MLP Structure Research Articles

TOR-GAN: A Transformer-Based OFDM Signals Reconstruction GAN

Reconstruction techniques for communication signals represent a significant research focus within the field of signal processing. To overcome the difficulty and low precision in reconstructing OFDM signals, we introduce a signal reconstruction technique called TOR-GAN (Transformer-Based OFDM Signal Reconstruction GAN). Reconstructing IQ sequences using a CNN and RNN presents challenges in capturing the correlations between two signals. To tackle this issue, the VIT (vision in transformer) approach was introduced into the discriminator network. The IQ signal is treated as a single-channel, two-dimensional image, divided into blocks of 2 × 2 pixels, with absolute position embedding added. The generator network maps the input noise to the same dimension as the IQ signal dimension × embedding vector dimension and adds two identical position embedding data points to the network learning. In the transformer network, prob sparse attention is employed as a replacement for multi-head attention to tackle the issue of high computational complexity. Finally, combined with the MLP structure, the transformer-based generator and discriminator are designed. The signal similarity evaluation index was constructed, and experiments showed that the reconstructed signal under QPSK and BPSK modulation had good reconstruction quality in the time-domain waveform, constellation diagram, and spectrogram at a high SNR. Compared with other reconstruction algorithms, the proposed algorithm improved the quality of the reconstructed signal while reducing the complexity of the algorithm.

Learning Global and Multi-granularity Local Representation with MLP for Sequential Recommendation

Sequential recommendation aims to predict the next item of interest to users based on their historical behavior data. Usually, users’ global and local preferences jointly affect the final recommendation result in different ways. Most existing works use transformers to globally model sequences, which makes them face the dilemma of quadratic computational complexity when dealing with long sequences. Moreover, the scope setting of the user’s local preference is usually static and single, and cannot cover richer multi-level local semantics. To this end, we proposed a parallel architecture for capturing global representation and M ulti-granularity L ocal dependencies with M LP for sequential Rec ommendation ( MLM4Rec ). For global representation, we utilize modified MLP-Mixer to capture global information of user sequences due to its simplicity and efficiency. For local representation, we incorporate convolution into MLP and propose a multi-granularity local awareness mechanism for capturing richer local semantic information. Moreover, we introduced a weight pooling method to adaptively fuse local-global representations instead of directly concatenation. Our model has the advantages of low complexity and high efficiency thanks to its simple MLP structure. Experimental results on three public datasets demonstrate the effectiveness of our proposed model. Our code is available here 1 .

LWR-Net: Robust and Lightweight Place Recognition Network for Noisy and Low-Density Point Clouds.

Point cloud-based retrieval for place recognition is essential in robotic applications like autonomous driving or simultaneous localization and mapping. However, this remains challenging in complex real-world scenes. Existing methods are sensitive to noisy, low-density point clouds and require extensive storage and computation, posing limitations for hardware-limited scenarios. To overcome these challenges, we propose LWR-Net, a lightweight place recognition network for efficient and robust point cloud retrieval in noisy, low-density conditions. Our approach incorporates a fast dilated sampling and grouping module with a residual MLP structure to learn geometric features from local neighborhoods. We also introduce a lightweight attentional weighting module to enhance global feature representation. By utilizing the Generalized Mean pooling structure, we aggregated the global descriptor for point cloud retrieval. We validated LWR-Net's efficiency and robustness on the Oxford robotcar dataset and three in-house datasets. The results demonstrate that our method efficiently and accurately retrieves matching scenes while being more robust to variations in point density and noise intensity. LWR-Net achieves state-of-the-art accuracy and robustness with a lightweight model size of 0.4M parameters. These efficiency, robustness, and lightweight advantages make our network highly suitable for robotic applications relying on point cloud-based place recognition.

An informative teaching model of English-Chinese translation in universities based on the perceptual machine model

Abstract With the development of globalization, there is an increasing demand for English-Chinese translation talents in the country and society. In this paper, a text classifier based on multilayer perceptron is proposed, an MLP structure containing two layers of feedforward neural network is designed for the characteristics of English-Chinese translation corpus, the training samples are selected with text height mainly below 60 pixels, and the best sliding window of 13×13 is selected by comparison, supplemented by the smoothing method of mathematical morphology. The constructed MLP-based information-based teaching model is then applied to teaching English-Chinese translation in colleges and universities and is mainly evaluated in terms of student evaluation and teaching effectiveness. There were 28.48 percentage points more English-Chinese translation students who thought the class was more interesting than the original, while 20.56% and 20.50% more students evaluated the class as more helpful and less helpful, respectively. The average score of English-Chinese translation students increased by 15.88% compared to the original one through the comparison test between traditional and informational teaching modes. The MLP-based informatization teaching can mobilize students’ enthusiasm to participate in English-Chinese translation learning, strengthen the guidance of translation methods, and improve students’ translation levels.

MLP-Like Model With Convolution Complex Transformation for Auxiliary Diagnosis Through Medical Images.

Medical images such as facial and tongue images have been widely used for intelligence-assisted diagnosis, which can be regarded as the multi-label classification task for disease location (DL) and disease nature (DN) of biomedical images. Compared with complicated convolutional neural networks and Transformers for this task, recent MLP-like architectures are not only simple and less computationally expensive, but also have stronger generalization capabilities. However, MLP-like models require better input features from the image. Thus, this study proposes a novel convolution complex transformation MLP-like (CCT-MLP) model for the multi-label DL and DN recognition task for facial and tongue images. Notably, the convolutional Tokenizer and multiple convolutional layers are first used to extract the better shallow features from input biomedical images to make up for the loss of spatial information obtained by the simple MLP structure. Subsequently, the Channel-MLP architecture with complex transformations is used to extract deep-level contextual features. In this way, multi-channel features are extracted and mixed to perform the multi-label classification of the input biomedical images. Experimental results on our constructed multi-label facial and tongue image datasets demonstrate that our method outperforms existing methods in terms of both accuracy (Acc) and mean average precision (mAP).

Multi-objective optimization of viscosity and thermal conductivity of TiO2/BioGlycol-water nanofluids with sorting non-dominated genetic algorithm II coupled with response surface methodology

The study aimed to optimize of viscosity and thermal conductivity (TC) properties of TiO2/BG-Water (20:80) and TiO2/BG-Water (30:70) nanofluids (NFs). In this study, simulation methods (NSGA-II, RSM, ANN) are used with the aim of faster and more accurate results, saving time and exorbitant laboratory costs. TC and viscosity of TiO2/BG-Water (20:80) NF in RSM have regression coefficients R2 = 0.9913 and R2 = 0.9982, respectively. For TiO2/BG-Water (30:70) NF in RSM, TC and viscosity have regression coefficients R2 = 0.9911 and R2 = 0.9993, respectively. These results indicate the high accuracy of the predicted model. Using ANN, the MLP structure with the best efficiency for each NFs is introduced. In the MLP model, TC and viscosity of TiO2/BG-Water (20:80) NF have regression coefficient R2 = 0.9997 and TC and viscosity of TiO2/BG-Water (30:70) NF have regression coefficient R2 = 0.9998. Finally, after defining the variables and objective functions in the NSGA-II method, optimization of multi-objective was performed using RSM and MOPSO. Pareto fronts are introduced to evaluate optimal viscosity and TC responses. The results of the Pareto front show that optimum TC and viscosity of NFs occur when T and solid volume fraction (SVF) are at their maximum. The exact results of this study from the prediction and optimization of TC and viscosity of these NFs, can help craftsmen to better understand the properties of heat transfer.

CO2 capture by benzene‐based hypercrosslinked polymer adsorbent: Artificial neural network and response surface methodology

AbstractIn this research, porous benzene‐based hypercrosslinked polymeric adsorbents with different morphological properties were synthesized through Friedel–Crafts alkylation reaction. The resulting samples were applied for CO2 capture at different operational conditions. Two modelling approaches, including artificial neural network (radial basis function [RBF] and multi layer perceptron [MLP]) and response surface methodology (RSM), were employed to investigate the effect of independent parameters on adsorption capacity. A semi‐empirical quadratic model for adsorption capacity was presented based on RSM‐central composite design technique. Additionally, the optimal structure of RBF was determined with 200 neurons, and the optimal structure of MLP was determined with three hidden layers and 10, 8, and 7 neurons. The modelling results demonstrate the better prediction of MLP and RBF approaches than the RSM method with correlation coefficient values of 0.999, 0.989, and 0.931, respectively. Finally, process optimization was carried out using RSM optimization module and the optimized values of synthesis time, crosslinker ratio (formaldehyde dimethyl acetal [FDA]/benzene), adsorption time, pressure, and temperature were obtained at 10.11 h, 1, 220 s, 9 bar, and 55°C, respectively. The optimum value of CO2 uptake capacity was obtained around 167 (mg/g).

Optimization of density and coefficient of thermal expansion of MWCNT in thermal oil nanofluid and modeling using MLP and response surface methodology

In this study, the rheological properties including viscosity, thermal conductivity (TC), density and coefficient of thermal expansion of MWCNT nanofluid based on Thermal Oil (THO) were investigated. The laboratory study was carried out by a two-step method and at a solid volume fraction of SVF= 0.1–1% and a temperature of T = 20 °C. RSM and MLP intelligent methods were used for modeling and optimization. The convergence presented for TC, viscosity, density and thermal expansion (TE) coefficient have a convergence coefficient of 0.9961, 0.9969, 0.9999 and 0.9999, respectively, which shows the high accuracy of the relations. The results show that increasing the SVF increases the TC, viscosity and density of NFs and decreases the TE coefficient of NFs. Increasing temperature increases TC and TE coefficient and decreases the viscosity and density of NF. The MLP structure was obtained with two hidden layers Tansig and Logsig and with convergence coefficients of 0.9980, 0.9449 and 1 respectively for TC, viscosity, density and TE coefficient. For viscosity and density, both RSM and ANN models show high accuracy, but the MLP model for TC and RSM model for TE coefficient had higher accuracy.

PCAPN: An Enhanced Feature Extraction Framework for Point Cloud

Point cloud is a widely used geometric data structure in the missions of 3D reconstruction, digital city and geologic survey etc. Extracting sufficient information from the point cloud is the key to deal with the aforementioned missions. However, huge number of points lead to computational complexity and inefficiency during the training process. To deal with this problem, this paper proposes a novel framework name Principal Component Analysis Point Net (PCAPN) for the feature extraction of point cloud. Firstly, a sampling module namely Component Point Sampling (CPS) is designed for generating several candidate sets of points by different scales, which defines the centroids of local regions. Secondly, a feature extraction framework based on MLP structure is adopted for extracting the feature vectors from the points set generated from the sampling module. Finally, the extracted feature vectors are concatenated together, and then put into a fully connected layer for classification. The proposed framework was evaluated on 2 benchmarks, i.e. ShapeNet part dataset and ModelNet40. The experimental results show that our framework is efficient and robust. In particular, the results are significantly better than those obtained by the state-of-the-art frameworks. Our network is 4.6% more accurate than PointNet and 1.1% higher than PointNet++.

Application of Artificial Neural Networks to the Analysis of Friction Behaviour in a Drawbead Profile in Sheet Metal Forming.

Drawbeads are used when forming drawpieces with complex shapes to equalise the flow resistance of a material around the perimeter of the drawpiece or to change the state of stress in certain regions of the drawpiece. This article presents a special drawbead simulator for determining the value of the coefficient of friction on the drawbead. The aim of this paper is the application of artificial neural networks (ANNs) to understand the effect of the most important parameters of the friction process (sample orientation in relation to the rolling direction of the steel sheets, surface roughness of the counter-samples and lubrication conditions) on the coefficient of friction. The intention was to build a database for training ANNs. The friction coefficient was determined for low-carbon steel sheets with various drawability indices: drawing quality DQ, deep-drawing quality DDQ and extra deep-drawing quality EDDQ. Equivalents of the sheets tested in EN standards are DC01 (DQ), DC03 (DDQ) and DC04 (EDDQ). The tests were carried out under the conditions of dry friction and the sheet surface was lubricated with machine oil LAN46 and hydraulic oil LHL32, commonly used in sheet metal forming. Moreover, various specimen orientations (0° and 90°) in relation to the rolling direction of the steel sheets were investigated. Moreover, a wide range of surface roughness values of the counter-samples (Ra = 0.32 μm, 0.63 μm, 1.25 μm and 2.5 μm) were also considered. In general, the value of the coefficient of friction increased with increasing surface roughness of the counter-samples. In the case of LAN46 machine oil, the effectiveness of lubrication decreased with increasing mean roughness of the counter-samples Ra = 0.32-1.25 μm. With increasing drawing quality of the sheet metal, the effectiveness of lubrication increased, but only in the range of surface roughness of the counter-samples in which Ra = 0.32-1.25 μm. This study investigated different transfer functions and training algorithms to develop the best artificial neural network structure. Backpropagation in an MLP structure was used to build the structure. In addition, the COF was calculated using a parameter-based analytical equation. Garson partitioning weight was used to calculate the relative importance (RI) effect on coefficient of friction. The Bayesian regularization backpropagation (BRB)-Trainbr training algorithm, together with the radial basis normalized-Radbasn transfer function, scored best in predicting the coefficient of friction with R2 values between 0.9318 and 0.9180 for the training and testing datasets, respectively.

Investigation the effects of different nanoparticles on density and specific heat: Prediction using MLP artificial neural network and response surface methodology

Increasing energy efficiency is of particular importance and therefore the use of nanofluids has been considered due to their thermophysical properties. in this study Response surface methodology (RSM) and MLP artificial neural network (ANN) models are applied for the determination of the density and specific heat capacity. Results indicate that the maximum residual value for the RSM model of density was ±1.5 and for the RSM model of specific heat was + 0.008 and − 0.006. The optimum MLP structure for density is created with 5 neurons in the first layer and 2 neurons in the second layer. For the MLP structure of specific heat, 4 neurons in the first layer and 5 neurons in the second layer are considered. The best MLP structures have MSE, MAE and R2 equal to 0.324113, 0.395038 and 0.9985 for density and 1.89E-05, 3.24E-03 and 0.9974 for thermal conductivity, respectively. The maximum influence of volume fraction of nanoparticles (φ) on the density belongs to TiN50–EG nanofluid and the minimum one belongs to Si3N420–EG nanofluid. TiN50–EG and TiN20–EG nanofluids have the minimum effect on the specific heat.

Extraction and characterization of a functional protein from Millettia speciosa Champ. leaf

Natural plant-derived protein with excellent bioactivities has attracted much attention so a functional protein with molecular weight of 15.2 kDa was extracted from Millettia speciosa Champ. leaf for the first time. Under the pH of 12.0, solid-liquid ratio of 1:40 (w/v), extraction time of 2.0 h, and extraction temperature of 50 °C, the highest extracting efficiency (79.25 ± 0.78%) of the Millettia speciosa Champ. leaf protein (MLP) was achieved. The main structure of MLP contained β-fold and β-corner by Fourier transform infrared spectroscopy (FTIR) and Circular dichroism (CD) spectra analysis. Additionally, MLP was predominant with glutamic acid, aspartic acid, and leucine, which could be considered as a high quality natural protein. MLP showed great water holding capacity (WHC), oil absorption capacity (OAC), as well as emulsifying and foaming properties. Simultaneously, MLP exhibited considerable antioxidant activity. These results suggested that MLP could be utilised as a promising ingredient of functional foods.

Sound-spectrogram based automatic bird species recognition using MLP classifier

Bioacoustics play a major role in the field of ornithology, ecology, animal behaviour study, habitat monitoring, species conservation and design of deterrent system. This work focusses on the design and implementation and performance evaluation of an automatic, more efficient and flexible bird sound based recognition system for classifying eight species of popular Eurasian birds the standard online annotated databases. A dedicated virtual instrument tool with effective GUI is developed that acquires, pre-processes the sound samples and generates statistically evaluated short term Fourier transform spectrogram based feature matrix, suited for characterization of vocalization patterns of bird species. Using the well-labelled feature data of sound records, multi-layer perceptron artificial neural network (MLP-NN) classifier model is designed, trained, tested and optimized using feedforward-backpropagation supervised learning algorithm. Various experiments, following a systematic approach, are conducted to optimize the structure of MLP with respect of number of neurons in the hidden layer, epochs and learning rate for attaining enhanced recognition accuracy (96.1%), recall (82.6%) and precision (84.5%). The performance of the optimal model is also analysed in terms of recognition capabilities of individual bird species that indicate promising results. A few of the birds are recognized accurately and precisely when present as compared to the others. The tendency of the model to wrongly identify or miss bird species also remained low. The model performance over the unseen dataset also remained satisfactory with cross-validation classification accuracy of 81.4%. The system being scalable, can easily be reused in future to retrain the model over the large set of sound samples from real world recordings with improved acoustic features for achieving very high classification accuracy and reliability.

Thermal Conversion Behavior of Medium–Low–Temperature Coal Tar Pitch During Liquid–Phase Carbonization Process

AbstractMedium–low temperature coal tar pitch (MLP) is a type of aromatic carbonaceous material obtained by the low temperature distillation of low rank coal, and is considered as a desirable raw material for the production of artificial carbon/graphite material. The thermal conversion behaviors of MLP during the liquid‐phase carbonization process were analyzed in detail in this study. Further, the changes of molecular structure and carbon microcrystalline structure of MLP and liquid‐phase carbonization products derived from it were quantitatively investigated by optical microscopy, fourier transform infrared spectroscopy, Raman spectroscopy, X‐ray diffraction, and curve‐fitting method. Results showed that both branched chain and the contents of C=O functional groups in the MLP played a key role in the reaction reactivity. Moreover, temperature of 450 °C was found to be a rapid changing temperature point on the enhancement of thermal conversion of MLP. The stacking height (Lc), parallel layers (N), average number of aromatic rings in each layer (n), ratio of ‘impurity’ structure (RI), ratio of amorphous carbon structure (RAC), ratio of graphite carbon structure (RGC), and ratio of defects graphite carbon structure (RDGC) exhibited a significant change with the improvement of liquid‐phase carbonization temperature.

Forecasting the estimated time of arrival for a cargo dispatch delivered by a freight train along a railway section

This paper reports a method for predicting the expected time of arrival (ETA) of a cargo dispatch taking into consideration determining the duration at which a freight train travels along a railroad section where trains move not complying with a departure schedule. A characteristic feature of railroads with such a traffic system is the difficulty in predicting the stages of a transportation process, which necessitates the development of effective methods of forecasting. Based on correlation analysis, we have determined the dependence of the general macro-characteristics of train flow and individual parameters of a freight train on the duration of its movement along a section. It has been proposed to represent the dependence of predicted duration of train movement along a railroad section on the following factors: traffic intensity and density along a section, the proportion of passenger trains in total train flows, the length of a train and its gross weight. All experimental studies are based on actual data on the operation of the distance Osnova-Lyubotyn at the railroad network AO Ukrzaliznytsya.Based on a comparative analysis, using the indicators for accuracy and adequacy of several regression methods to predict ETA of cargo dispatch, we have chosen the regression model based on an artificial neural network MLP. To derive the MLP structure, a cross-validation method has been applied, which implies the validation of a mathematical model reliability based on the criteria of accuracy MAE and adequacy ‒ F-test. The structure of MLP has been obtained, which consists of five hidden layers. We predicted the time that it would take for a train to travel in facing direction along the Osnova-Lyubotyn section. For a given projection, the value for MAE was 0.0845, which is a rather high accuracy for this type of problems, and confirms the effectiveness of MLP application to solve the task on predicting a cargo dispatch ETA.The current study provides a possibility to design in the future an automated system for predicting a cargo dispatch ETA for a mixed-traffic railroad system in which freight trains depart not complying with a regulatory schedule.

VARIATION ON THE NUMBER OF HIDDEN NODES THROUGH MULTILAYER PERCEPTRON NETWORKS TO PREDICT THE CYCLE TIME

Multilayer Perceptron Network (MLP) has a better prediction Multilayer Perceptron Network (MLP) has a better prediction performance compared to other networks since the structure ofthe MLP is suitable for training processes in solving prediction problems. However, to the best of our knowledge, there is no ruleof thumb in determining the number of hidden nodes within the MLP structure. Researchers normally test with various numbers of hidden nodes to obtain the lowest square error value for optimal prediction results since none of the approaches has yet to be claimed as the best practice. Thus, the aim of this study is to determine the best MLP network by varying the number of hidden nodes of developed networks to predict cycle time for producing a new audio product on a production line. The networks were trained and validated through 100 sets of production lots from a selected audio manufacturer. As a result, the 3-2-1 MLP network was the best network based on the lowest square error value compared to the 3-1-1 and 3-3-1 networks. The 3-2-1 predicted the best cycle time of 5 seconds to produce a new audio product. Hence, the prediction result could facilitate production planners in managing assembly processes on the production line.

Mining of electricity prices in energy markets using a hybrid linear ARMA and nonlinear functional link neural network trained by evolutionary unscented H-infinity filter

This paper presents a hybrid autoregressive moving average (ARMA) and a nonlinear functional link neural network for electricity price forecasting in an Energy market. The functional neural block helps to introduce nonlinearity by expanding the input space to higher dimensional space through a basis function without using any hidden layers like MLP structure. Unlike the conventional functional link artificial neural network (FLANN), the input layer comprises the inputs and tangent hyperbolic functions of the linear combination of the inputs known as the basis functions. The proposed hybrid neural network is trained by an unscented H-infinity filter to provide an accurate forecasting of day ahead electricity prices. The noise covariance parameters of the unscented H-infinity filter are further optimised with an adaptive differential evolution strategy. The studies on PJM, Spanish and Australian energy markets exhibit excellent forecasting results over different seasonal horizons for one day ahead of time.

Optimization of a multi-class MLP ECG classifier using FCM

ECG pattern classification using MLP is an effective and robust technique. Due to the inherent structure of MLP and training algorithm, MLPs tend to be slow and bulky in terms of the hidden layer neurons. This condition is aggravated further if the input data dimension is large as in the case of ECG. The paper addresses this problem by optimizing the MLP using an additional clustering of the feature extracted data. Considerable reduction in the size of MLP was recorded (max. 67.9%) with a reduction in training time (maximum of 59.81%). The experimentation used benchmark arrhythmia database from Physionet Massachusetts institute of technology–Beth Israel hospital (MIT-BIH). Four feature extraction methodologies were subjected to fuzzy c-means clustering for obtaining optimized MLPs. Ten statistical morphological features were also considered for designing the MLP classifiers. Keywords: ECG, MLP, DCT, DWT, FCM, morphological features.

Optimization of a multi-class MLP ECG classifier using FCM

ECG pattern classification using MLP is an effective and robust technique. Due to the inherent structure of MLP and training algorithm, MLPs tend to be slow and bulky in terms of the hidden layer neurons. This condition is aggravated further if the input data dimension is large as in the case of ECG. The paper addresses this problem by optimizing the MLP using an additional clustering of the feature extracted data. Considerable reduction in the size of MLP was recorded (max. 67.9%) with a reduction in training time (maximum of 59.81%). The experimentation used benchmark arrhythmia database from Physionet Massachusetts institute of technology-Beth Israel hospital (MIT-BIH). Four feature extraction methodologies were subjected to fuzzy c-means clustering for obtaining optimized MLPs. Ten statistical morphological features were also considered for designing the MLP classifiers.

Previsão de Vazões com a Metodologia DPFT e com Redes Neurais Artificiais

The comparison of two flood forecasting methodologies, using models with few parameters to calibrate is presented. In DPFT method (First Differences of the Transfer Function) one uses a multi-events rainfall-runoff set and an iterative algorithm for the simultaneous identification of the average Unit Hydrograph (Transfer Function) of the watershed and of a series of effective precipitations associated to each event. This last particularity allows to the calibration and comparison of different Loss Function models, relating observed precipitations to calculated effective precipitations identified by the methodology. In this case a simple Loss Function model is used, the GR3 model. The DPFT methodology is compared with a simple Artificial Neural Network ( ANN ) model approach, where the non-linearities inerent to the rainfall-runoff process are represented by a MLP structure with 4 neurons concerning to the discharges and the precipitations. The results of application and comparison of the methods in two medium sized watersheds are presented and analysed. In this study case, the DPFT methodology showed a slightly better fitting than the ANN model, despite more irregular results. Keywords: methodology DPFT, artificial neural networks