer neural network model is constructed, the weight parameters of hidden layer are randomly set, the optimal output parameters and solution of the first segment are obtained by the extreme learning machine algorithm, then we calculate the initial value of the second segment according to the jumping equation, and the remaining segments are solved in turn in the same way. Although the single hidden layer neural network method proposed can solve linear equations with high accuracy, it is not suitable for solving nonlinear equations. Therefore, we propose the physics-informed neural network combined with learning rate attenuation strategy to solve the nonlinear impulsive differential equations, then the Adam algorithm and L-BFGS algorithm are combined to find the optimal approximate solution of each segment. Numerical examples show that the single hidden layer neural network method with Legendre polynomials as the activation function and the physics-informed neural network method combined with learning rate attenuation strategy can solve linear and nonlinear impulsive differential equations with higher accuracy.

Association between short birth spacing and child malnutrition in Bangladesh: a propensity score matching approach

ObjectivesThis study aimed to explore the effects of short birth spacing (SBS), which is defined as a period of less than 33 months between two successive births, on multiple concurrent...

Image classification system for COVID-19 patient X-Ray results using median filter with deep learning method

An image is a representation of an object that is rewritten on a medium with specific values (intensity) and has x and y coordinates. X-ray images are one type of medical image that can be used to detect and study a disease. However, X-ray images can sometimes appear blurry, making interpretation a bit challenging. Furthermore, there is variation in X-ray attenuation between normal tissue and tissue affected by a disease. By using the median filter and implementing deep learning as a method for classifying images based on feature extraction and weights in artificial neural networks. The stages carried out include preprocessing, with training and testing, feature extraction using layers of artificial neural networks that have undergone the filtering process. Then, a custom classifier layer is created to train the classification of COVID and normal classes using data from bottleneck.npy. The results of training and testing yielded an average accuracy of 95%, while the F1 Score evaluation result was 0.9."

Deep L1 Stochastic Optimal Control Policies for Planetary Soft Landing

In this paper, a novel deep-learning-based solution is introduced to the Powered Descent Guidance problem, grounded in principles of nonlinear Stochastic Optimal Control (SOC) and Feynman–Kac theory. Our algorithm solves the problem by framing it as an L1-SOC problem for minimum fuel consumption. Additionally, it can handle practically useful control constraints and nonlinear dynamics and enforces state constraints as soft constraints. This is achieved by building off of recent work on deep forward-backward stochastic differential equations and differentiable neural-network layers for nonconvex optimization based on stochastic search. In contrast to previous approaches, our algorithm does not require convexification of the constraints or linearization of the dynamics and is empirically shown to be robust to stochastic disturbances and the initial conditions of the spacecraft. After training offline, our control policy can be activated once the spacecraft is within a prespecified radius of the landing zone and at a prespecified altitude, in other words, the base of an inverted cone with the tip at the landing zone. We demonstrate empirically that our controller can successfully and safely land all trajectories initialized in the vicinity of the base of this cone as well as with randomization in the starting velocities and spacecraft mass while minimizing fuel consumption.

Implementing perceptrons by means of water-based computing

Water-based computing emerged as a branch of membrane computing in which water tanks act as permeable membranes connected via pipes. Valves residing at the pipes control the flow of water in terms of processing rules. Resulting water tank systems provide a promising platform for exploration and for case studies of information processing by flow of liquid media like water. We first discuss the possibility of realizing a single layer neural network using tanks and pipes systems. Moreover, we discuss the possibility to create a multi-layer neural network, which could be used to solve more complex problems. Two different implementations are considered: in a first solution, the weight values of the connections between the network nodes are represented by tanks. This means that the network diagram includes multiplication structures between the weight tanks and the input tanks. The second solution aims at simplifying the network proposed in the previous implementation, by considering the possibility to modify the weight values associated to neuron by varying the diameter of the connecting pipes between the tanks. The multiplication structures are replaced with a timer that regulates the opening of the outlet valves of all the tanks. These two implementations can be compared to evaluate their efficiency, and considerations will be made regarding the simplicity of implementation.

Exploring Brain Dynamics via EEG and Steady-State Activation Map Networks in Music Composition.

In recent years, the integration of brain-computer interface technology and neural networks in the field of music generation has garnered widespread attention. These studies aimed to extract individual-specific emotional and state information from electroencephalogram (EEG) signals to generate unique musical compositions. While existing research has focused primarily on brain regions associated with emotions, this study extends this research to brain regions related to musical composition. To this end, a novel neural network model incorporating attention mechanisms and steady-state activation mapping (SSAM) was proposed. In this model, the self-attention module enhances task-related information in the current state matrix, while the extended attention module captures the importance of state matrices over different time frames. Additionally, a convolutional neural network layer is used to capture spatial information. Finally, the ECA module integrates the frequency information learned by the model in each of the four frequency bands, mapping these by learning their complementary frequency information into the final attention representation. Evaluations conducted on a dataset specifically constructed for this study revealed that the model surpassed representative models in the emotion recognition field, with recognition rate improvements of 1.47% and 3.83% for two different music states. Analysis of the attention matrix indicates that the left frontal lobe and occipital lobe are the most critical brain regions in distinguishing between 'recall and creation' states, while FP1, FPZ, O1, OZ, and O2 are the electrodes most related to this state. In our study of the correlations and significances between these areas and other electrodes, we found that individuals with musical training exhibit more extensive functional connectivity across multiple brain regions. This discovery not only deepens our understanding of how musical training can enhance the brain's ability to work in coordination but also provides crucial guidance for the advancement of brain-computer music generation technologies, particularly in the selection of key brain areas and electrode configurations. We hope our research can guide the work of EEG-based music generation to create better and more personalized music.

Transfer learning model for estimating site amplification factors from limited microtremor H/V spectral ratios

Summary Site amplification factors (SAFs) of seismic ground motions are essential in evaluating and estimating seismic hazards. In our previous study, the authors proposed a simple and cost-effective method to estimate a SAF based on a deep neural network (DNN) model and microtremor horizontal-to-vertical spectral ratio (MHVR). Since the previous DNN model was based on the observed SAFs and MHVRs within a limited district in Japan, the applicability of the previous model to non-source regions with different site conditions was limited. This study explored the application of a transfer learning (TL) technique to adapt an existing (pre-trained) DNN model to new regions and a different database. The SAFs obtained through generalized spectral inversion technique (GIT) at the seismic observation stations (K-NET and KiK-net) in Japan were collated as the ground truth for site effects. MHVRs recorded at the stations in several districts of Japan were collected to construct a dataset for the development of the TL model. Subsequently, a TL model was constructed, leveraging the neural network layers and their weights from the pre-trained model while incorporating additional neural network layers to enhance the performance. During the training process, a total data set of 112 sites was divided into training set, validation set, and external test set by 1:1:5. Utilizing a cross-validation strategy, the residuals between pSAFs (pseudo-SAFs) estimated by the TL model and the observed SAFs were analyzed for the external test set containing 80 sites. The results showed that the TL model outperformed the pre-trained DNN model. The cross-validation results demonstrated that almost consistent prediction results were obtained regardless of any combination of 16 sites selected as the training set. Furthermore, by contrasting the influence of varying training set sizes on the performance of the TL model and comparing the TL model to a DNN model with an extended training set, the effectiveness of constructing the model with the limited number of data (16 sites) was ascertained. Finally, the effectiveness and limitations of the TL model were evaluated using MHVRs with peak frequencies falling outside the training set's range.

PMONN: an optical neural network for photonic integrated circuits based on micro-resonator.

We propose an improved optical neural network (ONN) circuit architecture based on conventional micro-resonator ONNs, called the Phase-based Micro-resonator Optical Neural Network (PMONN). PMONN's core architecture features a Convolutions and Batch Normalization (CB) unit, comprising a phase-based (PB) convolutional layer, a Depth-Point-Wise (DPW) convolutional layer, and a reconstructed Batch Normalization (RBN) layer. The PB convolution kernel uses modulable phase shifts of Add-drop MRRs as learnable parameters and their optical transfer function as convolution weights. The DPW convolution kernel amplifies PB convolution weights by learning the amplification factors. To address the internal covariate shift during training, the RBN layer normalizes DPW outputs by reconstructing the BN layer of the electronic neural network, which is then merged with the DPW layer in the test stage. We employ the tunable DAs in the architecture to implement the merged layer. PMONN achieves 99.15% and 91.83% accuracy on MNIST and Fashion-MNIST datasets, respectively. This work presents a method for implementing an optical neural network on the improved architecture based on MRRs and increases the flexibility and reusability of the architecture. PMONN has potential applications as the backbone for future optical object detection neural networks.

Prediction Enhancement of Metasurface Absorber Design Using Adaptive Cascaded Deep Learning (ACDL) Model

This paper presents a customized adaptive cascaded deep learning (ACDL) model for the design and performance prediction of metasurface absorbers. A multi-resonant metasurface absorber structure is introduced, with 10 target-driven design parameters. The proposed deep learning model takes advantage of cascading several sub-deep neural network (DNN) layers with forward noise mitigation capabilities. The inherent appearance of sparse data is dealt with in this work by proposing a trained data-adaptive selection technique. On the basis of the findings, the prediction response is quite fast and accurate enough to retrieve the design parameters of the studied metasurface absorber with two patches of 4000- and 7000-sample datasets. The training loss taken from the second DNN of our proposed model showed logarithmic mean squared errors of 0.039 and 0.033 when using Keras and the adaptive method, respectively, with a dataset split of 4000. On the contrary, for a dataset split of 7000, the errors were 0.049 with Keras and 0.045 with the adaptive method. On the other hand, the validation loss was evaluated using the mean square error method, which resulted in a loss of 0.044 with the 4000-sample datasets split with the Keras method, while this was 0.020 with the adaptive method. When extending the dataset to 7000 samples, the validation loss with the Keras splitting method was 0.0073, while it was improved, reaching 0.006, with the proposed adaptive method, and achieved a prediction accuracy of 94%. This proposed deep learning model can be deployed in the design process and synthesis of multi-resonant metasurface absorber structures. The proposed model shows the advantages of making the design process more efficient in sparse dataset handling, being an efficient approach in multi-resonance metasurface data pre-processing, being less time consuming, and being computationally valuable.