Search results

This study aims to construct a sentiment series generation method for danmu comments based on deep learning, and explore the features of sentiment series after clustering.

This study consisted of two main parts: danmu comment sentiment series generation and clustering. In the first part, the authors proposed a sentiment classification model based on BERT fine-tuning to quantify danmu comment sentiment polarity. To smooth the sentiment series, they used methods, such as comprehensive weights. In the second part, the shaped-based distance (SBD)-K-shape method was used to cluster the actual collected data.

The filtered sentiment series or curves of the microfilms on the Bilibili website could be divided into four major categories. There is an apparently stable time interval for the first three types of sentiment curves, while the fourth type of sentiment curve shows a clear trend of fluctuation in general. In addition, it was found that “disputed points” or “highlights” are likely to appear at the beginning and the climax of films, resulting in significant changes in the sentiment curves. The clustering results show a significant difference in user participation, with the second type prevailing over others.

Their sentiment classification model based on BERT fine-tuning outperformed the traditional sentiment lexicon method, which provides a reference for using deep learning as well as transfer learning for danmu comment sentiment analysis. The BERT fine-tuning–SBD-K-shape algorithm can weaken the effect of non-regular noise and temporal phase shift of danmu text.

To improve the position tracking efficiency of the upper-limb rehabilitation robot for stroke hemiplegia patients, the optimization Learning rate of the membership function based on the fuzzy impedance controller of the rehabilitation robot is propose.

First, the impaired limb’s damping and stiffness parameters for evaluating its physical recovery condition are online estimated by using weighted least squares method based on recursive algorithm. Second, the fuzzy impedance control with the rule has been designed with the optimal impedance parameters. Finally, the membership function learning rate online optimization strategy based on Takagi-Sugeno (TS) fuzzy impedance model was proposed to improve the position tracking speed of fuzzy impedance control.

This method provides a solution for improving the membership function learning rate of the fuzzy impedance controller of the upper limb rehabilitation robot. Compared with traditional TS fuzzy impedance controller in position control, the improved TS fuzzy impedance controller has reduced the overshoot stability time by 0.025 s, and the position error caused by simulating the thrust interference of the impaired limb has been reduced by 8.4%. This fact is verified by simulation and test.

The TS fuzzy impedance controller based on membership function online optimization learning strategy can effectively optimize control parameters and improve the position tracking speed of upper limb rehabilitation robots. This controller improves the auxiliary rehabilitation efficiency of the upper limb rehabilitation robot and ensures the stability of auxiliary rehabilitation training.

An abundance of techniques has been presented so forth for waste classification but, they deliver inefficient results with low accuracy. Their achievement on various repositories is different and also, there is insufficiency of high-scale databases for training. The purpose of the study is to provide high security.

In this research, optimization-assisted federated learning (FL) is introduced for thermoplastic waste segregation and classification. The deep learning (DL) network trained by Archimedes Henry gas solubility optimization (AHGSO) is used for the classification of plastic and resin types. The deep quantum neural networks (DQNN) is used for first-level classification and the deep max-out network (DMN) is employed for second-level classification. This developed AHGSO is obtained by blending the features of Archimedes optimization algorithm (AOA) and Henry gas solubility optimization (HGSO). The entities included in this approach are nodes and servers. Local training is carried out depending on local data and updations to the server are performed. Then, the model is aggregated at the server. Thereafter, each node downloads the global model and the update training is executed depending on the downloaded global and the local model till it achieves the satisfied condition. Finally, local update and aggregation at the server is altered based on the average method. The Data tag suite (DATS_2022) dataset is used for multilevel thermoplastic waste segregation and classification.

By using the DQNN in first-level classification the designed optimization-assisted FL has gained an accuracy of 0.930, mean average precision (MAP) of 0.933, false positive rate (FPR) of 0.213, loss function of 0.211, mean square error (MSE) of 0.328 and root mean square error (RMSE) of 0.572. In the second level classification, by using DMN the accuracy, MAP, FPR, loss function, MSE and RMSE are 0.932, 0.935, 0.093, 0.068, 0.303 and 0.551.

The multilevel thermoplastic waste segregation and classification using the proposed model is accurate and improves the effectiveness of the classification.

Classification of remote sensing images (RSI) is a challenging task in computer vision. Recently, researchers have proposed a variety of creative methods for automatic recognition of RSI, and feature fusion is a research hotspot for its great potential to boost performance. However, RSI has a unique imaging condition and cluttered scenes with complicated backgrounds. This larger difference from nature images has made the previous feature fusion methods present insignificant performance improvements.

This work proposed a two-convolutional neural network (CNN) fusion method named main and branch CNN fusion network (MBC-Net) as an improved solution for classifying RSI. In detail, the MBC-Net employs an EfficientNet-B3 as its main CNN stream and an EfficientNet-B0 as a branch, named MC-B3 and BC-B0, respectively. In particular, MBC-Net includes a long-range derivation (LRD) module, which is specially designed to learn the dependence of different features. Meanwhile, MBC-Net also uses some unique ideas to tackle the problems coming from the two-CNN fusion and the inherent nature of RSI.

Extensive experiments on three RSI sets prove that MBC-Net outperforms the other 38 state-of-the-art (STOA) methods published from 2020 to 2023, with a noticeable increase in overall accuracy (OA) values. MBC-Net not only presents a 0.7% increased OA value on the most confusing NWPU set but also has 62% fewer parameters compared to the leading approach that ranks first in the literature.

MBC-Net is a more effective and efficient feature fusion approach compared to other STOA methods in the literature. Given the visualizations of grad class activation mapping (Grad-CAM), it reveals that MBC-Net can learn the long-range dependence of features that a single CNN cannot. Based on the tendency stochastic neighbor embedding (t-SNE) results, it demonstrates that the feature representation of MBC-Net is more effective than other methods. In addition, the ablation tests indicate that MBC-Net is effective and efficient for fusing features from two CNNs.

The paper aims to tackle the classification of Remote Sensing Images (RSIs), which presents a significant challenge for computer algorithms due to the inherent characteristics of clustered ground objects and noisy backgrounds. Recent research typically leverages larger volume models to achieve advanced performance. However, the operating environments of remote sensing commonly cannot provide unconstrained computational and storage resources. It requires lightweight algorithms with exceptional generalization capabilities.

This study introduces an efficient knowledge distillation (KD) method to build a lightweight yet precise convolutional neural network (CNN) classifier. This method also aims to substantially decrease the training time expenses commonly linked with traditional KD techniques. This approach entails extensive alterations to both the model training framework and the distillation process, each tailored to the unique characteristics of RSIs. In particular, this study establishes a robust ensemble teacher by independently training two CNN models using a customized, efficient training algorithm. Following this, this study modifies a KD loss function to mitigate the suppression of non-target category predictions, which are essential for capturing the inter- and intra-similarity of RSIs.

This study validated the student model, termed KD-enhanced network (KDE-Net), obtained through the KD process on three benchmark RSI data sets. The KDE-Net surpasses 42 other state-of-the-art methods in the literature published from 2020 to 2023. Compared to the top-ranked method’s performance on the challenging NWPU45 data set, KDE-Net demonstrated a noticeable 0.4% increase in overall accuracy with a significant 88% reduction in parameters. Meanwhile, this study’s reformed KD framework significantly enhances the knowledge transfer speed by at least three times.

This study illustrates that the logit-based KD technique can effectively develop lightweight CNN classifiers for RSI classification without substantial sacrifices in computation and storage costs. Compared to neural architecture search or other methods aiming to provide lightweight solutions, this study’s KDE-Net, based on the inherent characteristics of RSIs, is currently more efficient in constructing accurate yet lightweight classifiers for RSI classification.

Web users rely heavily on online content make decisions without assessing the veracity of the content. The online content comprising text, image, video or audio may be tampered with to influence public opinion. Since the consumers of online information (misinformation) tend to trust the content when the image(s) supplement the text, image manipulation software is increasingly being used to forge the images. To address the crucial problem of image manipulation, this study focusses on developing a deep-learning-based image forgery detection framework.

The proposed deep-learning-based framework aims to detect images forged using copy-move and splicing techniques. The image transformation technique aids the identification of relevant features for the network to train effectively. After that, the pre-trained customized convolutional neural network is used to train on the public benchmark datasets, and the performance is evaluated on the test dataset using various parameters.

The comparative analysis of image transformation techniques and experiments conducted on benchmark datasets from a variety of socio-cultural domains establishes the effectiveness and viability of the proposed framework. These findings affirm the potential applicability of proposed framework in real-time image forgery detection.

This study bears implications for several important aspects of research on image forgery detection. First this research adds to recent discussion on feature extraction and learning for image forgery detection. While prior research on image forgery detection, hand-crafted the features, the proposed solution contributes to stream of literature that automatically learns the features and classify the images. Second, this research contributes to ongoing effort in curtailing the spread of misinformation using images. The extant literature on spread of misinformation has prominently focussed on textual data shared over social media platforms. The study addresses the call for greater emphasis on the development of robust image transformation techniques.

This study carries important practical implications for various domains such as forensic sciences, media and journalism where image data is increasingly being used to make inferences. The integration of image forgery detection tools can be helpful in determining the credibility of the article or post before it is shared over the Internet. The content shared over the Internet by the users has become an important component of news reporting. The framework proposed in this paper can be further extended and trained on more annotated real-world data so as to function as a tool for fact-checkers.

In the current scenario wherein most of the image forgery detection studies attempt to assess whether the image is real or forged in an offline mode, it is crucial to identify any trending or potential forged image as early as possible. By learning from historical data, the proposed framework can aid in early prediction of forged images to detect the newly emerging forged images even before they occur. In summary, the proposed framework has a potential to mitigate physical spreading and psychological impact of forged images on social media.

This study focusses on copy-move and splicing techniques while integrating transfer learning concepts to classify forged images with high accuracy. The synergistic use of hitherto little explored image transformation techniques and customized convolutional neural network helps design a robust image forgery detection framework. Experiments and findings establish that the proposed framework accurately classifies forged images, thus mitigating the negative socio-cultural spread of misinformation.

Higher energy conversion efficiency of internal combustion engine can be achieved with optimal control of unsteady in-cylinder flow fields inside a direct-injection (DI) engine. However, it remains a daunting task to predict the nonlinear and transient in-cylinder flow motion because they are highly complex which change both in space and time. Recently, machine learning methods have demonstrated great promises to infer relatively simple temporal flow field development. This paper aims to feature a physics-guided machine learning approach to realize high accuracy and generalization prediction for complex swirl-induced flow field motions.

To achieve high-fidelity time-series prediction of unsteady engine flow fields, this work features an automated machine learning framework with the following objectives: (1) The spatiotemporal physical constraint of the flow field structure is transferred to machine learning structure. (2) The ML inputs and targets are efficiently designed that ensure high model convergence with limited sets of experiments. (3) The prediction results are optimized by ensemble learning mechanism within the automated machine learning framework.

The proposed data-driven framework is proven effective in different time periods and different extent of unsteadiness of the flow dynamics, and the predicted flow fields are highly similar to the target field under various complex flow patterns. Among the described framework designs, the utilization of spatial flow field structure is the featured improvement to the time-series flow field prediction process.

The proposed flow field prediction framework could be generalized to different crank angle periods, cycles and swirl ratio conditions, which could greatly promote real-time flow control and reduce experiments on in-cylinder flow field measurement and diagnostics.

This paper aims to review in-memory computing (IMC) for machine learning (ML) applications from history, architectures and options aspects. In this review, the authors investigate different architectural aspects and collect and provide our comparative evaluations.

Collecting over 40 IMC papers related to hardware design and optimization techniques of recent years, then classify them into three optimization option categories: optimization through graphic processing unit (GPU), optimization through reduced precision and optimization through hardware accelerator. Then, the authors brief those techniques in aspects such as what kind of data set it applied, how it is designed and what is the contribution of this design.

ML algorithms are potent tools accommodated on IMC architecture. Although general-purpose hardware (central processing units and GPUs) can supply explicit solutions, their energy efficiencies have limitations because of their excessive flexibility support. On the other hand, hardware accelerators (field programmable gate arrays and application-specific integrated circuits) win on the energy efficiency aspect, but individual accelerator often adapts exclusively to ax single ML approach (family). From a long hardware evolution perspective, hardware/software collaboration heterogeneity design from hybrid platforms is an option for the researcher.

IMC’s optimization enables high-speed processing, increases performance and analyzes massive volumes of data in real-time. This work reviews IMC and its evolution. Then, the authors categorize three optimization paths for the IMC architecture to improve performance metrics.

Online news websites provide huge amounts of timely news, bringing the challenge of recommending personalized news articles. Generative adversarial networks (GAN) based on collaborative filtering (CFGAN) can achieve effective recommendation quality. However, CFGAN ignores item contents, which contain more latent preference features than just user ratings. It is important to consider both ratings and item contents in making preference predictions. This study aims to improve news recommendation by proposing a GAN-based news recommendation model considering both ratings (implicit feedback) and the latent features of news content.

The collaborative topic modeling (CTM) can improve user preference prediction by combining matrix factorization (MF) with latent topics of item content derived from latent topic modeling. This study proposes a novel hybrid news recommendation model, Hybrid-CFGAN, which modifies the architecture of the CFGAN model with enhanced preference learning from the CTM. The proposed Hybrid-CFGAN model contains parallel neural networks – original rating-based preference learning and CTM-based preference learning, which consider both ratings and news content with user preferences derived from the CTM model. A tunable parameter is used to adjust the weights of the two preference learnings, while concatenating the preference outputs of the two parallel neural networks.

This study uses the dataset collected from an online news website, NiusNews, to conduct an experimental evaluation. The results show that the proposed Hybrid-CFGAN model can achieve better performance than the state-of-the-art GAN-based recommendation methods. The proposed novel Hybrid-CFGAN model can enhance existing GAN-based recommendation and increase the performance of preference predictions on textual content such as news articles.

As the existing CFGAN model does not consider content information and solely relies on history logs, it may not be effective in recommending news articles. Our proposed Hybrid-CFGAN model modified the architecture of the CFGAN generator by adding a parallel neural network to gain the relevant information from news content and user preferences derived from the CTM model. The novel idea of adjusting the preference learning from two parallel neural networks – original rating-based preference learning and CTM-based preference learning – contributes to improve the recommendation quality of the proposed model by considering both ratings and latent preferences derived from item contents. The proposed novel recommendation model can improve news recommendation, thereby increasing the commercial value of news media platforms.

The Chinese housing market has witnessed rapid growth during the past decade and the significance of housing price forecasting has undoubtedly elevated, becoming an important issue to investors and policymakers. This study aims to examine neural networks (NNs) for office property price index forecasting from 10 major Chinese cities for July 2005–April 2021.

The authors aim at building simple and accurate NNs to contribute to pure technical forecasts of the Chinese office property market. To facilitate the analysis, the authors explore different model settings over algorithms, delays, hidden neurons and data-spitting ratios.

The authors reach a simple NN with three delays and three hidden neurons, which leads to stable performance of about 1.45% average relative root mean square error across the 10 cities for the training, validation and testing phases.

The results could be used on a standalone basis or combined with fundamental forecasts to form perspectives of office property price trends and conduct policy analysis.