Deep Learning Methods In Terms Research Articles

Detection of COVID-19, pneumonia, and tuberculosis from radiographs using AI-driven knowledge distillation

Chest radiography is an essential diagnostic tool for respiratory diseases such as COVID-19, pneumonia, and tuberculosis because it accurately depicts the structures of the chest. However, accurate detection of these diseases from radiographs is a complex task that requires the availability of medical imaging equipment and trained personnel. Conventional deep learning models offer a viable automated solution for this task. However, the high complexity of these models often poses a significant obstacle to their practical deployment within automated medical applications, including mobile apps, web apps, and cloud-based platforms. This study addresses and resolves this dilemma by reducing the complexity of neural networks using knowledge distillation techniques (KDT). The proposed technique trains a neural network on an extensive collection of chest X-ray images and propagates the knowledge to a smaller network capable of real-time detection. To create a comprehensive dataset, we have integrated three popular chest radiograph datasets with chest radiographs for COVID-19, pneumonia, and tuberculosis. Our experiments show that this knowledge distillation approach outperforms conventional deep learning methods in terms of computational complexity and performance for real-time respiratory disease detection. Specifically, our system achieves an impressive average accuracy of 0.97, precision of 0.94, and recall of 0.97.

T-S2Inet: Transformer-based sequence-to-image network for accurate nanopore sequence recognition.

Nanopore sequencing is a new macromolecular recognition and perception technology that enables high-throughput sequencing of DNA, RNA, even protein molecules. The sequences generated by nanopore sequencing span a large time frame, and the labor and time costs incurred by traditional analysis methods are substantial. Recently, research on nanopore data analysis using machine learning algorithms has gained unceasing momentum, but there is often a significant gap between traditional and deep learning methods in terms of classification results. To analyze nanopore data using deep learning technologies, measures such as sequence completion and sequence transformation can be employed. However, these technologies do not preserve the local features of the sequences. To address this issue, we propose a sequence-to-image (S2I) module that transforms sequences of unequal length into images. Additionally, we propose the Transformer-based T-S2Inet model to capture the important information and improve the classification accuracy. Quantitative and qualitative analysis shows that the experimental results have an improvement of around 2% in accuracy compared to previous methods. The proposed method is adaptable to other nanopore platforms, such as the Oxford nanopore. It is worth noting that the proposed method not only aims to achieve the most advanced performance, but also provides a general idea for the analysis of nanopore sequences of unequal length. The main program is available at https://github.com/guanxiaoyu11/S2Inet.

Prawn morphometrics and weight estimation from images using deep learning for landmark localization

Accurate morphometric analyses and weight estimation are useful in aquaculture for optimizing feeding, predicting harvest yields, identifying desirable traits for selective breeding, grading processes, and monitoring the health status of production animals. However, the collection of phenotypic data through traditional manual approaches at industrial scales and in real-time is time-consuming, labour-intensive, and prone to errors. Digital imaging of individuals and subsequent training of prediction models using Deep Learning (DL) has the potential to rapidly and accurately acquire phenotypic data from aquaculture species. In this study, we applied a novel DL approach to automate morphometric analysis and weight estimation using the black tiger prawn (Penaeus monodon) as a model crustacean. The DL approach comprises two main components: a feature extraction module that efficiently combines low-level and high-level features using the Kronecker product operation; followed by a landmark localization module that then uses these features to predict the coordinates of key morphological points (landmarks) on the prawn body. Once these landmarks were extracted, weight was estimated using a weight regression module based on the extracted landmarks using a fully connected network. For morphometric analyses, we utilized the detected landmarks to derive five important prawn traits. Principal Component Analysis (PCA) was also used to identify landmark-derived distances, which were found to be highly correlated with shape features such as body length, and width. We evaluated our approach on a large dataset of 8164 images of the Black tiger prawn (Penaeus monodon) collected from Australian farms. Our experimental results demonstrate that the novel DL approach outperforms existing DL methods in terms of accuracy, robustness, and efficiency.

SE-NDEND: A novel symmetric watermarking framework with neural network-based chaotic encryption for Internet of Medical Things

The development of the Internet of Medical Things heavily relies on big data, and data security based on medical images has become a growing concern in society. Digital watermarking serves as a crucial technique for protecting and tracing medical image data copyright, as well as enabling forensic analysis. However, existing deep watermarking methods often neglect the protection of watermarks after extraction, leading to potential copyright disputes. To address this issue, this paper proposes SE-NDEND, a novel symmetric watermarking framework with neural network-based chaotic encryption for the Internet of Medical Things that significantly enhances the effectiveness and security of watermarking while maintaining robustness. Specifically, the SE-NDEND leverages neural networks to simulate chaotic systems and generate chaotic sequences, mitigating the complexity and high cost of implementing chaotic systems using hardware circuits. Moreover, we introduce a new noise layer with Moiré distortion that interacts with the decoder, forming a symmetric network structure that bolsters the robustness of watermarking. Parameters are jointly trained and shared during the training process to counteract potential interference from the noise layer. Experimental results validate the effectiveness of SE-NDEND in enhancing copyright protection, traceability, and forensic capabilities, surpassing existing deep learning methods in terms of visual quality (with PSNR of 45.8492 dB and SSIM of 0.9874), security, and robustness. The proposed framework can find application in protecting medical image data in the Internet of Medical Things.

Multivariate solar power time series forecasting using multilevel data fusion and deep neural networks

Accurate forecasting of regional solar photovoltaic power (SPVP) generation is essential for efficient energy management and planning. Existing approaches have shown the effectiveness of decomposing the time series to model the stochastic variability in SPVP data. However, these approaches have limitations in extracting and exploiting both spatial and temporal information from complex and high-dimensional data from multiple sources with intricate relationships, which can impact the accuracy of predictions. In this paper, we propose a novel approach called multilevel data fusion and neural basis expansion analysis (MF-NBEA) for forecasting aggregated regional-level SPVP generation. MF-NBEA integrates exogenous data at multiple levels, uses supervised and unsupervised encoders to provide compact data representation, and enhances model learning from complex data by incorporating spatial information. It also includes a sequence analyser module based on a neural network decomposition mechanism to learn the variability in data and incorporates a residuals learner module to improve overall predictions. We evaluate MF-NBEA using two real-world datasets and find that it outperforms state-of-the-art deep learning methods in terms of forecast accuracy. Furthermore, MF-NBEA facilitates information fusion and knowledge extraction to provide interpretable predictions regarding trend, seasonality, and residual components. The insights gained from our approach inform decision-making for energy management and planning, and can lead to more efficient and sustainable resource utilisation.

Simultaneous alignment and surface regression using hybrid 2D–3D networks for 3D coherent layer segmentation of retinal OCT images with full and sparse annotations

Layer segmentation is important to quantitative analysis of retinal optical coherence tomography (OCT). Recently, deep learning based methods have been developed to automate this task and yield remarkable performance. However, due to the large spatial gap and potential mismatch between the B-scans of an OCT volume, all of them were based on 2D segmentation of individual B-scans, which may lose the continuity and diagnostic information of the retinal layers in 3D space. Besides, most of these methods required dense annotation of the OCT volumes, which is labor-intensive and expertise-demanding. This work presents a novel framework based on hybrid 2D–3D convolutional neural networks (CNNs) to obtain continuous 3D retinal layer surfaces from OCT volumes, which works well with both full and sparse annotations. The 2D features of individual B-scans are extracted by an encoder consisting of 2D convolutions. These 2D features are then used to produce the alignment displacement vectors and layer segmentation by two 3D decoders coupled via a spatial transformer module. Two losses are proposed to utilize the retinal layers’ natural property of being smooth for B-scan alignment and layer segmentation, respectively, and are the key to the semi-supervised learning with sparse annotation. The entire framework is trained end-to-end. To the best of our knowledge, this is the first work that attempts 3D retinal layer segmentation in volumetric OCT images based on CNNs. Experiments on a synthetic dataset and three public clinical datasets show that our framework can effectively align the B-scans for potential motion correction, and achieves superior performance to state-of-the-art 2D deep learning methods in terms of both layer segmentation accuracy and cross-B-scan 3D continuity in both fully and semi-supervised settings, thus offering more clinical values than previous works.

Skin Disease Classification Using CNN Algorithms


 INTRODUCTION: Dermatological disorders, particularly human skin diseases, have become more common in recent decades. Environmental factors, socioeconomic problems, a lack of a balanced diet, and other variables have all contributed to an increase in skin diseases in recent years. Skin diseases can cause psychological suffering in addition to physical injury, especially in people with scarred or disfigured faces.
 OBJECTIVES: The use of artificial intelligence or computer-based technologies in the detection of face skin disorders has advanced dramatically over time. Even for highly experienced doctors and dermatologists, identifying skin disorders can be tricky since many skin diseases have a visual affinity with the surrounding skin and lesions.
 METHODS: Today, the majority of skincare specialists rely on time-consuming, traditional methods to identify disorders. Even though several research have demonstrated promising results on the picture classification job, few studies compare well-known deep learning models with various metrics for categorizing human skin disorders.
 RESULTS: This study examines and contrasts various skin illnesses in terms of cosmetics and common skin concerns. Our dataset includes over 25000 of the eight most common skin disorders. Convolutional neural networks have shown imaging performance that is comparable to or greater than that of humans. We used 11 different network algorithms to identify the illnesses in the sample and compared the results.
 CONCLUSION: To adjust the format of incoming photographs, we do certain image pre-processing and image scaling for each model. ResNet152 beat other deep learning methods in terms of recall, accuracy, and precision on a test dataset of 1930 images.

An Unsupervised Fundus Image Enhancement Method with Multi-Scale Transformer and Unreferenced Loss

Color fundus images are now widely used in computer-aided analysis systems for ophthalmic diseases. However, fundus imaging can be affected by human, environmental, and equipment factors, which may result in low-quality images. Such quality fundus images will interfere with computer-aided diagnosis. Existing methods for enhancing low-quality fundus images focus more on the overall visualization of the image rather than capturing pathological and structural features at the finer scales of the fundus image sufficiently. In this paper, we design an unsupervised method that integrates a multi-scale feature fusion transformer and an unreferenced loss function. Due to the loss of microscale features caused by unpaired training, we construct the Global Feature Extraction Module (GFEM), a combination of convolution blocks and residual Swin Transformer modules, to achieve the extraction of feature information at different levels while reducing computational costs. To improve the blurring of image details caused by deep unsupervised networks, we define unreferenced loss functions that improve the model’s ability to suppress edge sharpness degradation. In addition, uneven light distribution can also affect image quality, so we use an a priori luminance-based attention mechanism to improve low-quality image illumination unevenness. On the public dataset, we achieve an improvement of 0.88 dB in PSNR and 0.024 in SSIM compared to the state-of-the-art methods. Experiment results show that our method outperforms other deep learning methods in terms of vascular continuity and preservation of fine pathological features. Such a framework may have potential medical applications.

State of charge estimation for lithium-ion batteries based on cross-domain transfer learning with feedback mechanism

When the deep learning model is applied to estimate battery state of charge (SOC), the information inside the training set cannot be leveraged thoroughly, which would cause poor SOC estimation accuracy and robustness on the testing set. To solve the problem, this paper proposes an adaptive convolutional neural network-gated recurrent unit with Kalman filter and feedback mechanism (Fb-Ada-CNN-GRU-KF) for SOC estimation considering distribution difference of data segments inside the training set through transfer learning and extracting the spatial information through convolutional layer. Furthermore, the feedback mechanism provides the model more information to learn to correct the systematic error, and the KF in the proposed model works as a post data processor to obtain a steady and smooth SOC estimation results. Experimental and comparison results show that the proposed model for SOC estimation outperforms the existing deep learning methods in terms of the accuracy, generalization and stability.

Forecasting ENSO using convolutional LSTM network with improved attention mechanism and models recombined by genetic algorithm in CMIP5/6

El Niño-Southern Oscillation (ENSO) has a profound impact on global climate, and the ability to forecast it effectively over the long term is essential. In recent years, deep learning methods have demonstrated superior prediction outcomes compared to conventional numerical models. However, due to limited observational data, most of these deep learning methods learn information from simulation data derived from physical models, and ensuring the quality of such simulation data can prove challenging. As a result, the models in CMIP5/6 were recombined using genetic algorithms (GAs) to create our training dataset. A deep learning model was then used to learn features from the output of these combined CMIP models so that these physical numerical models can complement each other. To address the issue of inadequate spatiotemporal feature extraction present in many deep learning methods, we devised an ENSO deep learning regression model. An improved self-attention mechanism was introduced into the convolutional LSTM (long short-term memory) network to enable our model to better extract local and global spatiotemporal features. With our method and model, we were able to obtain less erroneous forecast results, and our method surpassed other state-of-the-art deep learning methods in terms of its capability for long-term forecasting.

UAV Abnormal State Detection Model Based on Timestamp Slice and Multi-Separable CNN

With the rapid development of UAVs (Unmanned Aerial Vehicles), abnormal state detection has become a critical technology to ensure the flight safety of UAVs. The position and orientation system (POS) data, etc., used to evaluate UAV flight status are from different sensors. The traditional abnormal state detection model ignores the difference of POS data in the frequency domain during feature learning, which leads to the loss of key feature information and limits the further improvement of detection performance. To deal with this and improve UAV flight safety, this paper presents a method for detecting the abnormal state of a UAV based on a timestamp slice and multi-separable convolutional neural network (TS-MSCNN). Firstly, TS-MSCNN divides the POS data reasonably in the time domain by setting a set of specific timestamps and then extracts and fuses the key features to avoid the loss of feature information. Secondly, TS-MSCNN converts these feature data into grayscale images by data reconstruction. Lastly, TS-MSCNN utilizes a multi-separable convolution neural network (MSCNN) to learn key features more effectively. The binary and multi-classification experiments conducted on the real flight data, Air Lab Fault and Anomaly (ALFA), demonstrate that the TS-MSCNN outperforms traditional machine learning (ML) and the latest deep learning methods in terms of accuracy.

DSR-GCN: Differentiated-Scale Restricted Graph Convolutional Network for Few-Shot Hyperspectral Image Classification

Graph convolution networks (GCNs) have shown great potentials for few-shot hyperspectral image (HSI) classification. Mainstream GCNs construct graph according to single scale segmentation, which usually ignores subtle adjacency relation between small regions, leading to unreliable initial local graph. To overcome the above issue, we propose a differentiated-scale restricted GCN (DSR-GCN) for HSI classification. Firstly, we propose a differentiated-scale graph construction method considering both the subtle and relative wider range spectral-spatial relation. Secondly, restricted fusion loss is designed to restrict the fusion of features extracted with differentiated-scale GCN branches. Finally, we design a lightweight spatial-spectral siamese network to remedy local pixel-level features. The proposed DSR-GCN can better model spatial structure with a reliable and refined graph, and it can capture more discriminate features in few-shot learning (FSL) scenario. Extensive experiments conducted on four benchmark data sets demonstrate that DSR-GCN outperforms the other deep learning methods in terms of classification accuracy and generalization performance, with the improvements in terms of OA around 6.20%~23.41% (Indian Pines), 4.45%~ 16.48% (University of Pavia), 4.25%~11.85% (Salinas), and 2.0%~17.23% (University of Houston) under 5 labeled samples per class.

Systematic Review of Deep Learning Techniques for Lung Cancer Detection

Cancer is the leading cause of deaths across the globe and 10 million people died of cancer and particularly 2.21 million new cases registered besides 1.80 million deaths, according to WHO, in 2020. Malignant cancer is caused by multiplication and growth of lung cells. In this context, exploiting technological innovations for automatic detection of lung cancer early is to be given paramount importance. Towards this end significant progress has been made and deep learning model such as Convolutional Neural Network (CNN) is found superior in processing lung CT or MRI images for disease diagnosis. Lung cancer detection in the early stages of the disease helps in better treatment and cure of the disease. In this paper, we made a systematic review of deep learning methods for detection of lung cancer. It reviews peer reviewed journal papers and conferences from 2012 to 2021. Literature review throws light on synthesis of different existing methods covering machine learning (ML), deep learning and artificial intelligence (AI). It provides insights of different deep learning methods in terms of their pros and cons and arrives at possible research gaps. This paper gives knowledge to the reader on different aspects of lung cancer detection which can trigger further research possibilities to realize models that can be used in Clinical Decision Support Systems (CDSSs) required by healthcare units.

Freeway Traffic Speed Prediction under the Intelligent Driving Environment: A Deep Learning Approach

The intelligent transportation system (ITS) has been proven capable of effectively addressing traffic congestion issues. For vehicles to perform effectively and improve mobility under the intelligent driving environment, real-time prediction of traffic speed is undoubtedly essential. Considering the complex spatiotemporal dependency inherent in traffic data, conventional prediction models encounter many limitations. To improve the prediction performance and investigate the temporal features, this study focuses on emerging deep neural networks (DNNs) using the Caltrans Performance Measurement System (PeMS) data. This research also establishes an intelligent driving environment in the simulation and compares the traditional car-following model with deep learning methods in terms of multiple performance metrics. The results indicate that both supervised learning and unsupervised learning are superior to the simulation-based model on the freeway, and the two deep learning networks are almost identical to one another. Besides, the result reveals that all models have their latent features for different time dimensions under the low traffic loads, transition states, and heavy traffic loads. This is critical in the application of prediction technologies in ITS. The findings can assist transportation researchers and traffic engineers in both traffic operation and management, such as bottleneck identification, platooning control, and route planning.

Transfer Learning for Signal Detection in Wireless Networks

The last decade has witnessed the rapid growth of deep learning (DL) applications in wireless communications, especially for channel estimation and signal detection. However, conventional DL techniques are usually trained for a specific scenario. Therefore, it is required to be retrained when applying for new wireless systems. Unfortunately, acquiring enough training data for good detection performance may be costly and time-consuming. Even if the training data is sufficient, conventional DL techniques usually require a long training time. Consequently, this can significantly impact the performance of DL techniques and hinder their robustness. To address all these limitations, this letter introduces a transfer learning framework that can transfer knowledge from a source system to improve the training process of target systems with limited training data. Our simulation results demonstrate that the proposed solution outperforms the conventional deep learning methods in terms of channel estimation and signal detection.

A novel multi-atlas segmentation approach under the semi-supervised learning framework: Application to knee cartilage segmentation

Background and objective: Multi-atlas based segmentation techniques, which rely on an atlas library comprised of training images labeled by an expert, have proven their effectiveness in multiple automatic segmentation applications. However, the usage of exhaustive patch libraries combined with the voxel-wise labeling incur a large computational cost in terms of memory requirements and execution times. Methods: To confront this shortcoming, we propose a novel two-stage multi-atlas approach designed under the Semi-Supervised Learning (SSL) framework. The main properties of our method are as follows: First, instead of the voxel-wise labeling approach, the labeling of target voxels is accomplished here by exploiting the spectral content of globally sampled datasets from the target image, along with their spatially correspondent data collected from the atlases. Following SSL, voxels classification is boosted by incorporating unlabeled data from the target image, in addition to the labeled ones from atlas library. Our scheme integrates constructively fruitful concepts, including sparse reconstructions of voxels from linear neighborhoods, HOG feature descriptors of patches/regions, and label propagation via sparse graph constructions. Segmentation of the target image is carried out in two stages: stage-1 focuses on the sampling and labeling of global data, while stage-2 undertakes the above tasks for the out-of-sample data. Finally, we propose different graph-based methods for the labeling of global data, while these methods are extended to deal with the out-of-sample voxels. Results: A thorough experimental investigation is conducted on 76 subjects provided by the publicly accessible Osteoarthritis Initiative (OAI) repository. Comparative results and statistical analysis demonstrate that the suggested methodology exhibits superior segmentation performance compared to the existing patch-based methods, across all evaluation metrics (DSC:88.89%, Precision: 89.86%, Recall: 88.12%), while at the same time it requires a considerably reduced computational load (>70% reduction on average execution time with respect to other patch-based). In addition, our approach is favorably compared against other non patch-based and deep learning methods in terms of performance accuracy (on the 3-class problem). A final experimentation on a 5-class setting of the problems demonstrates that our approach is capable of achieving performance comparable to existing state-of-the-art knee cartilage segmentation methods (DSC:88.22% and DSC:85.84% for femoral and tibial cartilage respectively).

Multi-Input Attention Network for Dehazing of Remote Sensing Images

The non-uniform haze distribution in remote sensing images, together with the complexity of the ground information, brings many difficulties to the dehazing of remote sensing images. In this paper, we propose a multi-input convolutional neural network based on an encoder–decoder structure to effectively restore remote sensing hazy images. The proposed network can directly learn the mapping between hazy images and the corresponding haze-free images. It also effectively utilizes the strong haze penetration characteristic of the Infrared band. Our proposed network also includes the attention module and the global skip connection structure, which enables the network to effectively learn the haze-relevant features and better preserve the ground information. We build a dataset for training and testing our proposed method. The dataset consists of remote sensing images with two different resolutions and nine bands, which are captured by Sentinel-2. The experimental results demonstrate that our method outperforms traditional dehazing methods and other deep learning methods in terms of the final dehazing effect, peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and feature similarity (FSIM).

Classification of Breast Cancer Histopathological Images Using DenseNet and Transfer Learning.

Breast cancer is one of the most common invading cancers in women. Analyzing breast cancer is nontrivial and may lead to disagreements among experts. Although deep learning methods achieved an excellent performance in classification tasks including breast cancer histopathological images, the existing state-of-the-art methods are computationally expensive and may overfit due to extracting features from in-distribution images. In this paper, our contribution is mainly twofold. First, we perform a short survey on deep-learning-based models for classifying histopathological images to investigate the most popular and optimized training-testing ratios. Our findings reveal that the most popular training-testing ratio for histopathological image classification is 70%: 30%, whereas the best performance (e.g., accuracy) is achieved by using the training-testing ratio of 80%: 20% on an identical dataset. Second, we propose a method named DenTnet to classify breast cancer histopathological images chiefly. DenTnet utilizes the principle of transfer learning to solve the problem of extracting features from the same distribution using DenseNet as a backbone model. The proposed DenTnet method is shown to be superior in comparison to a number of leading deep learning methods in terms of detection accuracy (up to 99.28% on BreaKHis dataset deeming training-testing ratio of 80%: 20%) with good generalization ability and computational speed. The limitation of existing methods including the requirement of high computation and utilization of the same feature distribution is mitigated by dint of the DenTnet.

A new data augmentation method for EEG features based on the hybrid model of broad-deep networks

Decades after data augmentation was first proposed in brain-computer interface (BCI), the authenticity and performance still do not meet rational requirements, which is directly related to the fact that the augmentation methods do not provide real electroencephalograph (EEG) trials. Here we show how to generate a numerous authentic EEG from the original calibrated EEG by using a novel hybrid model of broad-deep networks, which eliminates the lack of authenticity of data generated by GAN and other methods. First, we design an EEG evoked experiment with a complex boundary avoidance task to collect the EEG of different subjects. This experiment can effectively highlight the differences of EEG features of different subjects that makes the results more reliable when using our novel hybrid model to measure the similarity. A new hybrid model of broad-deep networks is proposed to measure the similarity of different subjects in this study. And the EEG features of the two subjects with the highest similarity are combined to generate an augmented feature set. On the condition of satisfying the authenticity of EEG, the augmented feature set is significantly better than the original feature set in data dimension and quality. Finally, we verify the classification effect of the augmented feature set, and the results show that the proposed method can effectively generate real EEG data and improve the classification performance to a high reliability level for complex boundary avoidance tasks under limited EEG conditions. In addition, we observe the obvious advantages of this model over traditional deep learning methods in terms of training time and memory overhead.

Image super-resolution reconstruction algorithm based on significant network connection-collaborative migration structure

Although deep learning is widely used in the field of image super-resolution reconstruction, the number of network layers and the complexity of deep learning models continue to increase, and the speed at which the difficulty of model training increases has exceeded the speed of network performance improvement. This situation prevents deep learning frameworks from fully utilizing their generalization ability in image super-resolution reconstruction. Moreover, the existing image reconstruction methods have the problem of partial information loss. Therefore, this paper first proposes a structure that simulates the feature extraction function in the visual attention mechanism in the convolutional neural network. We call this a significant network connection. The feature information extracted by the network architecture is more significant, while other information is less significant. This setup has a reduced impact on the resulting image reconstruction effect. Then, a network architecture focusing on collaborative information migration is proposed. This architecture can obtain the implicit domain of the intermediate state of the image domain to be reconstructed, and it can make the two networks use the learned hidden domain during the reconstruction process. The dual networks trained in this way are more symmetrical. This approach can better maintain the common feature information of the reconstructed image and effectively solve the problem of partial feature information loss in the image. The experimental results show that the texture, artificial effects and noise of the reconstructed image obtained by the method proposed in this paper are significantly improved over those of the images produced by other mainstream methods. In addition, the method proposed in this paper exhibits a certain degree of improvement over other deep learning methods in terms of model training speed and feature information retention.