Traditional Artificial Neural Network Research Articles

Prediction of anticancer drug sensitivity using an interpretable model guided by deep learning

BackgroundThe prediction of drug sensitivity plays a crucial role in improving the therapeutic effect of drugs. However, testing the effectiveness of drugs is challenging due to the complex mechanism of drug reactions and the lack of interpretability in most machine learning and deep learning methods. Therefore, it is imperative to establish an interpretable model that receives various cell line and drug feature data to learn drug response mechanisms and achieve stable predictions between available datasets.ResultsThis study proposes a new and interpretable deep learning model, DrugGene, which integrates gene expression, gene mutation, gene copy number variation of cancer cells, and chemical characteristics of anticancer drugs to predict their sensitivity. This model comprises two different branches of neural networks, where the first involves a hierarchical structure of biological subsystems that uses the biological processes of human cells to form a visual neural network (VNN) and an interpretable deep neural network for human cancer cells. DrugGene receives genotype input from the cell line and detects changes in the subsystem states. We also employ a traditional artificial neural network (ANN) to capture the chemical structural features of drugs. DrugGene generates final drug response predictions by combining VNN and ANN and integrating their outputs into a fully connected layer. The experimental results using drug sensitivity data extracted from the Cancer Drug Sensitivity Genome Database and the Cancer Treatment Response Portal v2 reveal that the proposed model is better than existing prediction methods. Therefore, our model achieves higher accuracy, learns the reaction mechanisms between anticancer drugs and cell lines from various features, and interprets the model’s predicted results.ConclusionsOur method utilizes biological pathways to construct neural networks, which can use genotypes to monitor changes in the state of network subsystems, thereby interpreting the prediction results in the model and achieving satisfactory prediction accuracy. This will help explore new directions in cancer treatment. More available code resources can be downloaded for free from GitHub (https://github.com/pangweixiong/DrugGene).

Competition of systematically optimized deep neural networks for the estimation of tensile behavior of aluminum-magnesium alloy

Aluminum-magnesium (Al-Mg) alloys are prevalently employed within the aerospace sector. This research engaged a suite of deep learning approaches, encompassing the Artificial Neural Network (ANN), Gated Recurrent Unit (GRU) networks, Long-Short Term Memory (LSTM), and simple Recurrent Neural Network (RNN) to evaluate their predictive efficacy regarding the tensile strength and stiffness of Al-Mg alloys obtained from molecular dynamics simulation. The Taguchi method was initially applied to refine the architecture of each deep neural network (DNN), followed by a comparative analysis of their optimized configurations. The findings of this investigation revealed that the refined simple RNN and LSTM models exhibited superior predictive accuracy for estimating the strength and stiffness of the alloy, respectively. Moreover, the study elucidated that DNNs equipped with memory capabilities outstripped traditional ANNs in forecasting the tensile properties of Al-Mg alloys.

Artificial intelligence based succored power prediction in stand alone PV system

ABSTRACT Accurate forecasting of solar power generation is crucial for effective management of renewable energy resources. This literature review emphasises the effectiveness of Artificial Neural Networks (ANNs) in predicting solar energy potential. While traditional ANNs with single hidden layers and predefined neuron configurations have shown success in various applications, they are not suitable for predicting nonlinear time-series databases in the context of this research. The limitations of traditional ANNs in predicting solar power potential necessitate the exploration of alternative approaches. Configuring the structure of traditional ANNs for solar power prediction is a time-consuming process that often requires manual or trial-and-error methods. To address this issue, optimisation techniques are incorporated into the ANN model. Various traditional methods have been explored in this research, but none has achieved satisfactory performance. Drawing from experiences with traditional techniques, a novel Evolution of Cub to Predator (ECP) technique is proposed. The investigation results demonstrate that the incorporation of optimisation techniques yields superior performance compared to traditional ANNs. The proposed ECP technique achieves an impressive prediction accuracy of 97.2%, surpassing existing optimisation techniques. These findings highlight the potential of the ECP technique for accurate solar power prediction and its contribution to optimising renewable energy resource management.

Convolutional neural network-based wind pressure prediction on low-rise buildings

A new model of wind pressure prediction for low-rise buildings is built based on deep convolutional neural network (CNN), and the model is trained and tested by the aerodynamic data from international open database. The prediction accuracy is compared with that of the literature, and then comprehensive error analysis on all 335 predicted taps under 3 untrained wind directions are carried out for mean and RMS pressure coefficients by various error metrics. The study shows that the distributions of mean and RMS coefficients predicted by the CNN model are closer to the experimental results in comparison with those predicted by the traditional artificial neural network (ANN) model and the common deep neural network (DNN) model without the convolution and pooling layers. The relative error on the corner tap is 13.5%, 4.3% and 0.1% by the ANN, DNN, and CNN models respectively. The mean-square-error (MSE) of the corner bay and the whole roof are also eminently reduced by the CNN model. By introducing the convolutional and pooling layers, more prediction errors are confined within 5% and basically no errors exceed 20%. Consequently, the CNN algorithm can be applied in solving various wind pressure prediction problems on buildings in the future work.

A Two-stage Method for Damage Detection in Z24 Bridge Based on K-nearest Neighbor and Artificial Neural Network

In this paper, we propose an effective approach for identifying damages in the Z24 bridge, a large-scale bridge in Switzerland. The dataset of the Z24 bridge is evaluated as a benchmark, reflecting the behavior of a real structure that has been used for numerous studies. However, most of the previous studies have only addressed the issues of updating the model or estimating the location and severity of damages. The core idea behind our proposed approach is to leverage the strengths of two effective Machine Learning (ML) algorithms: K-Nearest Neighbor (KNN) and Artificial Neural Network (ANN), to assess both the location and severity of damages in the Z24 bridge. First, we employ KNN, an unsupervised learning algorithm, for pinpointing the damage location. This strategy proves highly efficient, significantly reducing computation time by circumventing the need for a loss function during KNN training. By adopting this approach, KNN effectively mitigates the risk of encountering local minima in the ANN optimization process. Subsequently, we deploy ANN to determine the damage severity. When compared to previous studies on the Z24 bridge, our proposed method (KNN-ANN) exhibits promising results. Furthermore, our results illustrate that KNN-ANN consistently outperforms traditional ANN methodologies.

Spike-Weighted Spiking Neural Network with Spiking Long Short-Term Memory: A Biomimetic Approach to Decoding Brain Signals

Background. Brain–machine interfaces (BMIs) offer users the ability to directly communicate with digital devices through neural signals decoded with machine learning (ML)-based algorithms. Spiking Neural Networks (SNNs) are a type of Artificial Neural Network (ANN) that operate on neural spikes instead of continuous scalar outputs. Compared to traditional ANNs, SNNs perform fewer computations, use less memory, and mimic biological neurons better. However, SNNs only retain information for short durations, limiting their ability to capture long-term dependencies in time-variant data. Here, we propose a novel spike-weighted SNN with spiking long short-term memory (swSNN-SLSTM) for a regression problem. Spike-weighting captures neuronal firing rate instead of membrane potential, and the SLSTM layer captures long-term dependencies. Methods. We compared the performance of various ML algorithms during decoding directional movements, using a dataset of microelectrode recordings from a macaque during a directional joystick task, and also an open-source dataset. We thus quantified how swSNN-SLSTM performed compared to existing ML models: an unscented Kalman filter, LSTM-based ANN, and membrane-based SNN techniques. Result. The proposed swSNN-SLSTM outperforms both the unscented Kalman filter, the LSTM-based ANN, and the membrane based SNN technique. This shows that incorporating SLSTM can better capture long-term dependencies within neural data. Also, our proposed swSNN-SLSTM algorithm shows promise in reducing power consumption and lowering heat dissipation in implanted BMIs.

A Novel Ionospheric Inversion Model: PINN‐SAMI3 (Physics Informed Neural Network Based on SAMI3)

AbstractPurely data‐driven ionospheric modeling fails to adequately obey fundamental physical laws. To overcome this shortcoming, we propose a novel ionospheric inversion model, Physics‐Informed Neural Network based on fully physical models SAMI3 (PINN‐SAMI3). The model incorporates the governing equations of the ionospheric physical model SAMI3 into the neural network to reconstruct the temporal‐spatial distribution of ionospheric plasma parameters. The objective of this study is to investigate the feasibility of integrating physical models with machine learning for ionospheric modeling. The PINN‐SAMI3 framework enforces physical laws through the multiple ion species of continuity, momentum, temperature equations in the magnetic dipole coordinate system. The simulation results show that if sparse ion densities are used as training data, it is possible to retrieve ionospheric electron densities, ion velocities and ion temperatures, respectively. The optimal physical constraints have been also investigated for different inversion quantities. Furthermore, the impact of incorporating E × B velocity terms on inversion results during the periods of ionospheric calm and geomagnetic storm is analyzed. The PINN‐SAMI3 achieves good inversion results even using sparse data in comparison to the traditional artificial neural networks (ANN). The framework will contribute to advance the future space weather prediction capability with artificial intelligence (AI).

Damage Detection in Structural Health Monitoring Using an Integrated ANNIRSA Approach

In the past decade, artificial neural networks (ANNs) have been widely employed to address many problems. Despite their powerful problem-solving capabilities, ANNs are susceptible to a significant risk of stagnation in local minima due to using backpropagation algorithms based on gradient descent (GD) for optimal solution searching. In this paper, we introduce an enhanced version of the reptile search algorithm (IRSA), which operates in conjunction with an ANN to mitigate these limitations. By substituting GD with IRSA within an ANN, the network gains the ability to escape local minima, leading to improved prediction outcomes. To demonstrate the efficacy of IRSA in enhancing ANN’s performance, a numerical model of the Nam O Bridge is utilized. This model is updated to closely reflect actual structural conditions. Consequently, damage scenarios for single-element and multielement damage within the bridge structure are developed. The results confirm that ANNIRSA offers greater accuracy than traditional ANNs and ANNRSAs in predicting structural damage.

Efficient Spiking Neural Networks with Sparse Selective Activation for Continual Learning

The next generation of machine intelligence requires the capability of continual learning to acquire new knowledge without forgetting the old one while conserving limited computing resources. Spiking neural networks (SNNs), compared to artificial neural networks (ANNs), have more characteristics that align with biological neurons, which may be helpful as a potential gating function for knowledge maintenance in neural networks. Inspired by the selective sparse activation principle of context gating in biological systems, we present a novel SNN model with selective activation to achieve continual learning. The trace-based K-Winner-Take-All (K-WTA) and variable threshold components are designed to form the sparsity in selective activation in spatial and temporal dimensions of spiking neurons, which promotes the subpopulation of neuron activation to perform specific tasks. As a result, continual learning can be maintained by routing different tasks via different populations of neurons in the network. The experiments are conducted on MNIST and CIFAR10 datasets under the class incremental setting. The results show that the proposed SNN model achieves competitive performance similar to and even surpasses the other regularization-based methods deployed under traditional ANNs.

Gated Attention Coding for Training High-Performance and Efficient Spiking Neural Networks

Spiking neural networks (SNNs) are emerging as an energy-efficient alternative to traditional artificial neural networks (ANNs) due to their unique spike-based event-driven nature. Coding is crucial in SNNs as it converts external input stimuli into spatio-temporal feature sequences. However, most existing deep SNNs rely on direct coding that generates powerless spike representation and lacks the temporal dynamics inherent in human vision. Hence, we introduce Gated Attention Coding (GAC), a plug-and-play module that leverages the multi-dimensional gated attention unit to efficiently encode inputs into powerful representations before feeding them into the SNN architecture. GAC functions as a preprocessing layer that does not disrupt the spike-driven nature of the SNN, making it amenable to efficient neuromorphic hardware implementation with minimal modifications. Through an observer model theoretical analysis, we demonstrate GAC's attention mechanism improves temporal dynamics and coding efficiency. Experiments on CIFAR10/100 and ImageNet datasets demonstrate that GAC achieves state-of-the-art accuracy with remarkable efficiency. Notably, we improve top-1 accuracy by 3.10% on CIFAR100 with only 6-time steps and 1.07% on ImageNet while reducing energy usage to 66.9% of the previous works. To our best knowledge, it is the first time to explore the attention-based dynamic coding scheme in deep SNNs, with exceptional effectiveness and efficiency on large-scale datasets. Code is available at https://github.com/bollossom/GAC.

Efficient Spiking Neural Networks With Radix Encoding.

Spiking neural networks (SNNs) have advantages in latency and energy efficiency over traditional artificial neural networks (ANNs) due to their event-driven computation mechanism and the replacement of energy-consuming weight multiplication with addition. However, to achieve high accuracy, it usually requires long spike trains to ensure accuracy, usually more than 1000 time steps. This offsets the computation efficiency brought by SNNs because a longer spike train means a larger number of operations and larger latency. In this article, we propose a radix-encoded SNN, which has ultrashort spike trains. Specifically, it is able to use less than six time steps to achieve even higher accuracy than its traditional counterpart. We also develop a method to fit our radix encoding technique into the ANN-to-SNN conversion approach so that we can train radix-encoded SNNs more efficiently on mature platforms and hardware. Experiments show that our radix encoding can achieve 25× improvement in latency and 1.7% improvement in accuracy compared to the state-of-the-art method using the VGG-16 network on the CIFAR-10 dataset.

ANN-LSTM-A Water Consumption Prediction Based on Attention Mechanism Enhancement

To reduce the energy consumption of domestic hot water (DHW) production, it is necessary to reasonably select a water supply plan through early predictions of DHW consumption to optimize energy consumption. However, the fluctuations and intermittence of DHW consumption bring great challenges to the prediction of water consumption. In this paper, an ANN-LSTM-A water quantity prediction model based on attention mechanism (AM) enhancement is improved. The model includes an input layer, an AM layer, a hidden layer, and an output layer. Based on the combination of artificial neural network (ANN) and long short-term memory (LSTM) models, an AM is incorporated to address the issue of the traditional ANN model having difficulty capturing the long-term dependencies, such as lags and trends in time series, to improve the accuracy of the DHW consumption prediction. Through comparative experiments, it was found that the root mean square error of the ANN-LSTM-A model was 15.4%, 13.2%, and 13.2% lower than those of the ANN, LSTM, and ANN-LSTM models, respectively. The corresponding mean absolute error was 17.9%, 11.5%, and 8% lower than those of the ANN, LSTM, and ANN-LSTM models, respectively. The results showed that the proposed ANN-LSTM-A model yielded better performances in predicting DHW consumption than the ANN, LSTM, and ANN-LSTM models. This work provides an effective reference for the reasonable selection of the water supply plan and optimization of energy consumption.

Secure and privacy-preserving intrusion detection in wireless sensor networks: Federated learning with SCNN-Bi-LSTM for enhanced reliability

As the digital landscape expands rapidly due to technological advancements, cybersecurity concerns have become more prevalent. Intrusion Detection Systems (IDSs), which are crucial for identifying unusual network traffic indicative of malicious activity, have become a necessity. These systems can be either hardware or software-based. However, traditional IDS models often fail to adequately protect data privacy and detect complex, unique breaches, particularly within Wireless Sensor Networks (WSNs). To address these limitations, this paper proposes a novel Stacked Convolutional Neural Network and Bidirectional Long Short Term Memory (SCNN-Bi-LSTM) model for intrusion detection in WSNs. This model leverages Federated Learning (FL) to enhance intrusion detection performance and safeguard privacy. The FL-based SCNN-Bi-LSTM model is unique in its approach, allowing multiple sensor nodes to collaboratively train a central global model without revealing private data, thereby alleviating privacy concerns. The deep learning methodology of the SCNN-Bi-LSTM model effectively identifies sophisticated and previously unknown cyber threats by meticulously examining both local and temporal linkages in network patterns. The model has been specifically designed to detect and categorize different types of Denial of Service (DoS) attacks using specialized WSN-DS and CIC-IDS-2017 datasets. Compared to traditional Artificial Deep Neural Network (ADNN) models, our proposed FL-SCNN-Bi-LSTM model demonstrated superior detection rates for complex and unknown attacks, significantly improving IDS performance. The model achieved a notable classification accuracy of approximately 99.9% precision and recall on both datasets, substantially reducing false positives and negatives. Our research underscores the potential of federated learning and deep learning in enhancing the security and privacy of WSNs. The proposed FL-SCNN-Bi-LSTM architecture not only facilitates the identification of complex cyber threats but also exemplifies how deep learning techniques can be employed to bolster intrusion detection systems while preserving user data privacy.

Comparative Evaluation of Deep Learning Techniques in Streamflow Monthly Prediction of the Zarrine River Basin

Predicting monthly streamflow is essential for hydrological analysis and water resource management. Recent advancements in deep learning, particularly long short-term memory (LSTM) and recurrent neural networks (RNN), exhibit extraordinary efficacy in streamflow forecasting. This study employs RNN and LSTM to construct data-driven streamflow forecasting models. Sensitivity analysis, utilizing the analysis of variance (ANOVA) method, also is crucial for model refinement and identification of critical variables. This study covers monthly streamflow data from 1979 to 2014, employing five distinct model structures to ascertain the most optimal configuration. Application of the models to the Zarrine River basin in northwest Iran, a major sub-basin of Lake Urmia, demonstrates the superior accuracy of the RNN algorithm over LSTM. At the outlet of the basin, quantitative evaluations demonstrate that the RNN model outperforms the LSTM model across all model structures. The S3 model, characterized by its inclusion of all input variable values and a four-month delay, exhibits notably exceptional performance in this aspect. The accuracy measures applicable in this particular context were RMSE (22.8), R2 (0.84), and NSE (0.8). This study highlights the Zarrine River’s substantial impact on variations in Lake Urmia’s water level. Furthermore, the ANOVA method demonstrates exceptional performance in discerning the relevance of input factors. ANOVA underscores the key role of station streamflow, upstream station streamflow, and maximum temperature in influencing the model’s output. Notably, the RNN model, surpassing LSTM and traditional artificial neural network (ANN) models, excels in accurately mimicking rainfall–runoff processes. This emphasizes the potential of RNN networks to filter redundant information, distinguishing them as valuable tools in monthly streamflow forecasting.

Deep Ensemble Transformers for Dimensionality Reduction.

We propose deep ensemble transformers (DETs), a fast, scalable approach for dimensionality reduction problems. This method leverages the power of deep neural networks and employs cascade ensemble techniques as its fundamental feature extraction tool. To handle high-dimensional data, our approach employs a flexible number of intermediate layers sequentially. These layers progressively transform the input data into decision tree predictions. To further enhance prediction performance, the output from the final intermediate layer is fed through a feed-forward neural network architecture for final prediction. We derive an upper bound of the disparity between the generalization error and the empirical error and demonstrate that it converges to zero. This highlights the generalizability of our method to parameter estimation and feature selection problems. In our experimental evaluations, DETs outperform existing models in terms of prediction accuracy, representation learning ability, and computational time. Specifically, the method achieves over 95% accuracy in gene expression data and can be trained on average 50% faster than traditional artificial neural networks (ANNs).

Accuracy of a multipath ultrasonic flowmeter with deep learning based on the velocity profile

Purpose The asymmetry of the velocity profile caused by geometric deformation, complex turbulent motion and other factors must be considered to effectively use the flowmeter on any section. This study aims to better capture the flow field information and establish a model to predict the profile velocity, we take the classical double elbow as the research object and propose to divide the flow field into three categories with certain common characteristics. Design/methodology/approach The deep learning method is used to establish the model of multipath linear velocity fitting profile average velocity. A total of 480 groups of data are taken for training and validation, with ten integer velocity flow fields from 1 m/s to 10 m/s. Finally, accuracy research with relative error as standard is carried out. Findings The numerical experiment yielded the following promising results: the maximum relative error is approximately 1%, and in the majority of cases, the relative error is significantly lower than 1%. These results demonstrate that it surpasses the classical optimization algorithm Equal Tab (5%) and the traditional artificial neural network (3%) in the same scenario. In contrast with the previous research on a fixed profile, we focus on all the velocity profiles of a certain length for the first time, which can expand the application scope of a multipath ultrasonic flowmeter and promote the research on flow measurement in any section. Originality/value This work proposes to divide the flow field of double elbow into three categories with certain common characteristics to better capture the flow field information and establish a model to predict the profile velocity.

The use of memristive devices in machine vision systems

The comparison results of processing units with memristive devices versus modern hardware accelerators of artificial neural networks (ANNs) based on traditional electronic components, as presented in the review [1], demonstrate numerous advantages across all major indicators such as throughput, energy efficiency, accuracy, and others. This report analyzes the current state of memristive devices in addressing machine vision issues. Special attention is paid to the concept of [2] neuromorphic machine vision systems (MVS) based on memristive devices. This concept’s distinct feature lies in its fully analog system, commencing from information input to its output. It encompasses sensory and neural components. The sensory part is responsible for gathering visual information and transferring it to the neural segment for processing through the ANN model algorithm. A specific instance for implementing the input channel of the sensor component involves connecting a photodiode (PD) and a memristor in a single circuit. When the circuit is flipped in reverse bias and light falls on the PD, a photocurrent flows through it from the cathode to the anode. Depending on the light intensity and exposure time, this photocurrent alters the resistance of the memristor, thereby converting illumination into resistance. If visual information doesn’t require encoding with memristor resistances, they can be replaced with a load resistance of the same nominal value for all channels. Irrespective of the input channel variant, the signal encoding visual information is fed into the neural part without digitization. Memristors act as synapses as part of the neural part. They can be used to implement synapses in both traditional formal ANN architectures, where input information is multiplied by a pre-programmed weight, and synapses for spiking neural networks, where a memristor exhibits synaptic plasticity mechanisms similar to those in biological neural networks [3]. If the output from the suggested MVS input channel variants is connected to a device that operates on the “integrate and fire” principle, the device can be deemed as not only an input for a structured ANN, but also a presynaptic neuron for a spiking ANN. The neuron’s frequency of spikes will depend on the light intensity; the brighter the light, the higher the frequency of spikes and vice versa. Faster charge accumulation occurs in channels with low resistance. The complete analog machine vision system will function as a spike neural network, without incorporating any analog-to-digital or digital-to-analog converters. Compared to digital machine vision systems, this approach will significantly decrease energy consumption while producing wearable and on-board electronics with distinct tactical and technical features. This design can be tailored to the size of modern matrices of photo and video fixation devices and employed as a hardware accelerator for the ANN models currently used to process images, and it can serve as a foundation for advancing this area.

Predicting and Reconstructing Aerosol–Cloud–Precipitation Interactions with Physics-Informed Neural Networks

Interactions between clouds, aerosol, and precipitation are crucial aspects of weather and climate. The simple Koren–Feingold conceptual model is important for providing deeper insight into the complex aerosol–cloud–precipitation system. Recently, artificial neural networks (ANNs) and physics-informed neural networks (PINNs) have been used to study multiple dynamic systems. However, the Koren–Feingold model for aerosol–cloud–precipitation interactions has not yet been studied with either ANNs or PINNs. It is challenging for pure data-driven models, such as ANNs, to accurately predict and reconstruct time series in a small data regime. The pure data-driven approach results in the ANN becoming a “black box” that limits physical interpretability. We demonstrate how these challenges can be overcome by combining a simple ANN with physical laws into a PINN model (not purely data-driven, good for the small data regime, and interpretable). This paper is the first to use PINNs to learn about the original and modified Koren–Feingold models in a small data regime, including external forcings such as wildfire-induced aerosols or the diurnal cycle of clouds. By adding external forcing, we investigate the effects of environmental phenomena on the aerosol–cloud–precipitation system. In addition to predicting the system’s future, we also use PINN to reconstruct the system’s past: a nontrivial task because of time delay. So far, most research has focused on using PINNs to predict the future of dynamic systems. We demonstrate the PINN’s ability to reconstruct the past with limited data for a dynamic system with nonlinear delayed differential equations, such as the Koren–Feingold model, which remains underexplored in the literature. The main reason that this is possible is that the model is non-diffusive. We also demonstrate for the first time that PINNs have significant advantages over traditional ANNs in predicting the future and reconstructing the past of the original and modified Koren–Feingold models containing external forcings in the small data regime. We also show that the accuracy of the PINN is not sensitive to the value of the regularization factor (λ), a key parameter for the PINN that controls the weight for the physics loss relative to the data loss, for a broad range (from λ=1×103 to λ=1×105).

Optimized neural network-based state-of-the-art soft computing models for the bearing capacity of strip footings subjected to inclined loading

Determining the bearing capacity of a strip footing under inclined loading is crucial in designing foundations. Due to the complex correlations, the subject area remains predominantly unexplored, or it has been simulated using only limited datasets. This paper presents the development of a prediction model based on machine learning (ML), leveraging advanced hybrid artificial neural network (ANN) models for estimating the bearing capacity of strip footings under inclined loading. The ANN models are hybridized with four different optimization algorithms, ant colony optimization (ACO), artificial lion optimization (ALO), the imperialist competitive algorithm (ICA), and shuffled complex evolution (SCE), which enhance the accuracy and efficiency of the predictive capabilities of ANN. The models are trained on a dataset of 920 records, and their performance is evaluated using a range of significant performance metrics. The ANN-ICA model achieved the highest rank in the score analysis (R2 =0.912, RMSE=0.165 in testing), followed by ANN-ALO and ANN-ACO. To reinforce the trustworthiness of the predictions, external validation is employed, and visual analysis is conducted using the Taylor diagram. The findings suggest that the proposed models are robust, and the incorporation of optimization techniques has improved the performance of traditional ANNs. The research findings have significant implications for the field of geotechnical engineering, providing engineers and researchers with valuable insights into the applicability of hybrid artificial neural network (ANN) models and alternative machine learning (ML)-based prediction models in assessing the bearing capacity of strip footings.

Fast-SNN: Fast Spiking Neural Network by Converting Quantized ANN.

Spiking neural networks (SNNs) have shown advantages in computation and energy efficiency over traditional artificial neural networks (ANNs) thanks to their event-driven representations. SNNs also replace weight multiplications in ANNs with additions, which are more energy-efficient and less computationally intensive. However, it remains a challenge to train deep SNNs due to the discrete spiking function. A popular approach to circumvent this challenge is ANN-to-SNN conversion. However, due to the quantization error and accumulating error, it often requires lots of time steps (high inference latency) to achieve high performance, which negates SNN's advantages. To this end, this paper proposes Fast-SNN that achieves high performance with low latency. We demonstrate the equivalent mapping between temporal quantization in SNNs and spatial quantization in ANNs, based on which the minimization of the quantization error is transferred to quantized ANN training. With the minimization of the quantization error, we show that the sequential error is the primary cause of the accumulating error, which is addressed by introducing a signed IF neuron model and a layer-wise fine-tuning mechanism. Our method achieves state-of-the-art performance and low latency on various computer vision tasks, including image classification, object detection, and semantic segmentation. Codes are available at: https://github.com/yangfan-hu/Fast-SNN.