Regression Accuracy Research Articles

Infrared emissivity measurement methods considering target reflective characteristics

Emissivity measurements are of great significance for infrared thermal radiation and infrared remote sensing. However, traditional methods often face challenges such as difficulties in non-contact measurement, small measurement areas, and unsuitability for non-Lambertian surfaces. To address these issues, we propose the reflective distribution model integral method (RDMIM). This method is based on a reflective distribution model using the scattering-reflective deviation angle (SRDA). By regressing and integrating the object’s reflective distribution model, it achieves accurate non-contact measurement of non-Lambertian surfaces under normal temperature conditions. Additionally, the measurement scheme has been further optimized to improve measurement efficiency while ensuring the accuracy of the model regression. Finally, the proposed RDMIM method has been validated through experimental measurements. The results have shown that this method has advantages in non-contact and large-area measurements. Moreover, the systematic error is smaller when the reflective characteristics of the reference body and the target are relatively similar.

Music-induced emotion flow modeling by ENMI Network.

The relation between emotions and music is substantial because music as an art can evoke emotions. Music emotion recognition (MER) studies the emotions that music brings in the effort to map musical features to the affective dimensions. This study conceptualizes the mapping of music and emotion as a multivariate time series regression problem, with the aim of capturing the emotion flow in the Arousal-Valence emotional space. The Efficient Net-Music Informer (ENMI) Network was introduced to address this phenomenon. The ENMI was used to extract Mel-spectrogram features, complementing the time series data. Moreover, the Music Informer model was adopted to train on both time series music features and Mel-spectrogram features to predict emotional sequences. In our regression task, the model achieved a root mean square error (RMSE) of 0.0440 and 0.0352 in the arousal and valence dimensions, respectively, in the DEAM dataset. A comprehensive analysis of the effects of different hyperparameters tuning was conducted. Furthermore, different sequence lengths were predicted for the regression accuracy of the ENMI Network on three different datasets, namely the DEAM dataset, the Emomusic dataset, and the augmented Emomusic dataset. Additionally, a feature ablation on the Mel-spectrogram features and an analysis of the importance of the various musical features in the regression results were performed, establishing the effectiveness of the model presented herein.

Low-altitude UAV obstacle detection method based on position constraint and attention

UAVs are widely used in low-altitude fields for tasks such as power inspection, search and rescue, and reconnaissance. Pre-detection of obstacles during flight is a safety guarantee for completing the given tasks. In order to meet the requirements of obstacle detection accuracy and position regression accuracy when UAVs are flying at low altitudes, a low-altitude UAV obstacle detection method based on position constraint and attention improvement is proposed. Firstly, the shortcomings of the position regression loss function are analyzed, and based on this, a loss function of separating scale loss and integrating direction constraint is proposed to optimize the regression process; secondly, the attention mechanism CBAM is improved, and a dual attention mechanism is proposed to strengthen the feature suppression interference and improve the detection performance. The experimental results show that the improved algorithm has improved mAP 2.28%and mAP@0.5:0.95 2.7%, and has shown better low-altitude obstacle detection performance in both detection accuracy and position regression accuracy.

Accuracy and efficiency of 2D/3D single-vertebra spine navigation registration method based on dual-view feature fusion

Objective: To investigate the accuracy and efficiency of spine 2D/3D preoperative CT and intraoperative X-ray registration through a framework for spine 2D/3D single-vertebra navigation registration based on the fusion of dual-position image features. Methods: The preoperative CT and intraoperative anteroposterior (AP) and lateral (LAT) X-ray images of 140 lumbar spine patients who visited Huashan Hospital Affiliated to Fudan University from January 2020 to December 2023 were selected. In order to achieve rapid and high-precision single vertebra registration in clinical orthopedic surgery, a designed transformation parameter feature extraction module combined with a lightweight module of channel and spatial attention (CBAM) was used to accurately extract the local single vertebra image transformation information. Subsequently, the fusion regression module was used to complement the features of the anterior posterior (AP) and lateral (LAT) images to improve the accuracy of the registration parameter regression. Two 1×1 convolutions were used to reduce the parameter calculation amount, improve computational efficiency, and accelerate intraoperative registration time. Finally, the regression module outputed the final transformation parameters. Comparative experiments were conducted using traditional iterative methods (Opt-MI, Opt-NCC, Opt-C2F) and existing deep learning methods convolutional neural network (CNN) as control group. The registration accuracy (mRPD), registration time, and registration success rate were compared among the iterative methods. Results: Through experiments on real CT data, the image-guided registration accuracy of the proposed method was verified. The method achieved a registration accuracy of (0.81±0.41) mm in the mRPD metric, a rotational angle error of 0.57°±0.24°, and a translation error of (0.41±0.21) mm. Through experimental comparisons on mainstream models, the selected DenseNet alignment accuracy was significantly better than ResNet as well as VGG (both P<0.05). Compared to existing deep learning methods [mRPD: (2.97±0.99) mm, rotational angle error: 2.64°±0.54°, translation error: (2.15±0.41) mm, registration time: (0.03±0.05) seconds], the proposed method significantly improved registration accuracy (all P<0.05). The registration success rate reached 97%, with an average single registration time of only (0.04±0.02) seconds. Compared to traditional iterative methods [mRPD: (0.78±0.26) mm, rotational angle error: 0.84°±0.57°, translation error: (1.05±0.28) mm, registration time: (35.5±10.5) seconds], registration efficiency of the proposed method was significantly improved (all P<0.05). The dual-position study also compensated for the limitations in the single-view perspective, and significantly outperforms both the front and side single-view perspectives in terms of positional transformation parameter errors (both P<0.05). Conclusion: Compared to existing methods, the proposed CT and X-ray registration method significantly reduces registration time while maintaining high registration accuracy, achieving efficient and precise single vertebra registration.

A helmet detection algorithm based on improved Yolov8n

Abstract To address the issues of false positives and missed detections in current safety helmet algorithms, we propose a new detection method using an enhanced yolov8n model. An enhanced Yolov8n-based algorithm for detecting safety helmets is proposed. In this experiment, a new module called C2f_ECA is proposed. This module can extract channel information and enhance the performance of the network by replacing the original loss function of the algorithm with the EIoU loss function. The aspect ratio loss term is now determined by the difference between the predicted dimensions of width and height and those of the minimum bounding box. This modification speeds up the convergence of predicted boxes and improves their regression accuracy. In addition, a dataset of safety helmets is constructed in complex scenarios. The proposed algorithm enhances Yolov8n’s mAP50 by 1.6% and average accuracy by 1.6% compared to the original and traditional Yolov8 algorithms, respectively. It can efficiently address the issues of false positives and missed detections of safety helmets.

Investigating carbon peak prediction scenarios and implementation strategies in higher education institutions: A case study of Shandong Jianzhu university

The potential for energy conservation and emission reduction in Chinese universities is significant. The exploration of scenarios for predicting carbon peaking and identifying implementation pathways can offer valuable insights and demonstrations for achieving dual carbon goals. This paper develops a research methodology for carbon peaking in universities, with a focus on four main aspects: carbon emission accounting, the analysis of influencing factors, carbon emission prediction, and implementation pathways for carbon peaking. By utilizing the campus of Shandong Jianzhu University as a case study, the paper practically demonstrates the prediction of carbon peaking scenarios and offers technical solutions for universities. The results demonstrate that: (1) Using the Logarithmic Mean Divisia Index (LMDI) decomposition method to investigate the influencing factors of carbon emissions in universities, we confirm that the energy structure effect, economic effect, and population effect have a promoting impact on the increase of carbon emissions at Shandong Jianzhu University, while the energy intensity effect exhibits a significant inhibitory effect. (2) By employing Partial Least Squares (PLS) regression analysis of annual carbon emission statistics from universities, we construct a STIRPAT model with high predictive accuracy for carbon emission regression. The carbon emission prediction model for Shandong Jianzhu University is: lnC^t=-25.899+3.571lnPt+0.966lnARt+0.716lnTMt-0.279lnTEt (3) Both the low-carbon scenario and the enhanced scenario for Shandong Jianzhu University are capable of achieving carbon peaking by 2027. From the perspectives of energy-saving renovation of existing building envelopes, improving the efficiency of lighting systems, utilizing solar photovoltaic buildings, and electrifying cooking energy, implementation pathways for carbon peaking under low-carbon and enhanced scenarios are delineated

Model for atomization droplet size and energy distribution ratio at the distal end of an electrostatic nozzle

Electrostatic atomization minimum quantity lubrication (EMQL) employs the synergistic effect of multiple physical fields to atomize minute quantities of lubricant. This innovative methodology is distinguished by its capacity to ameliorate the atomization attributes of the lubricant substantially, which subsequently augments the migratory and infiltration proficiency of the droplets within the complex and demanding milieu of the cutting zone. Compared with the traditional minimum quantity lubrication (MQL), the EMQL process is further complicated by the multiphysical field influences. The presence of multiple physical fields not only increases the complexity of the forces acting on the liquid film but also induces changes in the physical properties of the lubricant itself, thus making the analysis of atomization characteristics and energy distribution particularly challenging. To address this objective reality, the current study has conducted a meticulous measurement of the volume average diameter, size distribution span, and the percentage concentration of inhalable particles of the charged droplets at various intercept positions of the EMQL nozzle. A predictive model for the volume-averaged droplet size at the far end of the EMQL nozzle was established with the observed statistical value F of 825.2125, which indicates a high regression accuracy of the model. Furthermore, based on the changes in the potential energy of surface tension, the loss of kinetic energy of gas, and the electric field work at different nozzle orifice positions in the EMQL system, an energy distribution ratio model for EMQL was developed. The energy distribution ratio coefficients under operating conditions of 0.1 MPa air pressure and 0 to 40 kV voltage on the 20 mm cross-section ranged from 3.094‰ to 3.458‰, while all other operating conditions and cross-sections had energy distribution ratios below 2.06‰. This research is expected to act as a catalyst for the progression of EMQL by stimulating innovation in the sphere of precision manufacturing, providing theoretical foundations, and offering practical guidance for the further development of EMQL technology.

Reducing Measurement Costs of Thermal Power: An Advanced MISM (Mamba with Improved SSM Embedding in MLP) Regression Model for Accurate CO2 Emission Accounting.

Current calculation methods for the carbon content as received (Car) of coal rely on multiple instruments, leading to high costs for enterprises. There is a need for a cost-effective model that maintains accuracy in CO2 emission accounting. This study introduces an MISM model using key parameters identified through correlation and ablation analyses. An Improved State-Space Model (ISSM) and an IS-Mamba module are integrated into a Multi-Layer Perceptron (MLP) framework, enhancing information flow and regression accuracy. The MISM model demonstrates superior performance over traditional methods, reducing the Root Mean Square Error (RMSE) by 22.36% compared to MLP, and by 9.65% compared to Mamba. Using only six selected parameters, the MISM model achieves a precision of 0.27% for the discrepancy between the calculated CO2 emissions and the actual measurements. An ablation analysis confirms the importance of certain parameters and the effectiveness of the IS-Mamba module at improving model performance. This paper offers an innovative solution for accurate and cost-effective carbon accounting in the thermal power sector, supporting China's carbon peaking and carbon neutrality goals.

Enhanced Hand–Eye Coordination Control for Six-Axis Robots Using YOLOv5 with Attention Module

Utilizing machine vision technology based on YOLOv5, a six-axis robot can quickly identify and classify targets. However, when the YOLOv5 model is used for the recognition and grasping of small workpieces, issues such as low precision and missed detections frequently occur. This paper proposes an enhanced object recognition algorithm, integrating a CBAM attention module and an improved loss function into YOLOv5 to control the hand–eye coordination of the six-axis robot during grasping. The CBAM attention module is incorporated into the backbone network of YOLOv5 to enhance its feature extraction capabilities, while the original loss function is modified to accelerate convergence and improve regression accuracy. An experimental platform for six-axis robot hand–eye coordination grasping was built, and grasping experiments were conducted. The proposed method significantly improves the robot’s grasping accuracy, with a 99.59% mAP0.5 and a 90.83% successful grasping rate, effectively addressing the challenges of low accuracy and missed detections in traditional systems.

Automated data-driven discovery of material models based on symbolic regression: A case study on the human brain cortex

We introduce a data-driven framework to automatically identify interpretable and physically meaningful hyperelastic constitutive models from sparse data. Leveraging symbolic regression, our approach generates elegant hyperelastic models that achieve accurate data fitting with parsimonious mathematic formulas, while strictly adhering to hyperelasticity constraints such as polyconvexity/ellipticity. Our investigation spans three distinct hyperelastic models—invariant-based, principal stretch-based, and normal strain-based—and highlights the versatility of symbolic regression. We validate our new approach using synthetic data from five classic hyperelastic models and experimental data from the human brain cortex to demonstrate algorithmic efficacy. Our results suggest that our symbolic regression algorithms robustly discover accurate models with succinct mathematic expressions in invariant-based, stretch-based, and strain-based scenarios. Strikingly, the strain-based model exhibits superior accuracy, while both stretch-based and strain-based models effectively capture the nonlinearity and tension-compression asymmetry inherent to the human brain tissue. Polyconvexity/ellipticity assessment affirm the rigorous adherence to convexity requirements both within and beyond the training regime. However, the stretch-based models raise concerns regarding potential convexity loss under large deformations. The evaluation of predictive capabilities demonstrates remarkable interpolation capabilities for all three models and acceptable extrapolation performance for stretch-based and strain-based models. Finally, robustness tests on noise-embedded data underscore the reliability of our symbolic regression algorithms. Our study confirms the applicability and accuracy of symbolic regression in the automated discovery of isotropic hyperelastic models for the human brain and gives rise to a wide variety of applications in other soft matter systems. Statement of significanceOur research introduces a pioneering data-driven framework that revolutionizes the automated identification of hyperelastic constitutive models, particularly in the context of soft matter systems such as the human brain. By harnessing the power of symbolic regression, we have unlocked the ability to distill intricate physical phenomena into elegant and interpretable mathematical expressions. Our approach not only ensures accurate fitting to sparse data but also upholds crucial hyperelasticity constraints, including polyconvexity, essential for maintaining physical relevance.

Permeability Characteristics of Improved Loess and Prediction Method for Permeability Coefficient

Due to its unique geotechnical properties, loess presents itself as a cost-effective and energy-efficient material for engineering construction, aiding in cost reduction and environmental sustainability. However, to meet engineering specifications, loess often requires enhancement. Evaluating its permeability properties holds significant importance for employing improved loess for construction materials in landfills and artificial water bodies. This study investigates the influence of dry densities, grain size characteristics, grain size distribution, and admixture contents and types on the permeability of improved loess, focusing on the Malan and Lishi loess. The falling head permeability test was conducted to analyze how each factor affects the permeability of the improved loess. The findings indicate that the permeability coefficient decreases with increased dry density and admixture content. Conversely, it demonstrates a linear increase with the average grain size (d50), restricted grain size (d60), and the product of the coefficient of uniformity and coefficient of curvature (Cu × Cc). The primary influencing factor is the type of admixture, followed by Cc and d60. Furthermore, this study developed a predictive model for permeability using a support vector machine (SVM), surpassing the predictive accuracy of linear regression and neural network models. The model provides a robust prediction for the permeability of superior loess material.

Refined Prior Guided Category-Level 6D Pose Estimation and Its Application on Robotic Grasping

Estimating the 6D pose and size of objects is crucial in the task of visual grasping for robotic arms. Most current algorithms still require the 3D CAD model of the target object to match with the detected points, and they are unable to predict the object’s size, which significantly limits the generalizability of these methods. In this paper, we introduce category priors and extract high-dimensional abstract features from both the observed point cloud and the prior to predict the deformation matrix of the reconstructed point cloud and the dense correspondence between the reconstructed and observed point clouds. Furthermore, we propose a staged geometric correction and dense correspondence refinement mechanism to enhance the accuracy of regression. In addition, a novel lightweight attention module is introduced to further integrate the extracted features and identify potential correlations between the observed point cloud and the category prior. Ultimately, the object’s translation, rotation, and size are obtained by mapping the reconstructed point cloud to a normalized canonical coordinate system. Through extensive experiments, we demonstrate that our algorithm outperforms existing methods in terms of performance and accuracy on commonly used benchmarks for this type of problem. Additionally, we implement the algorithm in robotic arm-grasping simulations, further validating its effectiveness.

Improved Pedestrian Vehicle Detection for Small Objects Based on Attention Mechanism

Abstract This study aims to solve the low detection accuracy and susceptibility to false detection and omission in pedestrian and vehicle detection by proposing an improved YOLOv5s algorithm. Firstly, a small target detection module is added to better acquire and determine the information of pedestrians from long-range vehicles. Secondly, the multi-scale channel attention CBAM attention module is added, and the dual attention mechanism is not only flexible and convenient, but also improves the computational efficiency. Finally, the MPDIoU loss function based on minimum point distance is introduced to replace the original GIoU loss function, and this change not only enhances the regression accuracy of the model. At the same time, the convergence speed of the model is accelerated. KITTI data set was used for experiments, and the experimental results showed that the average accuracy of the model trained by the improved YOLOv5s algorithm on the data set reached 84.9%, which was 3.7% higher than that of the original YOLOv5s algorithm. It is verified that the model is suitable for high accuracy of pedestrian and vehicle recognition in complex environments, and has high value for promotion.

A lightweight joint metric detection approach on YOLO for hot spots in photovoltaic modules

The hot spot effect is one of the primary causes of damage to photovoltaic (PV) modules and a significant factor contributing to the decline in their power generation capacity. Thermal imaging inspection of PV modules is an indispensable aspect of PV plant operation and maintenance. This article introduces a lightweight detection algorithm for hot spots in PV modules based on an enhanced version of YOLOv8. The algorithm incorporates a lightweight convolution operator, Volumetric Grid Spatial Cross Stage Partial, as a replacement for the original C2f operator, resulting in effective reduction of the model's parameter size and improved detection speed. Additionally, the Content-Aware ReAssembly of FEatures upsampling operator is utilized instead of the nearest-neighbor algorithm during the upsampling process, thereby minimizing the loss of hot spot feature information. Furthermore, a joint metric approach combining the Complete Intersection over Union metric and the normalized Wasserstein distance metric is proposed to calculate the localization loss of the target bounding boxes, thereby enhancing the regression accuracy of small hot spot prediction boxes. Experimental results demonstrate that the improved YOLOv8 network enables fast and accurate detection of hot spots. This method achieves an impressive average precision of 98.8% and a high detection speed of 218.3 fps.

Detection of Subtle ECG Changes Despite Superimposed Artifacts by Different Machine Learning Algorithms

Analyzing electrocardiographic (ECG) signals is crucial for evaluating heart function and diagnosing cardiac pathology. Traditional methods for detecting ECG changes often rely on offline analysis or subjective visual inspection, which may overlook subtle variations, particularly in the case of artifacts. In this theoretical, proof-of-concept study, we investigated the potential of five different machine learning algorithms [random forests (RFs), gradient boosting methods (GBMs), deep neural networks (DNNs), an ensemble learning technique, as well as logistic regression] to detect subtle changes in the morphology of synthetically generated ECG beats despite artifacts. Following the generation of a synthetic ECG beat using the standardized McSharry algorithm, the baseline ECG signal was modified by changing the amplitude of different ECG components by 0.01–0.06 mV. In addition, a Gaussian jitter of 0.1–0.3 mV was overlaid to simulate artifacts. Five different machine learning algorithms were then applied to detect differences between the modified ECG beats. The highest discriminatory potency, as assessed by the discriminatory accuracy, was achieved by RFs and GBMs (accuracy of up to 1.0), whereas the least accurate results were obtained by logistic regression (accuracy approximately 10% less). In a second step, a feature importance algorithm (Boruta) was used to discriminate which signal parts were responsible for difference detection. For all comparisons, only signal components that had been modified in advance were used for discretion, demonstrating that the RF model focused on the appropriate signal elements. Our findings highlight the potential of RFs and GBMs as valuable tools for detecting subtle ECG changes despite artifacts, with implications for enhancing clinical diagnosis and monitoring. Further studies are needed to validate our findings with clinical data.

Comparison of Linear Regression and LSTM (Long Short-Term Memory) in Cryptocurrency Prediction

Abstract Cryptocurrency, particularly Bitcoin, has become a major topic in the financial and digital trading sectors due to its ability to facilitate direct transactions without intermediaries and the transparency offered by blockchain technology. However, the high volatility of Bitcoin prices necessitates accurate prediction methods to support better investment decisions. This research aims to compare the accuracy of Linear Regression and Long Short-Term Memory (LSTM) methods in predicting Bitcoin prices using historical data from Yahoo Finance. The research process begins with the collection of historical Bitcoin price data from September 17, 2014, to July 15, 2024, followed by data processing that includes cleaning and splitting the dataset into training and test data. Linear Regression and LSTM models are applied to the training data and tested to evaluate their performance in price prediction. The research findings show that the LSTM model significantly outperforms the Linear Regression model in terms of prediction accuracy. The LSTM model records much lower Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), as well as perfect R² scores on both datasets, demonstrating its high precision in prediction. In contrast, the Linear Regression model shows higher errors and lower explanatory power of data variability. These findings indicate that LSTM is more effective in capturing temporal patterns and Bitcoin price fluctuations, offering better accuracy and potentially being more suitable for future cryptocurrency price analysis, providing better guidance for investors in this highly dynamic market.

A multiband terahertz metamaterial sensor evaluated using a multiresponse variable fusion framework for the precise detection of trace fluoroquinolone antibiotics

A specific structural multiband terahertz (THz) metamaterial sensor (TMS) was designed herein to improve the sensitivity of detecting multiple trace fluoroquinolone (FQ) antibiotics in water, and a multiple resonance response competition and selection fusion (MRR-CSF) framework was proposed for accurate FQ analysis. The TMS based on three-concentric C4-symmetrical split-ring (TC4) structures generated five bands at 0.797, 0.900, 1.097, 1.198, and 1.777 THz. The multiband resonance signals for three types of FQs, including enrofloxacin (ENR), pefloxacin (PEF), and nadifloxacin (NAD), were detected by the TMS, which increased as the antibiotic concentrations increased (5–140 ng/L). The proposed method can solve the problems of low accuracy and poor robustness of conventional single-response regression by combining selected and fused TMS-effective resonance response variables with three multivariate regression strategies. The calculated comprehensive metrics also provide an objective evaluation of multiband TMS performance. The designed TMS combined with the proposed MRR-CSF framework showed 21 %–27 %, 13 %–17 %, and 12 %–13 % improvements in the detection accuracy for ENR, PEF, and NAD, respectively, according to the univariate regression results. The limit of detection for the ENR, PEF, and NAD antibiotic solutions was achieved at 5 ng/L, and the reliable range of recoveries (100 % ± 25 %), average recoveries (>90 %), and relative standard deviation values (<5 %) were obtained for the concentration prediction of newly prepared FQs. This study demonstrates the promising prospect of applying the multiband TMS combined with the MRR-CSF framework for accurate biochemical sensing and detection.

A dielectric method for predicting pear firmness combining deep data augmentation and ensemble learning

Firmness is valuable in evaluating pear quality, as it determines ripeness and storability. This research presents a new method that combines deep data augmentation and ensemble learning for predicting firmness in three pear cultivars using dielectric spectra. Firstly, 126 sets of pear dielectric spectra and firmness data were obtained, and the characteristic dielectric frequencies were selected based on the prediction accuracy of partial least squares regression (PLSR) model established using different preprocessing methods. Then the generative adversarial networks (GAN), least squares GAN (LSGAN), and Wasserstein GAN (WGAN) models with different numbers of generated samples in the model's training set on the prediction accuracy of the PLSR model were compared. Finally, the optimal GAN model was combined with support vector regression (SVR), artificial neural networks (ANN), random forest (RF), and AdaBoost decision tree (AdaBoost-DT) to predict pear firmness. Results showed that the selected characteristic dielectric frequencies included two ε′ and eight ε" points. GAN, LSGAN, and WGAN all improved the prediction accuracy of the PLSR model, and the model accuracy improvement of GAN and LSGAN was proportional to the generated sample numbers added to the training set. GAN had the highest accuracy improvement for PLSR model, with the mean Rp2 (determination coefficient of prediction set) and RMSEP (root mean square error of prediction set) increasing up to 13.70% and −19.48% during 100–1000 epochs (added 126 sets of generated data in the GAN-PLSR model' training set), respectively. Ensemble learning models (RF, AdaBoost-DT) outperformed SVR and ANN. GAN-AdaBoost-DT model added 126 sets of generated data in the training set had the highest prediction accuracy, with average Rp2 and RMSEP of 0.90 and 3.35 N during 300–1000 epochs, respectively. This study proved the feasibility of deep data augmentation and ensemble learning methods to improve the prediction accuracy of pear firmness model without increasing the labor and time costs.

Spectrum is a picture: Feasibility study of two-dimensional convolutional neural networks in spectral processing

In chemometrics, a spectrum is usually represented in a form of numeric vector for further processing. The same approach has been used also for spectral data treatment by convolutional neural networks (CNN), initially purposed, however, for image processing. Analyzing spectral data as a two-dimensional picture rather than a one-dimensional vector potentially can improve the accuracy of regression and classification models. The purpose of this work was to test this assumption. As a first case study we have chosen one of the most difficult types of spectra to interpret – the Mössbauer spectral data. We explored the potential of 2D-CNN to predict Mössbauer spectral parameters numerically using image (.bmp) files with spectra. As a second case study, we address another challenging task – classification of biological tissues using their near-infrared spectra. In both cases, we compared the performance of CNN for two types of input data: spectra as pictures and as numeric vectors. It was found that in case of Mossbauer spectra, the use of 2D-CNN provides for better prediction of spectral parameters, while for NIR spectra of biological tissues the use of the traditional PLS-DA method turned out to be better for classification compared to the CNN approach.

Using Machine Learning Methods to Assess Module Performance Contribution in Modular Optimization Frameworks.

Modular algorithm frameworks not only allow for combinations never tested in manually selected algorithm portfolios, but they also provide a structured approach to assess which algorithmic ideas are crucial for the observed performance of algorithms. In this study, we propose a methodology for analyzing the impact of the different modules on the overall performance. We consider modular frameworks for two widely used families of derivative-free black-box optimization algorithms, the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and differential evolution (DE). More specifically, we use performance data of 324 modCMA-ES and 576 modDE algorithm variants (with each variant corresponding to a specific configuration of modules) obtained on the 24 BBOB problems for 6 different runtime budgets in 2 dimensions. Our analysis of these data reveals that the impact of individual modules on overall algorithm performance varies significantly. Notably, among the examined modules, the elitism module in CMA-ES and the linear population size reduction module in DE exhibit the most significant impact on performance. Furthermore, our exploratory data analysis of problem landscape data suggests that the most relevant landscape features remain consistent regardless of the configuration of individual modules, but the influence that these features have on regression accuracy varies. In addition, we apply classifiers that exploit feature importance with respect to the trained models for performance prediction and performance data, to predict the modular configurations of CMA-ES and DE algorithm variants. The results show that the predicted configurations do not exhibit a statistically significant difference in performance compared to the true configurations, with the percentage varying depending on the setup (from 49.1% to 95.5% for mod-CMA and 21.7% to 77.1% for DE).