Input Sequence Research Articles

Follower attraction in live streaming: Knowledge driven by PKM and data driven by EF-LSTM

In this study, we examine variables affecting follower attraction and determine the optimal model for its prediction. We identify 12 variables from the persuasion attempt perspective in the persuasion knowledge model and accordingly perform two progressive studies (study 1: knowledge driven, with coarse-grained variables, with unimodal models, without data fusion; study 2: knowledge-and-data driven, with fine-grained features, with multimodal models, with data fusion). In study 1, we compute 12 variables in accordance with hypotheses from nonfused trimodal data (verbal, vocal, and visual). XGBoost and Shapley additive explanations are used to evaluate the importance of these 12 variables, and economic models are adopted to test their statistical significance. In study 2, we reduce the data granularity of the 12 coarse-grained variables by generating 323 fine-grained features. We then process trimodal data fusion to consider intra- and intermodal interactions. Long Short-Term Memory (LSTM) models are applied to learn a mapping from the input sequences of trimodality to output values of follower increment. The results demonstrate that early fusion LSTM with fused trimodal features has a 66.3% prediction accuracy, outperforming LSTM with unimodal features and late fusion LSTM with nonfused trimodal features. To our knowledge, these results are among the first to unfold follower attraction strategies with unstructured video data, and this is the first study to provide knowledge-and-data-driven multimodal machine learning models for predicting follower attraction at different granularity levels.

Fine-grained hate speech detection in Arabic using transformer-based models

With the proliferation of social media platforms, characterized by features such as anonymity, user-friendly access, and the facilitation of online community building and discourse, the matter of detecting and monitoring hate speech has emerged as an increasingly formidable challenge for society, individuals, and researchers. Despite the crucial importance of hate speech detection task, the majority of work in this field has been conducted in English, with insufficient focus on other languages, particularly Arabic. Furthermore, most existing studies on Arabic hate speech detection have addressed this task as a binary classification problem, which is unreliable. Therefore, the aim of this study is to provide an enhanced model for detecting fine-grained hate speech in Arabic. To this end, three transformer-based models were evaluated to generate contextualized word embeddings from input sequence. Additionally, these models were combined with a bidirectional gated recurrent unit (BiGRU) layer to further improve the extracted semantic and context features. The experiments were conducted on an Arabic reference dataset provided by the open-source Arabic corpora and processing tools (OSACT-5) shared task. A comparative analysis indicates the efficiency of the proposed model over the baseline and related work models by achieving a macro F1-score of 61.68%.

Enhanced Compression of k-Mer Sets with Counters via de Bruijn Graphs.

An essential task in computational genomics involves transforming input sequences into their constituent k-mers. The quest for an efficient representation of k-mer sets is crucial for enhancing the scalability of bioinformatic analyses. One widely used method involves converting the k-mer set into a de Bruijn graph (dBG), followed by seeking a compact graph representation via the smallest path cover. This study introduces USTAR* (Unitig STitch Advanced constRuction), a tool designed to compress both a set of k-mers and their associated counts. USTAR leverages the connectivity and density of dBGs, enabling a more efficient path selection for constructing the path cover. The efficacy of USTAR is demonstrated through its application in compressing real read data sets. USTAR improves the compression achieved by UST (Unitig STitch), the best algorithm, by percentages ranging from 2.3% to 26.4%, depending on the k-mer size, and it is up to times faster.

Long-term rolling prediction of transformer power load capacity based on the informer model

Abstract There are currently no methods capable of long-term forecasting based on very long-term real-world data. Any false prediction may damage the electrical transformer. For this problem, a transformer power load based on the Informer model is used as a method of long-term rolling forecasting. This method uses the self-attention distillation mechanism in the Informer model to allow the decoder of each layer to shorten the length of the input sequence by half, thus greatly saving the encoder memory overhead, and this paper adds a rolling long-term prediction function, making the prediction decoding time extremely short. Taking the power load data of a certain province in China as a test case, the improved model Informer* was compared with the traditional Informer model. The results show that the prediction accuracy of the Informer* model is higher and the load prediction accuracy is effectively improved.

A Novel Source Code Representation Approach Based on Multi-Head Attention

Code classification and code clone detection are crucial for understanding and maintaining large software systems. Although deep learning surpasses traditional techniques in capturing the features of source code, existing models suffer from low processing power and high complexity. We propose a novel source code representation method based on the multi-head attention mechanism (SCRMHA). SCRMHA captures the vector representation of entire code segments, enabling it to focus on different positions of the input sequence, capture richer semantic information, and simultaneously process different aspects and relationships of the sequence. Moreover, it can calculate multiple attention heads in parallel, speeding up the computational process. We evaluate SCRMHA on both the standard dataset and an actual industrial dataset, and analyze the differences between these two datasets. Experiment results in code classification and clone detection tasks show that SCRMHA consumes less time and reduces complexity by about one-third compared with traditional source code feature representation methods. The results demonstrate that SCRMHA reduces the computational complexity and time consumption of the model while maintaining accuracy.

Intelligent forecasting temperature measurements of solar PV cells using modified recurrent neural network

For microgrids to operate optimally and minimize the effects of uncertainty, anticipating solar PV measurements is essential. For residential and commercial microgrids that use solar PV, the predicting of solar energy over a short period is crucial for managing grid-connected PVeffectively. Therefore, this work develops a Recurrent Neural Network (RNN) for forecasting temperature measurements as time series records, where a combination of long short-term memory (LSTM) architecture with RNN is used to process input measurements by updating the RNN state and winding over time degrees. Data from the entire prior time steps is stored in the RNN state. A dataset of temperature waveform measurements is used, which includes 2000 unnaturally produced signals of three channels with varying length. An LSTM neural network can be used to expect future values of a time series or sequence utilizing data from earlier time steps as input. Training of a regression LSTM neural network through the output of a sequence is performed, where the goals are the training sequence with records shifting one-time step, for training theLSTM neural architecture with time series forecasting. In other words, the weights of the LSTM neural structure learn to predict the following time step values of the input sequence at every time step. By considering the past forecasts as inputs, the closed-loop prediction forecasts the next time steps of sequences. The model makes the forecast without using the true data. The cross-entropy loss serves as the loss function. It is found that the mean RMSE overall test observations were about 0.5080 which promises to make better predictions from learning the temporal context of input sequences

A Fresh Look at Tomato Leaf Disease Recognition using Vision Transformers

Tomatoes is one of the major economically significant vegetables produced worldwide, contributing greatly to increased agricultural production and food security. However, tomato plants are unfortunately prone to a number of diseases, including several that target the leaves, which can significantly reduce crop productivity and quality. Recently, deep learning techniques have revolutionized the fields of computer vision and image analysis. By automatically learning hierarchical representations from raw pixel data. Transformer is a new deep learning technique that opens new possibilities for image understanding using self-attention mechanisms to capture global dependencies within input sequences. This approach is exemplified by the Vision Transformer (ViT). In this study, we utilize and evaluate the effectiveness of six variations of the Vision Transformer (ViT) architecture in the task of tomato leaf disease recognition. The variants include Mobile ViT, EANet, Swin ViT, ViT, Shift ViT, and Compact ViT. Utilizing a publicly available, multiple-source dataset of tomato leaf images containing various disease patterns. Performance for all models was evaluated and compared in classifying various types of tomato leaf diseases in terms of accuracy, loss, precision, recall, and F1-Score, and the results showed that. The CompactVit has achieved the best accuracy of 97% and precision of 97% and 96% for recall. While the mobile ViT has the lowest performance among all variations in tomato disease recognition, overall, ViT is showing its promise, and it can be utilized on a large scale for smart agriculture, which opens the door for further exploration of this area.

A closed-loop human-computer interactive design method based on sequential human intention prediction and knowledge recommendation

Designers working on large-scale complex engineering projects may encounter design errors due to limited experience and expertise. Therefore, during the design process, accurately identifying the designer’s design intention and recommending relevant design knowledge can enhance design efficiency and reduce errors. In this paper, we propose a closed-loop human–computer interactive design method based on sequential human intention prediction and knowledge recommendation. This method leverages the Function-Behaviour-Structure (FBS) model to reduce the dimensionality of design action sequences, so as to facilitate the analysis of potential design patterns. The processed action sequences are used to train Transformer to predict design intentions. In different input sequences, Transformer achieved a highest prediction accuracy of 92.09%. We construct a Design Knowledge Recommendation Framework (DKRF) and its corresponding design knowledge matching algorithm. This framework accurately recommends design knowledge to designers, solving design stagnation and improving design efficiency. Finally, a case study of four types of mechanical model design is conducted to demonstrate the feasibility and wide applicability of the proposed closed-loop human–computer interactive design method.

Chemical Gas Source Localization with Synthetic Time Series Diffusion Data Using Video Vision Transformer

Gas source localization is vital in emergency scenarios to enable swift and effective responses. In this study, we introduce a gas source localization model leveraging the video vision transformer (ViViT). Utilizing synthetic time series diffusion data, the source grid is predicted by classifying the grid with the highest probability of gas occurrence within the diffusion data coverage. Through extensive experimentation using the NBC-RAMS simulator, we generate large datasets of gas diffusion under varied experimental conditions and meteorological environments, enabling comprehensive model training and evaluation. Our findings demonstrate that the ViViT outperforms other deep learning models in processing time series gas data, showcasing a superior estimation performance. Leveraging a transformer architecture, the ViViT exhibits a robust classification performance even in scenarios influenced by weather conditions or incomplete observations. Furthermore, we conduct an analysis of accuracy and parameter count across various input sequence lengths, revealing the ability of the ViViT to maintain high computational efficiency while achieving accurate source localization. These results underscore the effectiveness of the ViViT as a model for gas source localization, particularly in situations demanding a rapid response in real-world environments, such as gas leaks or attacks.

The longest almost increasing subsequence problem with sliding windows

In this paper, we first define the longest almost increasing subsequence with sliding windows (LaISW), a generalization that combines the longest increasing subsequence with sliding windows (LISW) problem and the longest almost increasing subsequence (LaIS) problem. For a numeric sequence A, a window size w and a tolerance constant c, the goal of the LaISW problem is to identify the LaIS within all windows of size w in A. In an almost increasing subsequence, slight decreases smaller than c are permitted. We propose an efficient algorithm for solving the LaISW problem. Instead of constructing the entire row tower, our algorithm computes the change in drop out (occurrence) for each element in the row tower. The time complexity of our algorithm is O(nL), and the space complexity is O(L), where n and L represent the lengths of the input sequence and the LaISW answer, respectively.

PointTransform Networks for automatic depression level prediction via facial keypoints

According to physiological reports, individuals with different levels of depression present various facial dynamic patterns. Thus, researchers have examined facial changes to predict depression severity. However, they often use facial images or videos of the subjects, which increases the risk of privacy leakage. Therefore, we intend to use the mixer layer to process facial keypoint sequences to predict the depression levels. However, the mixer layer cannot guarantee that the output and input sequences have the same temporal properties, which prevents the residual connection from being additive in a physical sense. To this end, we construct a PointTransform Network (PTN). In the constructed model, a Mixer Attention Layer (MAL) and a Token Sequence Aggregation (TSA) module are proposed. The proposed MAL embeds the results of channel-communication and token-communication into the sequence in the form of weights through the attention mechanism, which maintains the temporal order among tokens in the sequence during the calculation process and compensates for the limitations of the mixer layer. The proposed TSA module can integrate the depression level prediction results of all facial keypoints through the attention mechanism. Therefore, the TSA module achieves the unity of decision-level fusion and tensor vectorization. Experiments are conducted on several benchmark databases and the results demonstrate the effectiveness of our method.

The importance of input sequence set to consensus-derived proteins and their relationship to reconstructed ancestral proteins.

A protein sequence encodes its energy landscape-all the accessible conformations, energetics, and dynamics. The evolutionary relationship between sequence and landscape can be probed phylogenetically by compiling a multiple sequence alignment of homologous sequences and generating common ancestors via Ancestral Sequence Reconstruction or a consensus protein containing the most common amino acid at each position. Both ancestral and consensus proteins are often more stable than their extant homologs-questioning the differences between them and suggesting that both approaches serve as general methods to engineer thermostability. We used the Ribonuclease H family to compare these approaches and evaluate how the evolutionary relationship of the input sequences affects the properties of the resulting consensus protein. While the consensus protein derived from our full Ribonuclease H sequence alignment is structured and active, it neither shows properties of a well-folded protein nor has enhanced stability. In contrast, the consensus protein derived from a phylogenetically-restricted set of sequences is significantly more stable and cooperatively folded, suggesting that cooperativity may be encoded by different mechanisms in separate clades and lost when too many diverse clades are combined to generate a consensus protein. To explore this, we compared pairwise covariance scores using a Potts formalism as well as higher-order sequence correlations using singular value decomposition (SVD). We find the SVD coordinates of a stable consensus sequence are close to coordinates of the analogous ancestor sequence and its descendants, whereas the unstable consensus sequences are outliers in SVD space.

A pyramidal residual attention model of short‐term wind power forecasting for wind farm safety

AbstractWind power fluctuation significantly impacts the safe and stable operation of the wind farm power grid. As the installed capacity of grid‐connected wind power expands to a certain threshold, these fluctuations can detrimentally affect the wind farm's operations. Consequently, wind power prediction emerges as a critical technology for ensuring safe, stable and efficient wind power generation. To optimize power grid dispatching and enhance wind farm operation and maintenance, precise wind power prediction is essential. In this context, we introduce a joint deep learning model that integrates a compact pyramid structure with a residual attention encoder, aiming to bolster wind farm operational safety and reliability. The model employs a compact pyramid architecture to extract multi‐time scale features from the input sequence, facilitating effective information exchange across different scales and enhancing the capture of long‐term sequence dependencies. To mitigate vanishing gradients, the residual transformer encoder is applied, augmenting the original attention mechanism with a global dot product attention pathway. This approach improves the gradient descent process, making it more accessible without introducing additional hyperparameters. The model's efficacy is validated using a dataset from an actual wind farm in China. Experimental outcomes reveal a notable enhancement in wind power prediction accuracy, thereby contributing to the operational safety of wind farms.

AFMF: Time series anomaly detection framework with modified forecasting

Forecasting-based method is one of prevalent unsupervised time series anomaly detection approaches. Currently, large portions of existing forecasting-based methods are devoted to discussing the feature extraction of input sequences and targeting at accurate predictions. In essence, their frameworks and core ideas are identical to the pure forecasting models. However, the distinctiveness of anomalies is affected by not only forecasting accuracy, but also many other factors. This paper summarizes three other dominant factors: (1) Scale disparity; (2) Discrete variate; (3) Input anomaly. They are common and non-negligible in real-world anomaly detection. Moreover, we propose AFMF: a time series Anomaly detection Framework with Modified Forecasting to solve them respectively by its three key components, i.e., Local Instance Normalization, Lopsided Forecasting and Progressive Adjacent Masking. The first two are refined descendants of existing mechanisms while the third component is completely novel. Extensive experiments on ten benchmarks verify that AFMF can be combined with any forecasting or forecasting-based anomaly detection method to achieve SOTA anomaly detection performances. The source code is available at https://github.com/OrigamiSL/AFMF.

Transient surrogate modeling of modally reduced structures with discontinuous loads and damping

In this work, a surrogate model for structural, transient and discontinuously excited finite element method simulations is developed. This allows to reduce the computational effort of repeated calculations of identical models under different load cases. The architecture of the surrogate combines fully connected neural network layers with long short-term memory layers. For the reproduction of different damping ratios, a categorical variable is added to the continuous input data. Based on a recursive flow of the predicted data back to the input layer, long-term dependencies do not vanish due to short-input sequences. The system dimension is reduced by applying the model-order reduction technique for modal decomposition. The high accuracy of the surrogate and the reduction of computational costs are shown on an academic example of a cantilever beam and a real-world example of a robot. The advantages of our approach are illustrated in comparison with state-of-the-art surrogates for transient finite element analysis. By using the surrogate proposed in this study, oscillations due to discontinuous excitation of mechanical structures can be reproduced. For this purpose, only short-input sequences are necessary since the excitation of the oscillations does not have to be part of the input sequence during the whole duration of the oscillations. Due to the categorical variable for the damping ratio, the surrogate can account for the influence of different damping in parameter studies.

ConsensusPrime—A Bioinformatic Pipeline for Efficient Consensus Primer Design—Detection of Various Resistance and Virulence Factors in MRSA—A Case Study

Background: The effectiveness and reliability of diagnostic tests that detect DNA sequences largely hinge on the quality of the used primers and probes. This importance is especially evident when considering the specific sample being analyzed, as it affects the molecular background and potential for cross-reactivity, ultimately determining the test’s performance. Methods: Predicting primers based on the consensus sequence of the target has multiple advantages, including high specificity, diagnostic reliability, broad applicability, and long-term validity. Automated curation of the input sequences ensures high-quality primers and probes. Results: Here, we present a use case for developing a set of consensus primers and probes to identify antibiotic resistance and virulence genes in Staphylococcus (S.) aureus using the ConsensusPrime pipeline. Extensive qPCR experiments with several S. aureus strains confirm the exceptional quality of the primers designed using the pipeline. Conclusions: By improving the quality of the input sequences and using the consensus sequence as a basis, the ConsensusPrime pipeline pipeline ensures high-quality primers and probes, which should be the basis of molecular assays.

An efficient CNN-transformer hybrid approach for water turbine unit failure prediction

Water Turbine Unit is the core equipment of water power generation. It is the most important research object in hydroelectric power. The analysis of daily monitoring data of Water Turbines can evaluate its running state and avoid the loss caused by failure. In this paper, A hybrid neural network architecture which CNN combines Transformer is proposed to evaluate the health state of water turbines based on multivariate long time series data. It has two core components: (i) dilated convolution applies to capture low-level and local semantic information, then the output of convolutional layers is divided into subseries-level patches by time, these patches are regarded as input tokens of Transformer layers; (ii) utilizing self-attention of Transformer to extract high-level and global semantic information. Patching design naturally has benefit as follow: (i) local semantic information is retained in Transformer input tokens and high-level semantic confirm to common sense of human understanding through low-level construction; (ii) the length of Transformer input sequence is greatly shorted and attention can be more concentrated compared with point-wise form. Meanwhile, the computation and memory usage reduces at the same time. The experimental result indicates that the hybrid architecture can achieves excellent performance in time series understanding.

Short-term multi-step-ahead sector-based traffic flow prediction based on the attention-enhanced graph convolutional LSTM network (AGC-LSTM)

AbstractAccurate sector-based air traffic flow predictions are essential for ensuring the safety and efficiency of the air traffic management (ATM) system. However, due to the inherent spatial and temporal dependencies of air traffic flow, it is still a challenging problem. To solve this problem, some methods are proposed considering the relationship between sectors, while the complicated spatiotemporal dynamics and interdependencies between traffic flow of route segments related to the sector are not taken into account. To address this challenge, the attention-enhanced graph convolutional long short-term memory network (AGC-LSTM) model is applied to improve the short-term sector-based traffic flow prediction, in which spatial structures of route segments related to the sector are considered for the first time. Specifically, the graph convolutional networks (GCN)-LSTM network model was employed to capture spatiotemporal dependencies of the flight data, and the attention mechanism is designed to concentrate on the informative features from key nodes at each layer of the AGC-LSTM model. The proposed model is evaluated through a case study of the typical enroute sector in the central–southern region of China. The prediction results show that MAE reduces by 14.4% compared to the best performing GCN-LSTM model among the other five models. Furthermore, the study involves comparative analyses to assess the influence of route segment range, input and output sequence lengths, and time granularities on prediction performance. This study helps air traffic managers predict flight situations more accurately and avoid implementing overly conservative or excessively aggressive flow management measures for the sectors.

Development of a Screen Keyboard System with Radially‐Arranged Keys and its Effectiveness for Typing Including Eye‐Gaze Input

AbstractThis study aims to develop an effective screen‐keyboard system from a new perspective of key arrangements. In the proposed screen keyboard system, keys are centralized by arranging them radially, with alphabetical keys arranged based on the articulation for speech production. A gesture input method and click input method are implemented considering the mora‐based Japanese sound system with syllables containing a consonant‐vowel (CV) structure. In a preliminary experiment, the cumulative pixel distances required to press keys for the input sequences were evaluated as a simulation to confirm the basic performance of the keyboard system. The Euclidean pixel distances between the keys on the keyboard images were accumulated and compared between the proposed keyboard system and a typical QWERTY keyboard. Typing experiments involving participants were performed to confirm the effectiveness of the proposed keyboard system. The experimental results reveal that the proposed screen keyboard system enables users to generate efficient typing performances using a computer mouse and eye gaze via short‐term training. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

Prediction of Airport Taxiway Settlement based on CNNBiLSTM-Attention

The accurate real-time prediction of complex time series in airport runway settlement is crucial. In order to achieve precise and reliable predictions of runway surface settlement, considering the nonlinear and time-dependent characteristics of settlement data under the coupling effects of dynamic loads, water levels, and other factors, a CNN-BiLSTM-Attention approach is proposed. This method first employs Convolutional Neural Networks (CNN) to extensively extract local temporal features from the input sequences. Subsequently, the extracted local temporal features are input to BiLSTM layers to learn the features of the sequences both in the forward and reverse directions. Finally, the output of the BiLSTM layers is fed into an attention mechanism layer. By computing attention weights, crucial information in the input sequences is weighted and added to enhance the model’s focus on key information. The final settlement prediction is then generated through a fully connected layer. To validate the proposed method, field settlement monitoring data from Runway L at Hangzhou Xiaoshan Airport are utilized as an engineering case study. The results indicate that the combined model captures local features of settlement sequences more effectively with a smaller dataset, achieving higher accuracy compared to a standalone BiLSTM model.