Search

Leveraging deep learning (DL) approaches in genomics data has led to significant advances in cancer prediction. The continuous availability of gene expression datasets over the preceding years has made them one of the most accessible sources of genome-wide data, advancing cancer bioinformatics research and advanced prediction of cancer genomic data. To contribute to this topic, the proposed work is based on DL prediction in both convolutional neural network (CNN) and recurrent neural network (RNN) for five classes in brain cancer using gene expression data obtained from Curated Microarray Database (CuMiDa). This database is used for cancer classification and is publicly accessible on the official CuMiDa website. This paper implemented DL approaches using a One Dimensional-Convolutional Neural Network (1D-CNN) followed by an RNN classifier with and without Bayesian hyperparameter optimization (BO). The accuracy of this hybrid model combination of (BO + 1D-CNN + RNN) produced the highest classification accuracy of 100% instead of the 95% for the ML model in prior work and 90% for the (1D-CNN + RNN) algorithm considered in the paper. Therefore, the classification of brain cancer gene expression according to the hybrid model (BO + 1D-CNN + RNN) provides more accurate and useful assessments for patients with different types of brain cancers. Thus, gene expression data are used to create a DL classification-based- hybrid model that will hold senior promise in the treatment of brain cancer. [ABSTRACT FROM AUTHOR]

We demonstrate the application of free-to-use and easy-to-implement Bayesian optimization (BO) software to streamline atomic layer deposition (ALD) process optimization. By employing machine learning-based Bayesian optimization algorithms, we enhanced the silicon surface passivation quality of titanium dioxide layers deposited using titanium tetraisopropoxide (TTIP). Unlike classical designs of experimental methods, such as Box–Behnken or Plackett–Burman designs, which require a predefined set of experiments and can become resource intensive, BO offers several advantages. It dynamically updates the search strategy based on previous outcomes, allowing for efficient exploration of parameter spaces with fewer experimental runs. This adaptive approach is particularly advantageous in small-scale experiments or laboratories where time, resources, and materials are limited. In a single-objective optimization experiment, we identified constrained search spaces that limited further optimization, underscoring the importance of properly defined parameter bounds prior to the optimization process. Our findings highlight that Bayesian optimization can not only reduce time and resource costs associated with ALD process optimization but also support faster discovery of more optimal ALD process parameters, even with minimal prior knowledge of the deposition process or precursor chemistry. [ABSTRACT FROM AUTHOR]

Abstract DC grid fault protection techniques have previously faced challenges such as fixed thresholds, insensitivity to high-resistance faults, and dependency on specific threshold settings. These limitations can lead to elevated fault currents in the grid, particularly affecting multi-modular converters (MMCs) vulnerability to large fault current transients. This paper proposes a novel approach that combines the disjoint-based Bootstrap Aggregating (Bagging) technique and Bayesian optimization (BO) for fault detection in DC grids. Disjoint partitions reduce variance and enhance Ensemble Artificial Neural Network (EANN) performance, while BO optimizes EANN architecture. The proposed approach uses multiple transient periods instead of a fixed time to train the model. Transient periods are segmented into multiple 1 ms intervals, and each interval trains a separate neural network. In this way, a robust local relay is created that does not require high-speed communication systems. Additionally, a discrete wavelet transform (DWT) is applied to select detailed coefficients of the transient fault current, measured at the DC line’s sending terminal for fault protection. EANN is trained in comprehensive offline data that considers noise impact. Simulation results demonstrate the scheme’s ability to detect faults as high as 400 Ω accurately. This makes it a robust, reliable, and effective solution for fault detection on high-voltage direct current (HVDC) transmission lines. Lastly, this research provides the first-ever scientometric analysis of HVDC transmission line fault protection using neural network algorithms, highlighting major research themes and trends. The scientometric analysis was based on a dataset of 136 available research articles from the Scopus database from the last ten years. Therefore, this research provides valuable insights into the use of ANN for HVDC transmission line fault protection.

Abstract Leveraging deep learning (DL) approaches in genomics data has led to significant advances in cancer prediction. The continuous availability of gene expression datasets over the preceding years has made them one of the most accessible sources of genome-wide data, advancing cancer bioinformatics research and advanced prediction of cancer genomic data. To contribute to this topic, the proposed work is based on DL prediction in both convolutional neural network (CNN) and recurrent neural network (RNN) for five classes in brain cancer using gene expression data obtained from Curated Microarray Database (CuMiDa). This database is used for cancer classification and is publicly accessible on the official CuMiDa website. This paper implemented DL approaches using a One Dimensional-Convolutional Neural Network (1D-CNN) followed by an RNN classifier with and without Bayesian hyperparameter optimization (BO). The accuracy of this hybrid model combination of (BO + 1D-CNN + RNN) produced the highest classification accuracy of 100% instead of the 95% for the ML model in prior work and 90% for the (1D-CNN + RNN) algorithm considered in the paper. Therefore, the classification of brain cancer gene expression according to the hybrid model (BO + 1D-CNN + RNN) provides more accurate and useful assessments for patients with different types of brain cancers. Thus, gene expression data are used to create a DL classification-based- hybrid model that will hold senior promise in the treatment of brain cancer.

Online monitoring of tool wear is an indispensable part of industrial production. It allows for real-time assessment of tool condition, enabling tool replacement at the optimal time. This not only reduces the time costs associated with downtime for inspections but also effectively prevents the deterioration of workpiece quality due to excessive wear. Therefore, online monitoring of tool wear is essential for ensuring machining quality and efficiency. This research presents a monitoring technique for tool wear while performing micro-milling by extracting features from machining signals. Initially, the wear mechanism, which includes initial, regular, and severe wear, is analyzed. Tool wear labels are then constructed. Subsequently, the recursive feature elimination (RFE) algorithm optimizes the original feature set from the processed signal, resulting in a subset of relevant features that accurately describe the tool wear. The optimal subset is utilized as an input vector for extreme gradient boosting (XGBoost) training. Next, the best hyperparameter combination for the monitoring model is determined using Bayesian optimization (BO), ensuring the generalization performance of the monitoring model. Finally, the two-edge wear monitoring model is established based on the RFE-BO-XGBoost. To validate the superiority of the RFE-BO-XGBoost model, the performance of the XGBoost model with default parameters in tool wear monitoring is also investigated. The results demonstrate that the constructed monitoring model can effectively identify the tool wear stage. More notably, it precisely predicts the wear status of two cutting edges simultaneously, enabling qualitative and quantitative monitoring of tool wear under normal manufacturing settings.

This paper introduces an Automated Machine Learning (AutoML) framework specifically designed to efficiently synthesize end-to-end multimodal machine learning pipelines. Traditional reliance on the computationally demanding Neural Architecture Search is minimized through the strategic integration of pre-trained transformer models. This innovative approach enables the effective unification of diverse data modalities into high-dimensional embeddings, streamlining the pipeline development process. We leverage an advanced Bayesian Optimization strategy, informed by meta-learning, to facilitate the warm-starting of the pipeline synthesis, thereby enhancing computational efficiency. Our methodology demonstrates its potential to create advanced and custom multimodal pipelines within limited computational resources. Extensive testing across 23 varied multimodal datasets indicates the promise and utility of our framework in diverse scenarios. The results contribute to the ongoing efforts in the AutoML field, suggesting new possibilities for efficiently handling complex multimodal data. This research represents a step towards developing more efficient and versatile tools in multimodal machine learning pipeline development, acknowledging the collaborative and ever-evolving nature of this field. [ABSTRACT FROM AUTHOR]

The utilization of smart wearable devices in healthcare environments is expected to increase as digital technology advances. Smart wearables and the data they provide have become prevalent in medical care applications. However, evaluating clinical applications is challenging due to the need to accumulate more information and the limitations of technical approaches. By applying measures of the established digital health platform that incorporates smart wearables for adolescent brain cognitive development (ABCD), we investigated the optimization of ordinary physical activity as determined by the individual participants. It aims to provide reliable leverage for physical metabolic classification. First, the adolescent physical activity datasets are evaluated statistically for this analysis. Then, multi-dimensional Bayesian optimization (BO) is applied to extract optimized MET (metabolic equivalent of task) values to categorize physical development behavior. The result shows that the practical MET level is overestimated, and the real value is lower than expected. [ABSTRACT FROM AUTHOR]

Stock forecasting is a complicated and daily challenge for investors because of the non-linearity of the market and the high volatility of financial assets such as stocks, bonds and other commodities. There is a need for a powerful and adaptive stock prediction model that handles complexities and provides accurate predictions. The support vector regression (SVR) model is one of the most prominent machine learning models for forecasting time series data. An ensemble hyperbolic tangent kernel SVR (HTK-SVR-BO) is proposed in this paper, combining Tanh and inverse Tanh kernels with Bayesian optimization. Combining the strengths of multiple kernels using the ensemble technique and then using optimization to identify the optimal values for each SVR model to enhance the ensemble model performance is possible. Our proposed model is compared with an ensemble SVR model (LPR-SVR-BO), which uses well-known SVR kernel types, including linear, polynomial and radial basis function (RBF). We apply the proposed models to Microsoft Corporation (MSFT) stock prices. The mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), R2 score (model accuracy) and mean absolute percentage error (MAPE) are the regression metrics used to compare the effectiveness of each ensemble model. In our comparison, HTK-SVR-BO performs better in terms of regression metrics compared to LPR-SVR-BO and achieves results of 0.27424, 0.13392, 0.36595, 0.99997 and 5.2331 respectively. According to the analysis, the proposed model is more predictive and may generalize to previously unknown data more effectively, so it can be accurate when forecasting future stock prices.

Stock price forecasting is a critical component of financial market analysis and plays a key role in formulating investment strategies, affecting a wide range of stakeholders. To address the inherent complexities and dynamic fluctuations in stock prices, innovative modeling techniques are required. Deep Learning (DL) approaches, particularly Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks, have shown notable promise in enhancing prediction accuracy for stock prices. This study introduces a novel hybrid Single Exponential Smoothing with Dual Attention based Bi-directional LSTM optimize with Bayesian Optimization (SES-DA-BiLSTM-BO) designed to predict future stock prices accurately using Life Insurance Corporation of India (LIC) stock price dataset with 1-minute, 3-minute, and 10-minute intervals. The model aims to forecast stock prices, identify market trends, and reveal influential factors more effectively. The performance of the proposed SES-DA-BiLSTM-BO model was evaluated using key metrics: Directional Accuracy (DA=0.4816), Mean Absolute Percentage Error (MAPE=0.0006), Win/Loss Ratio (WLR=0.7517), and Theil’s U-Statistic (TUS=0.0006). These results demonstrate its superior performance compared to traditional models such as Moving Average (MA), Auto-Regressive (AR), Auto-Regressive Moving Average (ARMA), Auto-Regressive Integrated Moving Average (ARIMA), Single Exponential Smoothing (SES) and LSTM models. By outperforming baseline model on various error metrics, the proposed SES-DA-BiLSTM-BO hybrid model exhibits improved predictive capabilities and lower loss functions, making its highly reliable for forecasting stock prices under diverse market conditions. The integration of statistical methods with deep learning in this hybrid approach offers a scalable and efficient solution for optimizing investment strategies, enhancing market efficiency, and contributing to economic stability by reducing the impact of market volatility.

Financial market prediction has shown considerable potential in the past few years from the combination of contemporary Deep Learning (DL) techniques and traditional time series forecasting methodologies. To predict the stock prices of three distinct companies General Electric (GE), Microsoft (MSFT), and Amazon (AMZN) datasets. This study presents a novel hybrid model that combines the Double Exponential Smoothing (DES) method with a Deep Learning (DL) model Dual Attention Encoder-Decoder based Bi-directional GRU, optimized using Bayesian Optimization (DES-DA-ED-Bi-GRU-BO). By combining the best features of contemporary and old methods, the hybrid model seeks to efficiently identify patterns and trends in stock data. When handling time series data, the DES method offers a reliable and flexible mechanism that considers trends and seasonality in the data. The DA-ED-Bi-GRU added to the deep learning model further improves its comprehension of intricate patterns found in the stock data. The parameters are adjusted using Bayesian optimization (BO) to maximize the model’s performance. Several performance indicators, such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R-Square ( $R^{2}$ ), and Theil’s U-Statistics (TUS), are used to assess the effectiveness of the model. These measures offer thorough insights into the precision, dependability, and accuracy of the model’s predictions. The experimental findings show that the proposed hybrid model has the ability to predict GE, MSFT, and AMZN stock values with reasonable accuracy. Along with the optimization framework, DL and conventional smoothing approaches combine to provide a potent forecasting tool that may help traders and investors make wise judgments.

Given their potential to demonstrate near-term quantum advantage, variational quantum algorithms (VQAs) have been extensively studied. Although numerous techniques have been developed for VQA parameter optimization, it remains a significant challenge. A practical issue is that quantum noise is highly unstable and thus it is likely to shift in real time. This presents a critical problem as an optimized VQA ansatz may not perform effectively under a different noise environment. For the first time, we explore how to optimize VQA parameters to be robust against unknown shifted noise. We model the noise level as a random variable with an unknown probability density function (PDF), and we assume that the PDF may shift within an uncertainty set. This assumption guides us to formulate a distributionally robust optimization problem, with the goal of finding parameters that maintain effectiveness under shifted noise. We utilize a distributionally robust Bayesian optimization solver for our proposed formulation. This provides numerical evidence in both the quantum approximate optimization algorithm and the variational quantum eigensolver with hardware-efficient ansatz, indicating that we can identify parameters that perform more robustly under shifted noise. We regard this work as the first step toward improving the reliability of VQAs influenced by shifted noise from the parameter optimization perspective.

In order to achieve rapid, non‐destructive, efficient, and accurate classification of paper cup samples, we propose a classification model that integrates shifted‐excitation Raman difference spectroscopy (SERDS) with self‐organizing map (SOM) and Bayesian optimization–support vector machine (BO‐SVM). We collected differential Raman data from 52 paper cup samples using SERDS, with an excitation wavelength range of 784–785 nm, a laser power of 440 mW, an integration time of 10 s, and a spectral range spanning from 280 to 2700 cm−1. Subsequently, principal component analysis (PCA) was applied to reduce the dimensionality of the data. The SOM clustering outcomes were utilized as the foundation for constructing the discriminant analysis (FDA) and BO‐SVM classification models. The primary constituent of the paper cup samples was identified as cellulose, while additional fillers such as talc, calcium carbonate, and kaolin were also present. The SOM clustering categorized the samples into seven distinct groups. The FDA model achieved a classification accuracy of 92.3%, and the BO‐SVM model reached a classification accuracy of 96.2%. The SOM clustering effectively discerned samples with different fillers, as evidenced by distinct peak numbers and shapes in the differential Raman spectra, thereby underscoring the practical significance of SOM clustering. In comparison with FDA, BO‐SVM exhibited enhanced classification accuracy and exceptional performance in handling outliers and linearly inseparable data, indicating its superior generalization capabilities. [ABSTRACT FROM AUTHOR]

In predicting building energy (affected by seasons), there are issues like inefficient hyperparameter optimization and inaccurate predictions, it is unclear whether spatial and temporal attention improves performance. This study proposes a method based on Bayesian Optimization (BO), Spatial-Temporal Attention (STA), and Long Short-Term Memory (LSTM). Seven improved LSTM models (BO-LSTM, SA-LSTM, TA-LSTM, STA-LSTM, BO-SA-LSTM, BO-TA-LSTM, and BO-STA-LSTM) are compared with LSTM and the impacts of seasonal variations on BO-STA-LSTM are analysed using different sample types and time domain analysis. To further demonstrate the efficiency of the proposed method, comparisons with convolutional neural network (CNN) and temporal convolutional network (TCN) are performed, and followed by validation with new datasets. The findings indicate that adding STA and BO to LSTM enhances the average R2 of prediction performance by 0.0885. BO alone contributes 0.0717, while adding attention improves 0.0560. BO-STA-LSTM achieves higher prediction accuracy for similar test and training samples or a test sample size of 14,016, effectively capturing seasonal, trend, and peak energy patterns. Additionally, BO-STA-LSTM outperforms CNN and TCN, demonstrating superior prediction accuracy.

Diabetic retinopathy (DR) is a cause of vision loss and blindness, and early detection of red lesions, particularly microaneurysms (MAs), is crucial to prevent its progression. In this study, we propose an effective deep learning-based DR segmentation and classification method that leverages hyper-parameter optimisation to achieve high detection accuracy. In our proposed method, we first pre-processed retinal images to reduce background variability. Then, we employed a squirrel search optimised bidirectional ConvLSTM U-net with an attention gate for DR segmentation in the main processing step. Finally, we used a Bayesian-optimised Convolutional Neural Network for classification. We selected the optimal architectures for our proposed CNNs and optimised their hyper-parameters using the Bayesian optimisation technique.Our study demonstrates that the proposed CNNs effectively classify diabetic retinopathy in fundus images when model hyper-parameters are tuned with the help of Bayesian optimisation. We conducted statistical tests based on the receiver operating characteristic curve, histogram, and ANOVA test to verify the effectiveness of our suggested models.We proposed a deep learning-based DR segmentation and classification method that leverages hyper-parameter optimisation to achieve high detection accuracy. This study highlights the potential of Bayesian optimisation as an effective technique to optimise hyper-parameters of deep learning networks for disease classification tasks. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy of the mine water surge prediction, based on the instability and randomness of the water surge data, a mine water surge prediction model CEEMDAN-BO-BiGRU was proposed, combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), Bayesian optimization (BO) and bi-directional gated circulation unit (BiGRU). The proposed model decomposed the influx of water data into a number of smoother intrinsic mode function (IMF) and residual components (Res) through the CEEMDAN, filtered data noise, extracted of data fluctuations in different time scales characteristics, reduced the prediction error. Bayesian optimization was used to iteratively seek the optimization of multiple hyperparameters of the BiGRU model to further improve the prediction accuracy of the model. Afterwards, each component was predicted 1 to 3 steps ahead of the prediction, and finally the prediction results of each component were summed to obtain the multi-step prediction results of the water influx. The CEEMDAN-BO-BiGRU prediction results were compared and analyzed with other prediction models. The results show that the CEEM-DAN-BO-BiGRU combined network model is more accurate for the prediction of mine water surges, and the method provides a new idea for the short-time prediction of water surges. [ABSTRACT FROM AUTHOR]

In the dynamically evolving landscape of cybersecurity, safeguarding IT infrastructures has emerged as an imperative to thwart the escalation of cyber-attacks. Anomaly-based Intrusion Detection Systems (IDS) play a pivotal role in identifying aberrant behaviours that elude conventional detection mechanisms. Nonetheless, these systems are not without their shortcomings, manifesting as elevated false alarm rates and a diminished efficacy in detecting sophisticated attacks. In response to these challenges, a hybrid approach, entailing Machine Learning (ML) techniques, was employed to augment the performance of anomaly-based IDS in terms of detection accuracy, False Positive (FP) Rate, and detection time. The approach encompassed a two-fold optimization strategy: initial feature selection predicated on feature importance derived from the XGBoost classifier, followed by Bayesian optimization (BO) for hyperparameter tuning. The optimization was conducted with respect to two objective functions, namely the ROC-AUC score and the Average Precision score, each serving to identify the optimal hyperparameters for their respective maximization. Classifiers, including Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Stochastic Gradient Descent (SGD), were subjected to training under configurations encompassing both the hyperparameters resultant from BO and the default hyperparameters, the latter serving as reference models. Evaluation, conducted through a multifaceted metric analysis, substantiated the superiority of the optimized models over their reference counterparts, with the optimized XGBoost models demonstrating the most commendable performance. This paradigm offers a promising avenue for enhancing detection precision and mitigating false alarms, thereby fortifying the security of computer systems. [ABSTRACT FROM AUTHOR]

Cutting tool wear monitoring (TWM) plays a significant role because it guarantees the machined surface integrity. Therefore, the present article proposed a TWM system using Bayesian optimized-support vector regression (BO-SVR) analysis. This objective was realized by acquiring machined surface texture video during machining from an in-situ CMOS camera, and subsequently, analyzing it by feature extraction, selection, predictive model training, model hyperparameters optimization, and model testing and validation. To develop an in-process TWM system, machined surface video is acquired during the machining process, and analyzed using Gabor wavelet (GW) and grey level co-occurrence matrix (GLCM) to extract the information related to roughness, feed marks, and waviness of texture. The significant features are selected using the fisher discriminant ratio (FDR) analysis. The in-process TWM system is trained using the FDR selected features and the predictive model hyperparameters such as C, gamma, epsilon, and kernel type are optimized using the Bayesian optimization algorithm, and their optimized results are 99.51, 0.55, 0.01186, and RBF. An optimized hyperparameters are used to establish an accurate and reliable in-process TWM system. The prediction model accuracy is compared with experimentally measured tool wear, the proposed BO-SVR model can predict tool wear with an RMSE of 0.026494. [ABSTRACT FROM AUTHOR]

Bayesian optimization (BO)-assisted screening was applied to identify improved reaction conditions toward a hundred-gram scale-up synthesis of 2,3,7,8-tetrathiaspiro[4.4]nonane (1), a key synthetic intermediate of 2,2-bis(mercaptomethyl)propane-1,3-dithiol [tetramercaptan pentaerythritol]. Starting from the initial training set (ITS) consisting of six trials sampled by random screening for BO, suitable parameters were predicted (78% conversion yield of spiro-dithiolane 1) within seven experiments. Moreover, BO-assisted screening with the ITS selected by Latin hypercube sampling (LHS) further improved the yield of 1 to 89% within the eight trials. The established conditions were confirmed to be satisfactory for a hundred grams scale-up synthesis of 1. [ABSTRACT FROM AUTHOR]

Introduction: Decoding brain states from subcortical local field potentials (LFPs) indicative of activities such as voluntary movement, tremor, or sleep stages, holds significant potential in treating neurodegenerative disorders and offers new paradigms in brain-computer interface (BCI). Identified states can serve as control signals in coupled human-machine systems, e.g., to regulate deep brain stimulation (DBS) therapy or control prosthetic limbs. However, the behavior, performance, and efficiency of LFP decoders depend on an array of design and calibration settings encapsulated into a single set of hyper-parameters. Although methods exist to tune hyper-parameters automatically, decoders are typically found through exhaustive trial-and-error, manual search, and intuitive experience. Methods: This study introduces a Bayesian optimization (BO) approach to hyper-parameter tuning, applicable through feature extraction, channel selection, classification, and stage transition stages of the entire decoding pipeline. The optimization method is compared with five real-time feature extraction methods paired with four classifiers to decode voluntary movement asynchronously based on LFPs recorded with DBS electrodes implanted in the subthalamic nucleus of Parkinson’s disease patients. Results: Detection performance, measured as the geometric mean between classifier specificity and sensitivity, is automatically optimized. BO demonstrates improved decoding performance from initial parameter setting across all methods. The best decoders achieve a maximum performance of 0.74 ± 0.06 (mean ± SD across all participants) sensitivity-specificity geometric mean. In addition, parameter relevance is determined using the BO surrogate models. Discussion: Hyper-parameters tend to be sub-optimally fixed across different users rather than individually adjusted or even specifically set for a decoding task. The relevance of each parameter to the optimization problem and comparisons between algorithms can also be difficult to track with the evolution of the decoding problem. We believe that the proposed decoding pipeline and BO approach is a promising solution to such challenges surrounding hyper-parameter tuning and that the study’s findings can inform future design iterations of neural decoders for adaptive DBS and BCI. [ABSTRACT FROM AUTHOR]

Copyright of Machine Tool & Hydraulics is the property of Guangzhou Mechanical Engineering Research Institute (GMERI) and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In materials science, traditional techniques for analyzing layered structures are essential for obtaining information about local structure, electronic properties and chemical states. While valuable, these methods often require high vacuum environments and have limited depth profiling capabilities. The grazing exit x-ray absorption near-edge structure (GE-XANES) technique addresses these limitations by providing depth-resolved insight at ambient conditions, facilitating in situ material analysis without special sample preparation. However, GE-XANES is limited by long data acquisition times, which hinders its practicality for various applications. To overcome this, we have incorporated Bayesian optimization (BO) into the GE-XANES data acquisition process. This innovative approach potentially reduces measurement time by a factor of 50. We have used a standard GE-XANES experiment, which serve as reference, to validate the effectiveness and accuracy of the BO-informed experimental setup. Our results show that this optimized approach maintains data quality while significantly improving efficiency, making GE-XANES more accessible to a wider range of materials science applications.

IntroductionDecoding brain states from subcortical local field potentials (LFPs) indicative of activities such as voluntary movement, tremor, or sleep stages, holds significant potential in treating neurodegenerative disorders and offers new paradigms in brain-computer interface (BCI). Identified states can serve as control signals in coupled human-machine systems, e.g., to regulate deep brain stimulation (DBS) therapy or control prosthetic limbs. However, the behavior, performance, and efficiency of LFP decoders depend on an array of design and calibration settings encapsulated into a single set of hyper-parameters. Although methods exist to tune hyper-parameters automatically, decoders are typically found through exhaustive trial-and-error, manual search, and intuitive experience.MethodsThis study introduces a Bayesian optimization (BO) approach to hyper-parameter tuning, applicable through feature extraction, channel selection, classification, and stage transition stages of the entire decoding pipeline. The optimization method is compared with five real-time feature extraction methods paired with four classifiers to decode voluntary movement asynchronously based on LFPs recorded with DBS electrodes implanted in the subthalamic nucleus of Parkinson’s disease patients.ResultsDetection performance, measured as the geometric mean between classifier specificity and sensitivity, is automatically optimized. BO demonstrates improved decoding performance from initial parameter setting across all methods. The best decoders achieve a maximum performance of 0.74 ± 0.06 (mean ± SD across all participants) sensitivity-specificity geometric mean. In addition, parameter relevance is determined using the BO surrogate models.DiscussionHyper-parameters tend to be sub-optimally fixed across different users rather than individually adjusted or even specifically set for a decoding task. The relevance of each parameter to the optimization problem and comparisons between algorithms can also be difficult to track with the evolution of the decoding problem. We believe that the proposed decoding pipeline and BO approach is a promising solution to such challenges surrounding hyper-parameter tuning and that the study’s findings can inform future design iterations of neural decoders for adaptive DBS and BCI.

This article presents an automated, model-free, data-driven method for the safe tuning of PID cascade controller gains based on Bayesian optimization. The optimization objective is composed of data-driven performance metrics and modeled using Gaussian processes. The safety requirement is imposed via a barrier-like term in the objective, which is introduced to account for operational changes in the system. We further introduce a data-driven constraint that captures the stability requirements from system data. Numerical evaluation shows that the proposed approach outperforms relay feedback autotuning and quickly converges to the global optimum, thanks to a tailored stopping criterion. We demonstrate the performance of the method through simulations and experiments. For experimental implementation, in addition to the introduced safety constraint, we integrate a method for automatic detection of the critical gains and extend the optimization objective with a penalty depending on the proximity of the current candidate points to the critical gains. The resulting automated tuning method optimizes system performance while ensuring stability and standardization. [ABSTRACT FROM AUTHOR]

Interturn short circuit (ITSC) fault diagnosis at its early stage is very essential to improve the security of permanent magnet synchronous motors. In this article, a Bayesian optimization (BO) based residual convolutional neural network (CNN) algorithm for ITSC fault diagnosis was proposed. There are mainly two aspects of the contributions. First, it is a challenge to apply a conventional CNN on time-series signal tasks as they can only look back on history with a liner size, which will limit the network depth. Moreover, to obtain enough characteristics from acquired signals, the proposed algorithm requires adequate network depth, which may result in degradation difficulties. To solve these problems, a residual connection is embedded into the dilated CNN. Second, with the increase of network depth, the possible combination of hyperparameters will grow geometrically, thus the hyperparameters tuning will cost considerable time. To overcome this, the BO algorithm is adopted to select the optimal hyperparameter combination automatically. To demonstrate the effectiveness of the proposed algorithm, a motor fault experiment with various operating conditions was conducted on the motor that can be set for 17 fault severity levels. The test results and comparisons with other five algorithms show the advantage of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Spiking Neural Networks are often touted as brain-inspired learning models for the third wave of Artificial Intelligence. Although recent SNNs trained with supervised backpropagation show classification accuracy comparable to deep networks, the performance of unsupervised learning-based SNNs remains much lower. This paper presents a heterogeneous recurrent spiking neural network (HRSNN) with unsupervised learning for spatio-temporal classification of video activity recognition tasks on RGB (KTH, UCF11, UCF101) and event-based datasets (DVS128 Gesture). We observed an accuracy of 94.32%for the KTHdataset, 79.58%and 77.53%for theUCF11 and UCF101 datasets, respectively, and an accuracy of 96.54% on the event-based DVS Gesture dataset using the novel unsupervised HRSNN model. The key novelty of the HRSNN is that the recurrent layer in HRSNN consists of heterogeneous neurons with varying firing/relaxation dynamics, and they are trained via heterogeneous spike-time-dependent-plasticity (STDP) with varying learning dynamics for each synapse. We show that this novel combination of heterogeneity in architecture and learning method outperforms current homogeneous spiking neural networks. We further show that HRSNN can achieve similar performance to state-of-the-art backpropagation trained supervised SNN, but with less computation (fewer neurons and sparse connection) and less training data. [ABSTRACT FROM AUTHOR]

This paper presents novel contributions to both theory and solution methodology in AI-based analysis of solid mechanics. The physics-informed neural network (PINN) method is developed for thermoelastic wave propagation and Moore-Gibson-Thompson (MGT) coupled thermoelasticity analysis of porous media, a first in the field. The coupled thermoelasticity governing equations, based on the MGT heat conduction model, are derived for a porous half-space, with the thermal relaxation coefficient and strain relaxation factor being considered. Mechanical and thermal shock loading boundary conditions are imposed. The behavior of a magnesium-made porous body is analyzed using the PINN method, with highly accurate results being achieved for the system of coupled PDEs. An adaptive hyperparameter tuning approach, integrating a generalized subset design (GSD) and Bayesian optimization algorithm, is used to automatically select the optimal structure based on the L2 relative error. This hybrid methodology eliminates manual adjustment concerns. The proposed method is verified through a thorough comparison with the Lord-Shulman theory of coupled thermoelasticity. The strength of the methodology lies in its ability to operate without domain data, with only boundary and initial points being required. Four example sets are examined to demonstrate the capabilities of the modified PINN, and high-quality predictions of dimensionless fields' variables over an extended time interval are obtained, confirming its extrapolation abilities. [ABSTRACT FROM AUTHOR]

Soil moisture content (SMC) acquisition is vital for crop stress diagnosis and precision irrigation. However, UAV remote sensing-based SMC monitoring usually suffers from low accuracy and spatio-temporal applicability. To address these issues, this study integrated crop physiological spectral response features into an ensemble learning model. A two-year field experiment (2022–2023) was conducted. First, fractional-order differentiation (FOD) and continuous wavelet transform (CWT) were used to enhance the responsiveness of summer maize canopy spectra to SMC and leaf physiological parameters (LPPs), including leaf area index (LAI), leaf chlorophyll content (LCC) and leaf water content (LWC). Afterwards, variable importance in projection (VIP) was adopted to characterize the spectral response properties of each parameter. Finally, a stacked ensemble learning model (SELM) based on Bayesian optimization (BO) was used to construct a SMC monitoring model, and the feasibility of fusing LPP spectral response features for SMC monitoring was evaluated. The results indicated that: (1) Changes in SMC had significant effect on LCC and LAI of summer maize, which in T4 (with field water-holding capacity of 80 %-95 %) were 14.07 % and 34.41 % higher than that in T1 (40 %-50 %), respectively. (2) Spectral transformation could significantly enhance the correlation between SMC and LPPs with canopy spectra (the average increase of R reached 0.23). (3) Consideration of the crop physiology spectral response could improve the SMC monitoring accuracy, the LCC-FOD-BO-SELM had excellent monitoring performance (R²= 0.78; RMSE = 0.019). The monitoring model fusing LPPs spectral response features had the highest accuracy (R²= 0.81; RMSE = 0.016). (4) BO could significantly reduce the overfitting problem of the monitoring model, with the maximum difference between the R2 of the training and test set after BO being only 0.01), thus improving the model's generalizability. This new approach to SMC monitoring using UAV hyperspectral data provide scientific support for precision agriculture and irrigation. [Display omitted] • FOD and CWT methods significantly enhance spectral characteristics. • LPPs spectral response features significantly improve SMC monitoring performance. • Bayesian optimization greatly improves the robustness of SELM [ABSTRACT FROM AUTHOR]

Spiking Neural Networks are often touted as brain-inspired learning models for the third wave of Artificial Intelligence. Although recent SNNs trained with supervised backpropagation show classification accuracy comparable to deep networks, the performance of unsupervised learning-based SNNs remains much lower. This paper presents a heterogeneous recurrent spiking neural network (HRSNN) with unsupervised learning for spatio-temporal classification of video activity recognition tasks on RGB (KTH, UCF11, UCF101) and event-based datasets (DVS128 Gesture). We observed an accuracy of 94.32% for the KTH dataset, 79.58% and 77.53% for the UCF11 and UCF101 datasets, respectively, and an accuracy of 96.54% on the event-based DVS Gesture dataset using the novel unsupervised HRSNN model. The key novelty of the HRSNN is that the recurrent layer in HRSNN consists of heterogeneous neurons with varying firing/relaxation dynamics, and they are trained via heterogeneous spike-time-dependent-plasticity (STDP) with varying learning dynamics for each synapse. We show that this novel combination of heterogeneity in architecture and learning method outperforms current homogeneous spiking neural networks. We further show that HRSNN can achieve similar performance to state-of-the-art backpropagation trained supervised SNN, but with less computation (fewer neurons and sparse connection) and less training data.

Reflectarrays (RAs) have been attracting considerable interest in recent years due to their appealing features, in particular, the possibility of realizing pencil-beam radiation patterns, as in the phased arrays, but without the necessity of incorporating the feeding networks. These characteristics make them attractive solutions, among others, for satellite communications or mobile radar antennas. Notwithstanding, available microstrip implementations are inherently narrow-band and heavily affected by conductor and surface wave losses. RAs based on grounded dielectric layers offer improved performance and flexibility in terms of shaping the phase reflection response. In either case, a large number of variables (induced by the need for independent adjustment of individual unit cell geometries) and the necessity of handling several requirements make the design process of reflectarrays a challenging endeavor. In particular, RA optimization is extremely expensive when conducted at the level of electromagnetic (EM) simulation models, otherwise necessary to ensure reliability. A practical solution is a surrogate-assisted design with the metamodels constructed for the RA unit elements. Unfortunately, conventional modeling methods require large numbers of training data samples to render accurate surrogates, which turns detrimental to the optimization process efficiency. This work proposes an alternative approach with the unit element representations constructed using deep learning with automated adjustment of the model architecture. As a result, design-ready surrogates can be constructed using only a few hundred samples, and the total RA optimization cost is reduced to only a handful of equivalent EM analyses of the entire array. Our approach is validated using an RA incorporating 3-D pyramidal-shaped elements and favorably compared to benchmark techniques. Experimental verification of the obtained design is discussed as well. [ABSTRACT FROM AUTHOR]

An RF integrated circuit design heavily relies upon experienced experts to iteratively tune the circuit parameters. A Bayesian optimization (BO) method is explored in existing works for automated analog and RF circuit synthesis. The overall performance can be further improved by constructing a model to exploit the correlation among different circuit specifications. In this article, we propose a BO approach for RF circuit synthesis via a multitask neural network enhanced Gaussian process (MTNN-GP). We present a novel multioutput GP model, in which the kernel functions of multiple outputs are constructed from a multitask neural network with shared hidden layers and task-specific layers. Therefore, the correlation between the outputs can be captured by the shared hidden layers. Our proposed MTNN-GP-based BO method is compared with several state-of-the-art BO methods on three real word RF circuits and achieves best performance. The experimental results demonstrate the effectiveness and efficiency of our proposed method. [ABSTRACT FROM AUTHOR]

The methods for remaining useful life (RUL) prediction of bearings are mainly based on the autoregressive strategies, among which the temporal convolutional network (TCN) has been recently developed and is widely believed as the high-performance one. These methods generally suffer from errors of prediction. In this article, we newly design the Bayesian-optimization-based adversarial TCN (AdTCN-BO), by embedding the TCN into the adversarial training framework as the generator. Within the framework, the discriminator is designed to continuously correct the output value of the generator in the training process, thus reducing the errors of prediction to a certain extent. Based on the AdTCN-BO, a novel RUL prediction approach for bearings is developed. An experimental verification is carried out to validate the effectiveness of the proposed approach, demonstrating that the AdTCN-BO framework is more accurate in contrast to the traditional data-driven methods of RUL prediction. [ABSTRACT FROM AUTHOR]

Some response surface functions in complex engineering systems are usually highly nonlinear, unformed, and expensive to evaluate. To tackle this challenge, Bayesian optimization (BO), which conducts sequential design via a posterior distribution over the objective function, is a critical method used to find the global optimum of black-box functions. Kernel functions play an important role in shaping the posterior distribution of the estimated function. The widely used kernel function, e.g., radial basis function (RBF), is very vulnerable and susceptible to outliers; the existence of outliers is causing its Gaussian process (GP) surrogate model to be sporadic. In this article, we propose a robust kernel function, asymmetric elastic net radial basis function (AEN-RBF). Its validity as a kernel function and computational complexity are evaluated. When compared with the baseline RBF kernel, we prove theoretically that AEN-RBF can realize smaller mean squared prediction error under mild conditions. The proposed AEN-RBF kernel function can also realize faster convergence to the global optimum. We also show that the AEN-RBF kernel function is less sensitive to outliers, and hence improves the robustness of the corresponding BO with GPs. Through extensive evaluations carried out on synthetic and real-world optimization problems, we show that AEN-RBF outperforms the existing benchmark kernel functions. Note to Practitioners—Some industrial systems cannot be accurately represented by physical models. In this situation, data-driven black-box optimization is necessary for advancing the system automation and intelligence. BO is one of the widely used strategies for learning the global optimum of black-box functions. BO has been applied to robotics, anomaly detection, automatic learning algorithm configuration, reinforcement learning, and deep learning. This article proposes one new kernel function, named after AEN-RBF. The new kernel function will make BO with GPs more robust to outliers and lower the data quality barrier of model training. This article was motivated by the hyperparameter tuning problem of deep learning models for image defect detection in advanced manufacturing, but the method can be easily extended to other applications where kernel functions are needed. Our proposed method is verified by synthetic and real-world optimization problems. [ABSTRACT FROM AUTHOR]

In this article, a new motion-aware sampling strategy (MASS) is presented to speed up the measurement of radiation patterns around cylindrical surfaces. Differently from preexisting sampling techniques, the MASS directly chooses positions that reduce the overall travel time of the field antenna, rather than minimizing the total number of samples. The proposed strategy employs a Gaussian process model that is adapted to the field over a cylindrical surface. Moreover, a new acquisition function for Bayesian active learning is developed in order to efficiently search the peaks of the measured field and predict their values. Next, the proposed strategy is tested on the experimental data from a radiation pattern of a comb generator. Finally, the results are compared to standard grid sampling and Bayesian optimization strategies. [ABSTRACT FROM AUTHOR]

In this letter, a new Bayesian optimization (BO) method with dynamic feasible region shrinkage (DFRS) technique for power amplifier (PA) design is proposed. As a powerful optimization tool, it provides a more effective way to optimize the performance of PA than the embedded commercial optimization tools. It also has a better convergence speed than the existing fixed mode acquisition function. Results show that the new technique provides a great optimization for PA design, not only for circuit optimization but also for electromagnetic (EM) optimization. [ABSTRACT FROM AUTHOR]

To maximize the maintenance willingness of the owner of transmission lines, this study presents a transmission maintenance scheduling model that considers the energy constraints of the power system and the security constraints of on- site maintenance operations. Considering the computational complexity of the mixed integer programming (MIP) problem, a machine learning (ML) approach is presented to solve the transmission maintenance scheduling model efficiently. The value of the branching score factor value is optimized by Bayesian optimization (BO) in the proposed algorithm, which plays an important role in the size of the branch-and-bound search tree in the solution process. The test case in a modified version of the IEEE 30-bus system shows that the proposed algorithm can not only reach the optimal solution but also improve the computational efficiency.

Bayesian optimization (BO)-assisted screening was applied to identify improved reaction conditions toward a hundred-gram scale-up synthesis of 2,3,7,8-tetrathiaspiro[4.4]nonane (1), a key synthetic intermediate of 2,2-bis(mercaptomethyl)propane-1,3-dithiol [tetramercaptan pentaerythritol]. Starting from the initial training set (ITS) consisting of six trials sampled by random screening for BO, suitable parameters were predicted (78% conversion yield of spiro-dithiolane 1) within seven experiments. Moreover, BO-assisted screening with the ITS selected by Latin hypercube sampling (LHS) further improved the yield of 1 to 89% within the eight trials. The established conditions were confirmed to be satisfactory for a hundred grams scale-up synthesis of 1.

Bayesian optimization (BO) is well known to be sample efficient for solving black-box problems. However, BO algorithms may get stuck in suboptimal solutions even with plenty of samples. Intrinsically, such a suboptimal problem of BO can attribute to the poor surrogate accuracy of the trained Gaussian process (GP), particularly that in the regions where the optimal solutions locate. Hence, we propose to build multiple GP models instead of a single GP surrogate to complement each other, thus resolving the suboptimal problem of BO. Nevertheless, according to the bias-variance tradeoff equation, the individual prediction errors can increase when increasing the diversity of models, which may lead even worse overall surrogate accuracy. On the other hand, based on the theory of the Rademacher complexity, it has been proven that exploiting the agreement of models on unlabeled information can reduce the complexity of hypothesis space, therefore achieving the required surrogate accuracy with fewer samples. Such value of model agreement has been extensively demonstrated for co-training style algorithms to boost model accuracy with a small portion of samples. Inspired by the above, we propose a novel BO algorithm labeled as co-learning BO (CLBO), which exploits both model diversity and agreement on unlabeled information to improve the overall surrogate accuracy with limited samples, therefore achieving more efficient global optimization. Through tests on five numerical toy problems and three engineering benchmarks, the effectiveness of the proposed CLBO has been well demonstrated. [ABSTRACT FROM AUTHOR]

Nanoparticle-enhanced lubricants (nanolubricants) achieve tribological properties that can signficantly outperform the unmodified base oil. However, this performance enhancement is known to be extremely senstive to the concentration of nanoparticles used. Here, we introduce a new approach to optimizing nanolubricants combining design of experiments (DOE) with machine learning, addressing the challenge of determining optimal nanoparticle concentration in base oils and demonstrating significant performance improvements. Combining Box-Behnken design with Bayesian optimization, we identify an optimal concentration of 0.04 wt% copper oxide (CuO) nanoparticles in a standard engine oil (SAE 20W50), resulting in a 42% reduction in coefficient of friction (COF) relative to the base oil. Surface analysis (AFM, SEM) confirmed improvement of anti-wear and anti-friction characteristics. Signficant enhacements of thermophysical properties were observed (demonstrated by increased flash and pour points, by 10% and 7% respectively), signifying greater thermal stability. The dynamic viscosity remained consistent, with a beneficial reduction observed under conditions of high temperature and shear, which aligns with the demands of engine lubricant standards for PC-12B oils, with a reduction from 3.49 to 3.18 cP (8%). This research introduces a novel hybrid optimization technique that provides a versatile framework for optimizing nanofluids, addressing the critical challenge of determining optimal nanoparticle concentration. [Display omitted] • A hybrid optimization strategy was developed to optimize the concentration of CuO nanoparticles in nanolubricants. • The optimized nanolubricant demonstrated an impressive 42% reduction in the coefficient of friction (COF). • AFM surface analysis confirmed enhancements in anti-wear and anti-friction properties. • Enhanced thermophysical properties underscored the positive impact of the hybrid optimization tool. • Dynamic viscosity remained stable, improving under high temperature and shear conditions in line with PC-12B oil standards. [ABSTRACT FROM AUTHOR]

Real-world systems, such as manufacturing or computer systems, can be modeled as multistate network (MSN) consisting of arcs with stochastic capacity. Network reliability for an MSN is described as the probability that the system can meet the demand. The network reliability for demand level d can be computed in terms of the minimal path (calledd-MP). However, efficiently calculating network reliability is challenging in large-scale networks. Deep learning approaches are rapidly advancing several areas of technology, with significant applications in image recognition, parameter adjustment, and autonomous driving. Hence, in this study, we adopt a deep neural network (DNN) model to predict network reliability for a given demand level. To train the DNN model, network information is first used as input data. Then, a DNN model is constructed, including the determination of related functions. Furthermore, Bayesian optimization (BO) is applied to determine related hyperparameters. A practical implementation using a bridge network demonstrates the feasibility of the DNN model. Finally, experiments involving two networks with more nodes and arcs indicate the computational efficiency of combining deep learning methods and the existing d-MP algorithm. [ABSTRACT FROM AUTHOR]

This article aims at developing a gradient boosting (GB) model with parameters optimized by the Bayesian optimization (BO) method for detecting various types of fault in an oil-immersed transformer and reactor. The developed model depicts higher accuracy in comparison with other machine learning (ML) techniques, such as XGBoost, AdaBoost, random forest, and support vector machine parameters being optimized by BO. A comparison of the skill score depicts the superiority of the developed Bayesian optimization-based gradient boosting (BO-GB) model with respect to Bayesian optimized other ML techniques. Further comparison of the model with conventional Duval pentagon verifies the accuracy of detecting the faults, which were missed by the former. Also, the software has been developed based on the proposed BO-GB method for the on-site detection of the fault. In addition, actual values of dissolved gases with respect to fault confirmed after the internal inspection of the transformer and reactor are tabulated and compared. The data so presented will help in utilizing the authentic dissolved gas data in addition to the already available IEC TC10 databases for future works. The study undertaken in this article will help utilities in diagnosing the incipient faults more accurately and precisely, thereby preventing hazardous operating conditions along with minimizing the downtime cost. [ABSTRACT FROM AUTHOR]

With ever-increasing transistor density in nanoscale-integrated circuits, the impact of process variations on circuit performance and chip yield becomes dominant. To prevent failures in the field, simulation-based circuit optimization tools are performed during design time as a countermeasure. However, the efficiency of these tools requires accurate modeling of failure probabilities. Especially, for high-sigma problems such as yield estimation of memory cells, which require very high production yield, the failure rate assessment must be highly accurate. Importance sampling (IS) methods are deployed in this context to uncover very rare failure events, which cannot be revealed by standard Monte Carlo methods. Besides a highly accurate yield prediction model, the limited time budget of the simulation-based analysis tools has to be taken into account. Especially in conjunction with yield optimization techniques, the required number of circuit simulations for the failure rate estimation has to be substantially reduced. In this article, we propose a yield optimization method, which is based on a Bayesian optimization (BO) failure rate estimation technique for high-sigma yield extraction. The BO-based IS method is combined with a kernel density estimator for finding the most probable failure events, which have significant contribution to the chip yield. By integration into a global optimization framework, we show how the proposed yield optimization method can be applied to high-sigma yield optimization problems, such as memory cells. The experimental results indicate that the proposed method consumes only 5% circuit simulations to achieve the same optimization effect as the state-of-the-art yield optimizer techniques. [ABSTRACT FROM AUTHOR]

In this article, we propose a performance-based autotuning method for cascade control systems, where the parameters of a linear axis drive motion controller from two control loops are tuned jointly. Using Bayesian optimization as all parameters are tuned simultaneously, the method is guaranteed to converge asymptotically to the global optimum of the cost. The data-efficiency and performance of the method are studied numerically for several training configurations and compared numerically to those achieved with classical tuning methods and to the exhaustive evaluation of the cost. On the real system, the tracking performance and robustness against disturbances are compared experimentally to nominal tuning. The numerical study and the experimental data both demonstrate that the proposed automated tuning method is efficient in terms of required tuning iterations, robust to disturbances, and results in improved tracking. [ABSTRACT FROM AUTHOR]

Bayesian optimization (BO) is a well-developed machine learning (ML) field for black-box function optimization. In BO, a surrogate predictive model, here a Gaussian process, is used to approximate the black-box function. The estimated mean and uncertainty of the surrogate model are paired with an acquisition function to decide where to sample next. In this study, we applied this technique to known ferromagnetic thin-film materials such as ferromagnetic (Fe100−yGay)1−xBx (x = 0−21 and y = 9−17) to demonstrate optimization of structure–property relationships, specifically the dopant concentration or stoichiometry effect on magnetostriction and ferromagnetic resonance linewidth. Our results demonstrated that BO can be deployed to optimize structure–property relationships in FeGaB and FeGaC thin films. We have shown through simulation that using BO methods to guide experiments reduced the number of samples required to statistically determine the maximum or minimum by 50% compared to traditional methods. Our results suggest that BO can be used to save time and resources to optimize ferromagnetic films. This method is transferrable to other ferromagnetic material structure–property relationships, providing an accessible implementation of ML to magnetic materials development. [ABSTRACT FROM AUTHOR]

Financial experts seek to predict the variability of financial markets to ensure investors’ successful investments. However, there has been a big trend in finance in the last few years, which are the ESG (Economic, Social and Governance) criteria, due to the growing importance of investments being socially responsible, and because of the financial impact companies suffer when not complying with them. Consequently, creating a stock portfolio should consider not only its financial performance but compliance with ESG criteria. Portfolio optimization (PO) techniques previously applied to ESG portfolios, are all closed-form analytical ones. But the real world is rather a black box with unknown analytical expressions. Thus, in this paper we use Bayesian optimization (BO), a sequential state-of-the-art design strategy to optimize black-boxes with unknown analytical and costly-to-compute expressions, to maximize the performance of a stock portfolio under the presence of ESG criteria soft constraints incorporated into the objective function. And we compare it to two other black-box techniques widely applied for the optimization of ‘conventional portfolios’ (non-ESG ones): the metaheuristics Genetic algorithm (GA) and Simulated Annealing (SA). Although BO has many theoretical advantages over GA and SA, it has never been applied to PO. Thus, this paper investigates whether BO can be used in the ESG PO framework as an alternative and compares it with GA and SA. This is the research gap to which this paper responds. To show the empirical performance of BO, we carry out four illustrative experiments and find evidence of BO outperforming the baselines. Thus we add another different optimization approach to the world of ESG investing: a black-box non-heuristic optimization approach through BO. Our study is the first paper that leverages BO and ESG scores into a PO technique. This paper opens the door to many new research lines in (ESG) portfolio optimization.