Bidirectional Prediction Research Articles

Adaptive Prediction Structure for Learned Video Compression

Learned video compression has developed rapidly and shown competitive rate-distortion performance compared with the latest traditional video coding standard H.266 (VVC). However, existing works were restricted to fixed prediction direction and GoP size. The inflexibility on prediction structure hinders learned video compression towards optimal compression efficiency in diverse motion scenarios. In this paper, we propose to advance learned video compression with adaptive prediction structure decision. Specifically, we propose a unified compression framework that supports both forward prediction and bi-directional prediction. The framework can flexibly switch to different prediction direction to achieve better prediction performance. Meanwhile, we propose a low-complexity prediction structure decision algorithm, where prediction direction and GoP size are adaptively determined based on motion complexity to achieve optimal compression efficiency. Experimental results demonstrate that the proposed unified framework with adaptive decision algorithm improves compression efficiency of pure forward prediction-based or bi-directional prediction-based framework with neglectable ( \(0.9\%\) ) encoding time increment. Meanwhile, it achieves comparable compression performance with VVC and recent learned video coding methods.

A Method for Filling Missing Values in Multivariate Sequence Bidirectional Recurrent Neural Networks Based on Feature Correlations

Multivariate real-life time series data often contain missing values. These missing values often affect subsequent prediction tasks. Traditional imputation methods generally consider only some of the characteristics of multivariate time series data. This can easily lead to inaccurate filling results. In this paper, a feature correlation-based bidirectional recurrent network (BRNN-FR) is proposed to solve the problem of missing values in multivariate sequence data. First, this method involves the design of a bidirectional prediction network based on time intervals and the use of forward and reverse time series information between data points to obtain the characteristics of data changes with time to the greatest extent. Second, considering the correlation between features, a combined feature selection strategy based on the Pearson correlation coefficient and mutual information was proposed. A multiple regression model was established to predict between features. Finally, a bidirectional network ensemble filling algorithm based on the relationships between features is established to predict missing values. Comprehensive experiments on four public datasets show that the mean absolute error (MAE), root mean square error (RMSE) and maximum R2 value (R2_score) of the BRNN-FR algorithm in the direct imputation test are better than those of the other comparison methods in most cases. BRNN-FR also achieved a better area under the curve (AUC) in the indirect comparison experiment of two classifications of in-hospital death after filling the medical dataset. Using the AIR air quality dataset and the power transformer temperature dataset from the ETTH1 interpolation regression to predict the next 3hours and 6hours of average numerical results, most of the optimal regression results are obtained.

LEC-Codec: Learning-Based Genome Data Compression.

In this paper, we propose a Learning-based gEnome Codec (LEC), which is designed for high efficiency and enhanced flexibility. The LEC integrates several advanced technologies, including Group of Bases (GoB) compression, multi-stride coding and bidirectional prediction, all of which are aimed at optimizing the balance between coding complexity and performance in lossless compression. The model applied in our proposed codec is data-driven, based on deep neural networks to infer probabilities for each symbol, enabling fully parallel encoding and decoding with configured complexity for diverse applications. Based upon a set of configurations on compression ratios and inference speed, experimental results show that the proposed method is very efficient in terms of compression performance and provides improved flexibility in real-world applications.

A novel data-driven framework for enhancing the consistency of deposition contours and mechanical properties in metal additive manufacturing

The accuracy and quality of part formation are crucial considerations. However, the laser directed energy deposition (L-DED) process often leads to irregular changes in deposition contours and mechanical properties across parts due to complex flow fields and temperature variations. Hence, to ensure the forming accuracy and quality, it is necessary to achieve precise monitoring and appropriate parameter adjustments during the processing. In this study, a machine vision method for real-time monitoring is proposed, which combines target tracking and image processing techniques to achieve accurate recognition of deposition contours under noisy conditions. Through comparative verification, the measurement accuracy reaches as high as 98.98 %. Leveraging the monitoring information, a bidirectional prediction neural network is proposed to accomplish layer-by-layer forward prediction of layer height. Meanwhile, inverse prediction is employed to determine the processing parameters required for achieving the desired layer height, facilitating the optimization of the deposition contours. It was found that as the processing parameters were adjusted layer-by-layer to achieve consistent deposition contours, there was also a tendency towards consistent changes in microstructure and mechanical properties. The standard deviations of primary dendrite arm spacing (PDAS) and ultimate tensile strength (UTS) at different positions decrease by over 52.2 % and 61.4 %, respectively. This study reveals the consistent patterns of variation in deposition contours and mechanical properties under data-driven variable parameter processing, laying an important foundation for future exploration of the complex process-structure-performance (PSP) relationship in L-DED.

Bi-Directional Prediction Model for Hot Pressing Production Parameters and Quality of High-Performance Bamboo-Based Fiber Composites Based on cHGWOSCA-SVR

In the hot press process of high-performance bamboo-based fiber composites, there is a highly nonlinear relationship between the production parameters of hot press and the quality parameters of the finished boards. Consequently, it is challenging to accurately predict the quality of the boards based on the given production parameters, and it is equally difficult to preset the production parameters to achieve the desired board quality. The current approach relies on manual experience, which may result in subpar board quality and material waste. To address these issues, this paper proposes a bi-directional prediction model based on cHGWO-SCA-SVR, using the collaboration-based hybrid GWO-SCA optimizer to optimize the relevant parameters of the SVR, and then accurately predicting the production parameters and the quality of the finished boards in both directions. Finally the cHGWO-SCA-SVR prediction model achieves an average of 0.9591 for the forward prediction model and lower MAE and MSE values compared to other models; for the reverse prediction model, it attains an average of 0.9553 and lower MAE and MSE values compared to other models. The results demonstrate the superiority of the cHGWO-SCA-SVR prediction model in comparison with other existing models, proving its significance in guiding the production of high-performance bamboo-based fiber composites by hot compression.

Speckle noise detection and correction for frequency-scanning interferometry in vibration measurement

Speckle noise significantly impairs the accuracy of vibration measurements for non-cooperative targets in frequency-scanning interferometry (FSI). To restore the real vibration, this paper proposes a comprehensive vibration demodulation algorithm based on the FSI. We established an FSI vibration model accounting for speckle noise. The extraction of instantaneous frequency from the interference signal is performed using a complex-shifted Morlet wavelet. An autoregressive differential model, incorporating a smooth weight function, is applied for bidirectional prediction to correct speckle noise. Finally, the true vibration of the target is restored through Kalman filtering. We validated the algorithm’s accuracy using a Monte Carlo strategy. The experimental results show that the surface vibration of black aluminum oxide alloy and the axial runout of the DC motor during operation was successfully captured, and the time–frequency analysis results showed that this method has strong robustness against dense speckle noise.

Relationships among self-esteem, depression and self-injury in adolescents: a longitudinal study.

Non-suicidal self-injury is a widespread mental health concern among adolescents. This study aimed to examine the relationship between self-esteem, depression, and self-injury among adolescents using a longitudinal research design. The Self-Esteem Scale (SES), Child Depression Inventory (CDI), and Adolescent Self-Injury Scale (ASIS) were used to follow up 1,265 junior middle school students on three occasions with six-month intervals. At all three time points, there were significant gender differences in self-esteem, depression, and self-injury. Self-esteem was negatively correlated with depression and self-injury at all three time points, while depression and self-injury were significantly positively correlated. Cross-lagged analysis revealed that self-esteem at Time 1 (T1) did not significantly predict self-injury at Time 2 (T2), but self-esteem (T2) significantly predicted self-injury at Time 3 (T3; β = -0.079, p < 0.05). Similarly, self-injury (T1) significantly predicted self-esteem (T2; β = -0.140, p < 0.001), and self-injury (T2) significantly predicted self-esteem (T3; β = -0.071, p < 0.01). Horizontal and longitudinal mediating analysis showed that depression served as a complete mediator in both the pathway from self-esteem to self-injury and from self-injury to self-esteem. Cross-lagged analysis showed that self-esteem (T1) significantly predicts depression (T2; β = -0.070, p < 0.05), which in turn predict self-injury (T3; β = 0.126, p < 0.001). Similarly, self-injury (T1) predicted depression (T2; β = 0.055, p < 0.05), which further predicted self-esteem (T3; β = -0.218, p < 0.001). The self-esteem, depression, and self-injury of adolescents are closely related; self-esteem and self-injury predict each other; self-esteem indirectly affects self-injury through depression; and self-injury indirectly affects self-esteem through depression. Based on the relationship of bi-directional prediction of self-esteem and self-injury mediated by depression, this study proposes a theoretical model of depression-mediated self-esteem and self-injury cycle.

Unsupervised Extractive Summarization with Learnable Length Control Strategies

Unsupervised extractive summarization is an important technique in information extraction and retrieval. Compared with supervised method, it does not require high-quality human-labelled summaries for training and thus can be easily applied for documents with different types, domains or languages. Most of existing unsupervised methods including TextRank and PACSUM rely on graph-based ranking on sentence centrality. However, this scorer can not be directly applied in end-to-end training, and the positional-related prior assumption is often needed for achieving good summaries. In addition, less attention is paid to length-controllable extractor, where users can decide to summarize texts under particular length constraint. This paper introduces an unsupervised extractive summarization model based on a siamese network, for which we develop a trainable bidirectional prediction objective between the selected summary and the original document. Different from the centrality-based ranking methods, our extractive scorer can be trained in an end-to-end manner, with no other requirement of positional assumption. In addition, we introduce a differentiable length control module by approximating 0-1 knapsack solver for end-to-end length-controllable extracting. Experiments show that our unsupervised method largely outperforms the centrality-based baseline using a same sentence encoder. In terms of length control ability, via our trainable knapsack module, the performance consistently outperforms the strong baseline without utilizing end-to-end training. Human evaluation further evidences that our method performs the best among baselines in terms of relevance and consistency.

Optimization model of process parameters for waterworks based on CNN-ISSA-BiGRU

After conducting an extensive investigation into the current situation of Chinese water supply plants, it was imperative to systematically optimize the critical operational parameters of various treatment units using a machine learning approach. This holistic optimization aimed to reduce operational costs and ensure the desired effluent quality. This work explored the impact of key process parameters on turbidity and CODMn through a pilot experiment, and a CNN-ISSA-BiGRU model was introduced for predicting the optimization of process parameters in water treatment plants. The CNN layer played a crucial role in preprocessing the original data and extracting short-term relationships within the dataset. The ISSA layer was responsible for optimizing the parameters of the BiGRU layer, and the BiGRU layer focused on the bi-directional prediction of data, fully extracting features between input variables. The application of the CNN-ISSA-BiGRU model to actual water plant operations led to a significant reduction in operation costs, up to 17.1%. Furthermore, a correlation analysis of pilot experiment data was conducted to determine the relative importance of each parameter on effluent quality. The results indicated that coagulant dosage had the greaexperiment impact on effluent turbidity, while disinfectant dosage had the utmost effect on CODMn. This comprehensive analysis provided valuable guidance for environmental regulators and operators, offering insights into the mechanisms that govern water treatment processes and aiding in the optimization of plant operations.

Fractional-order long-term price guidance mechanism based on bidirectional prediction with attention mechanism for electric vehicle charging

As the infiltration rate of renewable energy generation (REG) is increasing, the power balance of power systems needs to be maintained urgently. Electric vehicle (EV) as a widely distributed demand-side flexibility resource can be utilized to address the supply-demand balance of the power system. Current EV charging management strategies do not consider user-side demand flexibility and cannot accurately guide users to charge. To address the problem that EV charging demand cannot be accurately directed, this study proposes a fractional-order long-term price guidance mechanism (FLPGM) for EV charging with a bidirectional prediction based on an attention mechanism (AM). The FLPGM combines an AM with a bidirectional gated recurrent unit (BiGRU) and a long-short term memory (LSTM) and applies a high-precision fractional-order stochastic dynamic differentiation method that considers the effects of community health index, EV types, commuting distance, holidays, and power status on charging demand. FLPGM can equate customer charging demand with REG and reduce the cost of charging to customers, resulting in a balance between supply and demand in a long-term electricity market environment. In this study, FLPGM is applied to an electric vehicle charging model (EVCM) for experiments. The test results show that: the charging load of EVs under the FLPGM matches the energy supply by 98.85%; the EVCM under the FLPGM has a charging cost of only 67.61% of the cost in the natural state and can save 79.93% of the potential savings; the EVCM under the FLPGM maintains the balance between supply and demand while reducing charging costs for users, maximizing the benefits for all parties.

Inverse design of reflective metasurface antennas using deep learning from small‐scale statistically random pico‐cells

AbstractA small‐scale data‐selection method is proposed for deep‐learning‐based inverse design of reflective metasurface antennas (MAs). The overall design process involves discretization, coding, data selection, simulation, feature representation, training, prediction, and metasurface design. Unlike conventional method that selects the training data empirically after simulation from a large quantity of unit cells, the proposed statistically random data filtering (SRDF) method selects the training data set randomly before simulation by leveraging the geometric‐coding statistics of the unit cells. The advantages of the proposed data‐selection method include lower simulation cost (due to fewer simulations needed), improved generality, and higher design accuracy simultaneously. The design process starts from the discretization of a unit cell into a grid array of binaurally‐coded pico‐cells. Afterwards, the proposed SRDF method selects 0.01% of all the possible unit cells for simulation. With the simulated results, a deep‐convolutional‐variational‐autoencoder network is trained, leading to an accurate and bidirectional prediction of both the geometry and reflection phase of a unit cell in seconds. The proposed data‐selection method is validated by designing two reflective MAs with the predicted unit cells, each pointing a pencil beam off the boresight by 30° and 60°, respectively. Experiments are in line with full‐wave simulations.

Bi-directional prediction of hydrothermal carbonization characteristics of agroforestry and livestock wastes with variable components: Graph learning model-aided waste recycle

Agricultural resource management is essential in present-day society. The hydrothermal carbonization (HTC) is a promising technology for recovering valuable resources from wastes. However, there is a paucity of efficient approaches to predict and comprehend the agricultural waste HTC process. Machine learning modeling approaches have emerged as an important tool for simulating thermochemical reaction processes such as HTC. However, previous studies all limit to predicting products properties of single-component HTC, which are contrary to the complex composition of raw materials, and demonstrate poor generalization performance. Herein, the homogeneity graph data is employed to represent the HTC data with variable components. Then, based on graph encoder-decoder transformer method, a bi-directional prediction model is established to model the HTC process of agroforestry and livestock wastes. In the case of the forward prediction, the model can predict the composition and yield of hydrochar according to the given HTC conditions. The comparisons with the experimental data other than the training and validation datasets show that the prediction accuracy (R2) of ash, carbon and nitrogen contents of the hydrochar are 0.895, 0.908 and 0.840, respectively, which proves the good generalizability of the model. Moreover, owing to the embedding vector, which connects the reactant data with the product data in HTC, the model can also achieve backward prediction of HTC reaction conditions such as raw material composition and reaction temperature. Furthermore, by dimensionality reduction analysis of the embedding vector in the model, the similarity of any two HTC reactions can be quantified. The established HTC-graph neural network (GNN) model shows both high accuracy and good generalization ability in bi-directional prediction of HTC process. Based on the effectiveness of the proposed model and the given evaluation indexes, the corresponding HTC parameters can be determined, which greatly benefits to the resources saving and energy consumption reduction.

Topological Building Extraction With Bidirectional Prediction From Remote Sensing Images

Topological Building Extraction With Bidirectional Prediction From Remote Sensing Images

Decomposition and prediction of initial uniform bi-directional water waves using an array of wave-rider buoys

Prediction of incident waves to Wave Energy Converters (WEC) is essential to maximize the energy absorption by controlling the WEC. Nevertheless, little work has been done on the deterministic prediction of bi-directional waves whose wave directions of components are 180∘ opposite. To decompose and predict such bi-directional waves, an array of three wave-rider buoys are considered. Buoys on both sides are used for decomposing bi-directional waves into progressive and regressive wave components, and the surface elevation of the middle buoy is predicted by these decomposed waves. The deterministic wave prediction is based on the impulse response function, and a cosine-filtered impulse response function is proposed to reduce an error due to the truncation of the infinite length of the function. Predictions of initial uniform bi-directional waves are shown to demonstrate the performance of the impulse response function method to time-series prediction. Both numerical and experimental comparisons are carried out to validate the proposed methods. The experimental validation revealed that the proposed bi-directional prediction method can predict such waves with at most 5% Mean Absolute Error within this experiment.

Reciprocal Consistency Prediction Network for Multi-Step Human Trajectory Prediction

A reliable prediction of traffic participants’ trajectories is challenging for automated driving. We attempt to integrate human trajectory’s temporal and spatial reciprocal consistency into the prediction network structure. Moreover, humans are intention-driven agents, and modelling human intentions will continuously provide more accurate and detailed information for future trajectory prediction. In this paper, we propose a Reciprocal Consistency Prediction Network (RCPNet), a bi-directional prediction network consisting of the forward and the backward with a similar structure. The prior work of human trajectory prediction is mainly a unidirectional process, while RCPNet fully uses the past and future trajectory information at reverse temporal scales. By training the backward part to optimize the whole network parameters based on the reciprocal consistency, we can improve the prediction accuracy of the forward network. In particular, the framework incorporates a destination variational auto-encoder (DVAE) to estimate the target intention of humans and generate multi-modal trajectory prediction. In order to adapt to the offline prediction situation, we refer to the calculation method of correlation between two variables to present “Predictive Confidence” to clarify the reliability of offline predicted trajectories. Experimental results show that the RCPNet improves state-of-the-art performance on the SDD and ETH/UCY benchmarks.

Abnormal Data Recovery of Structural Health Monitoring for Ancient City Wall Using Deep Learning Neural Network

ABSTRACT Continuous structural health monitoring is of great importance to preventive conservation for ancient architectural heritages. However, abnormal monitoring data may trigger false alarming of structural damages. SHM of ancient buildings also needs abnormal data recovering. Most of the existing studies used the neural network with single input dimension and forward prediction to recover abnormal data, which is difficult to accurately predict long data sequences. This study developed a novel abnormal data recovery framework. The main novelty of the proposed framework is that the input and output configurations of the GRU model are optimized. Meanwhile, to make full use of the forward and backward information of the abnormal data sequence, bidirectional prediction is used to improve the prediction accuracy. The framework is implemented in the abnormal monitoring data recovering for an ancient city wall built 600 years ago in Beijing. Three types of abnormal data, including outlier, drift, and missing, are considered in this study. The results reveal that the proposed framework has high accuracy in abnormal data recovering of strain and crack width. The recovered data has the same regular diurnal variation as the normal monitoring data. The linear correlation between the structural responses and wall temperature gets obviously improved after data recovering. The proposed framework shows great capacity of abnormal data recovery for structural static responses of ancient buildings, which are usually influenced by environmental temperature variation.

Bidirectional prediction of facial and bony shapes for orthognathic surgical planning.

This paper proposes a deep learning framework to encode subject-specific transformations between facial and bony shapes for orthognathic surgical planning. Our framework involves a bidirectional point-to-point convolutional network (P2P-Conv) to predict the transformations between facial and bony shapes. P2P-Conv is an extension of the state-of-the-art P2P-Net and leverages dynamic point-wise convolution (i.e., PointConv) to capture local-to-global spatial information. Data augmentation is carried out in the training of P2P-Conv with multiple point subsets from the facial and bony shapes. During inference, network outputs generated for multiple point subsets are combined into a dense transformation. Finally, non-rigid registration using the coherent point drift (CPD) algorithm is applied to generate surface meshes based on the predicted point sets. Experimental results on real-subject data demonstrate that our method substantially improves the prediction of facial and bony shapes over state-of-the-art methods.

DeepAG: Attack Graph Construction and Threats Prediction With Bi-Directional Deep Learning

The complicated multi-step attacks, such as Advanced Persistent Threats (APTs), have brought considerable threats to cybersecurity because they are naturally varied and complex. Therefore, studying the strategies of adversaries and making predictions are still significant challenges for attack prevention. To address these problems, we propose <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> , a framework utilizing system logs to detect threats and predict the attack paths. <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> leverages transformer models to novelly detect APT attack sequences by modeling semantic information of system logs. On the other hand, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> utilizes Long Short-Term Memory (LSTM) network to propose bi-directional prediction for attack paths, which achieves higher performance than traditional BiLSTM. In addition, with previously detected attack sequences and predicted paths, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> constructs the attack graphs that attackers may follow to compromise the network. Furthermore, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> offers the mechanisms of Out-Of-Vocabulary (OOV) word processor and online update respectively to adapt new attack patterns that show up during detection and prediction stages. The experiments on open-source data sets show that more than 99% of over 15000 sequences can be detected accurately by <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> . Moreover, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepAG</monospace> can improve the baseline by 11.166% of accuracy in terms of prediction.

Bidirectional Spatio-Temporal Feature Learning With Multiscale Evaluation for Video Anomaly Detection

Video anomaly detection aims to detect the segments containing abnormal events from video sequence, which is a current research hotspot due to the importance in maintaining social security. Recent detection methods tend to build frame reconstruction or frame prediction model based on deep learning to learn features of events. The reconstruction-based methods reproduce the input frame one-to-one, inevitably losing some temporal features. The prediction-based methods predict frames according to the natural time order, but ignore the reverse time information, causing the deviation in information learning. Besides, anomaly evaluation methods based on patch-level error neglect the diversity of object sizes in complex scenes, and it is difficult to determine the optimal size of the error patch accurately. For these issues, we propose a bidirectional spatio-temporal feature learning framework with multi-scale anomaly evaluation strategy. A video sequence is input to a double-encoder double-decoder network, and bidirectional spatio-temporal features are obtained for bidirectional prediction by fusing forward and backward features extracted from the two encoders. The multi-scale anomaly evaluation method is implemented based on error pyramid and mean pooling, which effectively detects target objects with different sizes. Experiments on several publicly video datasets show that our method outperforms most of existing methods.

Dopamine mediates the bidirectional update of interval timing.

The role of dopamine (DA) as a reward prediction error (RPE) signal in reinforcement learning (RL) tasks has been well-established over the past decades. Recent work has shown that the RPE interpretation can also account for the effects of DA on interval timing by controlling the speed of subjective time. According to this theory, the timing of the dopamine signal relative to reward delivery dictates whether subjective time speeds up or slows down: Early DA signals speed up subjective time and late signals slow it down. To test this bidirectional prediction, we reanalyzed measurements of dopaminergic neurons in the substantia nigra pars compacta of mice performing a self-timed movement task. Using the slope of ramping dopamine activity as a readout of subjective time speed, we found that trial-by-trial changes in the slope could be predicted from the timing of dopamine activity on the previous trial. This result provides a key piece of evidence supporting a unified computational theory of RL and interval timing. (PsycInfo Database Record (c) 2022 APA, all rights reserved).