Sequence Prediction Problem Research Articles

Residual Spatiotemporal Convolutional Neural Network Based on Multisource Fusion Data for Approaching Precipitation Forecasting

Approaching precipitation forecast refers to the prediction of precipitation within a short time scale, which is usually regarded as a spatiotemporal sequence prediction problem based on radar echo maps. However, due to its reliance on single-image prediction, it lacks good capture of sudden severe convective events and physical constraints, which may lead to prediction ambiguities and issues such as false alarms and missed alarms. Therefore, this study dynamically combines meteorological elements from surface observations with upper-air reanalysis data to establish complex nonlinear relationships among meteorological variables based on multisource data. We design a Residual Spatiotemporal Convolutional Network (ResSTConvNet) specifically for this purpose. In this model, data fusion is achieved through the channel attention mechanism, which assigns weights to different channels. Feature extraction is conducted through simultaneous three-dimensional and two-dimensional convolution operations using a pure convolutional structure, allowing the learning of spatiotemporal feature information. Finally, feature fitting is accomplished through residual connections, enhancing the model’s predictive capability. Furthermore, we evaluate the performance of our model in 0–3 h forecasting. The results show that compared with baseline methods, this network exhibits significantly better performance in predicting heavy rainfall. Moreover, as the forecast lead time increases, the spatial features of the forecast results from our network are richer than those of other baseline models, leading to more accurate predictions of precipitation intensity and coverage area.

Modeling The Prediction of Hard Drive Capacity Usage on Server Computers Based on Linear Regression

Bank of XYZ has a server computer that is used to run several information technology application services such as ATMs and others. Because the server computer uses a hard drive, the full hard drive can cause problems with the service not operating properly. Full hard drives occur without being noticed. So that this makes the computer server problematic, resulting in customer dissatisfaction and decreased customer loyalty to Bank XYZ. To solve the problem at XYZ Bank, one of the machine learning algorithms can be used to predict hard drive capacity. The method used to predict hard drive storage or usage. The machine learning algorithm used is Multiple Linear Regression. The results of this study show that the linear regression model successfully predicts the use of hard drive capacity on server computers with a sufficient level of accuracy.But it is still not optimal because only a few servers can be predicted. For further research, may consider using the LSTM (Long Short-Term Memory) algorithm. LSTM is an algorithm that is well-suited for sequence prediction problems, including time series forecasting.

SODFormer: Streaming Object Detection With Transformer Using Events and Frames.

DAVIS camera, streaming two complementary sensing modalities of asynchronous events and frames, has gradually been used to address major object detection challenges (e.g., fast motion blur and low-light). However, how to effectively leverage rich temporal cues and fuse two heterogeneous visual streams remains a challenging endeavor. To address this challenge, we propose a novel streaming object detector with Transformer, namely SODFormer, which first integrates events and frames to continuously detect objects in an asynchronous manner. Technically, we first build a large-scale multimodal neuromorphic object detection dataset (i.e., PKU-DAVIS-SOD) over 1080.1 k manual labels. Then, we design a spatiotemporal Transformer architecture to detect objects via an end-to-end sequence prediction problem, where the novel temporal Transformer module leverages rich temporal cues from two visual streams to improve the detection performance. Finally, an asynchronous attention-based fusion module is proposed to integrate two heterogeneous sensing modalities and take complementary advantages from each end, which can be queried at any time to locate objects and break through the limited output frequency from synchronized frame-based fusion strategies. The results show that the proposed SODFormer outperforms four state-of-the-art methods and our eight baselines by a significant margin. We also show that our unifying framework works well even in cases where the conventional frame-based camera fails, e.g., high-speed motion and low-light conditions. Our dataset and code can be available at https://github.com/dianzl/SODFormer.

Short-Term Demand Prediction for On-Demand Food Delivery with Attention-Based Convolutional LSTM

Demand prediction for on-demand food delivery (ODFD) is of great importance to the operation and transportation resource utilization of ODFD platforms. This paper addresses short-term ODFD demand prediction using an end-to-end deep learning architecture. The problem is formulated as a spatial–temporal prediction. The proposed model is composed of convolutional long short-term memory (ConvLSTM), and convolutional neural network (CNN) units with encoder–decoder structure. Specifically, long short-term memory (LSTM) networks are a type of recurrent neural network capable of learning order dependence in sequence prediction problems. The convolution unit is responsible for capturing spatial attributes, while the LSTM part is adopted to learn temporal attributes. Additionally, an attentional model is designed and integrated to enhance the prediction performance by addressing the spatial variation in demand. The proposed approach is compared to several baseline models using a historical ODFD dataset from Shenzhen, China. Results indicate that the proposed model obtains the highest prediction accuracy by capturing both spatial and temporal correlations with attention information focusing on different parts of the input series.

Lung pressure predictive model using LSTM: A deep learning techniques

<p>The human body relies on controlled breathing to ensure oxygen reaches all cells while filtering out contaminants to protect the lungs. However, infections like the Delta virus and SARS-CoV2 (COVID-19) have led to Acute Respiratory Distress Syndrome (ARDS), requiring urgent medical care, including mechanical ventilation. The overwhelming number of patients has strained healthcare organizations and workers, necessitating advancements in automated healthcare technology. To address this challenge, we propose a novel solution to predict pressure in mechanical ventilation (MV) for various lung illnesses. The goal is to accurately predict the pressure within the respiratory circuit, which poses a challenging sequence prediction issue. To tackle this, we employ a cutting-edge deep learning approach known as Long Short-Term Memory (LSTM), which exhibits remarkable performance in selectively recalling patterns over time. While traditional recurrent neural networks (RNNs) can handle short-term patterns well, the introduced LSTM technique excels in managing complex sequence prediction problems. Comparing the proposed method with four existing algorithms, the researchers demonstrate that their approach achieves significantly higher accuracy. The impressively low error rate of 1.85 × 10<sup>−7</sup> showcases a substantial improvement over existing system. This groundbreaking advancement has the potential to alleviate the pressure on the current healthcare infrastructure and significantly improve care for patients in need of mechanical ventilation due to respiratory issues.</p>

Stock Price Prediction Using Stacked LSTM

Abstract: In this project we attempt to implement machine learning approach to predict stock prices. Machine learning is effectively implemented in forecasting stock prices. The objective is to predict the stock prices in order to make more informed and accurate investment decisions. We propose a stock price prediction system that integrates mathematical functions, machine learning, and other external factors for the purpose of achieving better stock prediction accuracy and issuing profitable trades. There are two types of stock trading. You may know of intraday trading by the commonly used term "day trading." Intraday traders hold securities positions from at least one day to the next and often for several days to weeks or months. LSTMs are very powerful in sequence prediction problems because they’re able to store past information. This is important in our case because the previous price of a stock is crucial in predicting its future price. While predicting the actual price of a stock is an uphill climb, we can build a model that will predict whether the price will go up or down for next 30 days

Critical measurement parameters estimation in liquid rocket engine using LSTM-based soft sensor

Liquid Rocket engines (LREs) need to be hot tested on the ground, in order to make them flight-worthy. Ground hot tests are critical, expensive, and involve considerable human efforts in acquiring the critical measurement parameters. Measurement sensors are prone to failure due to the hazardous propellant environment and are vulnerable to noise (internal and external) during the LRE ground hot tests. The usage of redundant sensors is not a viable solution for space applications due to cost, power, weight, and space constraints. Data-driven Soft Sensors (SS) are being used in different industrial processes. SS can provide an effective alternative and economical solution for space applications. The objective of the proposed work is to implement a suitable data-driven soft sensor for estimating three critical measurement parameters namely, turbine speed, chamber pressure (Pc), and propellant flow rate of LRE. Long Short-Term Memory (LSTM) networks are capable of learning order dependence in sequence prediction problems and are well-suited to make predictions based on time series data. As the selected LRE test data contain order dependence sequences, LSTM is well suited for the estimation of critical measurement parameters. LSTM-based soft sensor (LSTM-SS) was developed and trained using a large sample set of 59,80,000 data points collected over a period of 12 years of LRE ground hot tests. The maximum root mean square error (RMSE) of 6.06% was obtained for propellant flow-rate estimation, whereas the minimum RMSE of 0.22% for chamber pressure and RMSE of 0.87% for turbine speed were obtained. The obtained results show that the designed LSTM-SS has superior performance and better generalization for practical use.

Spatiotemporal prediction of microstructure evolution with predictive recurrent neural network

Prediction of microstructure evolution during material processing is essential to control the material properties. Simulation tools for microstructure evolution prediction based on physical concepts are computationally expensive and time-consuming. Therefore, they are not practical when either there is an urgent need for microstructure morphology during the process or there is a need to generate big microstructure datasets. Essentially, microstructure evolution prediction is a spatiotemporal sequence prediction problem, where the prediction of material microstructure is difficult due to different process histories and chemistry. We propose a Predictive Recurrent Neural Network (PredRNN) model for the microstructure prediction, which extends the inner-layer transition function of memory states in LSTMs to spatiotemporal memory flow. As a case study, we used a dataset from spinodal decomposition simulation of FeCrCo alloy created by the phase-field method for training and predicting future microstructures by previous observations. The results show that the trained network predicts quantitatively accurate microstructure morphologies while it is several orders of magnitude faster than the phase field method.

Cardiac disease classification from ecg signals using hybrid recurrent neural network method

Medical practitioners and cardiologists frequently employ electrocardiogram (ECG) analysis to assess cardiac health. An automated ECG categorization systems have difficulty detecting and analysing the signals of numerous waves of populations, particularly electrocardiogram (ECG) data. A variety of learning strategies have been researched for the interpretation of ECG categorization. One drawback of machine learning is the use of heuristic features with shallow feature learning architectures. In order to overcome this difficulty, a deep learning approach is used to automatically learn features as opposed to using conventional handmade features. In this work, the use of Long Short-Term Memory (LSTM) and Gated Recurrent Neural Network (GRNN) techniques is proposed to develop an accurate ECG categorization and monitoring system. The purpose of the gated recurrent neural network to processing the sequential data and check the error presented or not in the ECG signals. The long short-term memory is the type of the neural network approaches to identify the complex sequence prediction problems. The GRNN and LSTM models then receive the learnt features that were first captured by the CNN model. There are no custom features required for the ECG categorization in the model. The results section lists various cutting-edge models that the CNN-LSTM and CNN-GRNN models outperformed. The CNN-LSTM and CNN-GRNN designs outperform conventional ECG classification architectures with comparable hyper-parameters, according to the comparison results.

Short-term solar irradiance prediction based on spatiotemporal graph convolutional recurrent neural network

Solar irradiance data include temporal information and geospatial information, so solar irradiance prediction can be regarded as a spatiotemporal sequence prediction problem. However, at present, most of the research is based on time series prediction models, and the research studies on spatial-temporal series prediction models are relatively few. Therefore, it is necessary to integrate spatial-temporal information to construct a spatial-temporal sequence prediction model for research. In this paper, the spatial-temporal prediction model based on graph convolutional network (GCN) and long short-term memory network (LSTM) was established for short-term solar irradiance prediction. In this model, solar radiation observatories were modeled as undirected graphs, where each node corresponds to an observatory, and a GCN was used to capture spatial correlations between sites. For each node, temporal features were extracted by using a LSTM. In order to evaluate the prediction performance of this model, six solar radiation observatories located in the Xinjiang region of China were selected; together with widely used persistence model seasonal autoregressive integrated moving average and data-driven prediction models such as convolutional neural network, recurrent neural network, and LSTM, comparisons were made under different seasons and weather conditions. The experimental results show that the average root mean square error of the GCN-LSTM model at the six sites is 62.058 W/m2, which is reduced by 9.8%, 14.3%, 6.9%, and 3.3%, respectively, compared with other models; the average MAE is 25.376 W/m2, which is reduced by 27.7%, 26.5%, 20.1%, and 11%, respectively, compared with other models; the average R2 is 0.943, which is improved by 1.4%, 2.2%, 0.8%, and 0.4%, respectively, compared with other models.

Gated Attention Recurrent Neural Network: A Deeping Learning Approach for Radar-Based Precipitation Nowcasting

Precipitation nowcasting predicts the future rainfall intensity in local areas in a brief time that impacts directly on human life. In this paper, we express the precipitation nowcasting as a spatiotemporal sequence prediction problem. Predictive learning for a spatiotemporal sequence aims to construct a model of natural spatiotemporal processes to predict the future frames based on historical frames. The spatiotemporal process is an abstraction of some of the spatial things in nature that change with time, and they usually do not change very dramatically. To simplify the model and facilitate the training, we considered that the spatiotemporal process satisfies the generalized Markov properties. The natural spatiotemporal processes are nonlinear and non-stationary in many aspects. The processes are not satisfied with the first-order Markov properties when making predictions, such as the nonlinear movement, expansion, dissipation, and intensity enhancement of echoes. To describe such complex spatiotemporal variations, higher-order Markov models need to be used for the modeling. However, many of the previous models for spatiotemporal prediction constructed were based on first-order Markov properties, losing information on the higher-order variations. Thus, we propose a recurrent neural network which satisfies the multi-order Markov properties to create more accurate spatiotemporal predictions. In this network, the core component is the memory cell structure of the gated attention mechanism, which combines the current input information, extracts the historical state that best matches the existing input from the historical multi-period memory information, and then predicts the future. Through this principle of the gated attention, we could extract the historical state information that is richer and deeper to predict the future and more accurately describe the changing characteristics of motion. The experiments show that our GARNN network captures the spatiotemporal characteristics better and obtains excellent results in the precipitation forecasting with radar echoes.

An Attention Encoder-Decoder Dual Graph Convolutional Network with Time Series Correlation for Multi-Step Traffic Flow Prediction

Accurate traffic prediction is a powerful factor of intelligent transportation systems to make assisted decisions. However, existing methods are deficient in modeling long series spatio-temporal characteristics. Due to the complex and nonlinear nature of traffic flow time series, traditional methods of prediction tasks tend to ignore the heterogeneity and long series dependencies of spatio-temporal data. In this paper, we propose an attentional encoder-decoder dual graph convolution model with time-series correlation (AED-DGCN-TSC) for solving the spatio-temporal sequence prediction problem in the traffic domain. First, the time-series correlation module calculates the sequence similarity by fast Fourier transform and inverse fast Fourier transform, while obtaining multiple possible lengths as possible solutions for the sequence period length. Then, K possible periods fetches are selected and the corresponding sequences are weighted and aggregated to the target sequence. Then, the gated dual graph convolution recurrent unit uses the graph convolution operation, which combines the ideas of node embedding, and dual graph, as an operation inside the gated recurrent structure to capture the spatio-temporal heterogeneity relationship of long sequences. The gated decomposition recurrent module decomposes the time series into the period and trend terms, which are modelled by convolutional gated recurrent unit (ConvGRU) and then fused with features, respectively, and output after graph convolution. Finally, multi-step prediction of future traffic flow is performed in the form of encoder-decoder. Experimental evaluations are conducted on two real traffic datasets, and the results demonstrate the effectiveness of the proposed model.

A LSTM-based approach for modelling the movement uncertainty of indoor trajectories with mobile sensing data

• This work formulates uncertainty modelling of indoor trajectories as a sequence prediction problem and proposes a novel data-driven method. • Features are constructed by considering each step’s stride and heading angle, and then input into a LSTM-based neural network, to predict pointwise deviations of trajectories. • Experiments with real-world datasets, demonstrate our method robustly outperforms previous models, regarding balancing preciseness and completeness. Individuals’ indoor trajectories can be reconstructed with sensing data for indoor navigation. However, the movement uncertainty of such reconstructed indoor trajectories has seldom been modelled before, which seriously affects the reliability of indoor trajectory analytics. Previous methods for movement uncertainty modelling mainly focus on outdoor trajectories, and are based on various user-specified assumptions, which may not hold effective for indoor environments and hence introduce inaccuracy in determining sizes of uncertain regions for indoor trajectories. To address above challenge, this research formulates uncertainty modelling of indoor trajectories as a sequence prediction problem and proposes a novel data-driven method. We construct input features by considering each step’s stride and heading angle. The constructed features are input into a LSTM-based neural network, and pointwise deviations of trajectories are predicted. Our model is data-driven and can learn the unknow mapping from each step’s feature to its corresponding uncertainty via neural network training, thus avoiding errors introduced by inaccurate assumptions of previous methods. We conduct experiments with real-world datasets, and demonstrate our method robustly outperforms previous models, regarding balancing preciseness (over 0.15/m 2 density of covered ground-truth points) and completeness (over 80% ground-truth trajectory coverage). This paper provides a novel approach for generating uncertain regions by predicting pointwise deviations, and can benefit related applications of trajectory mining.

Deep learning model inspired by lateral line system for underwater object detection

Inspired by the lateral line systems of various aquatic organisms that are capable of hydrodynamic imaging using ambient flow information, this study develops a deep learning-based object localization model that can detect the location of objects using flow information measured from a moving sensor array. In numerical simulations with the assumption of a potential flow, a two-dimensional hydrofoil navigates around four stationary cylinders in a uniform flow and obtains two types of sensory data during a simulation, namely flow velocity and pressure, from an array of sensors located on the surface of the hydrofoil. Several neural network models are constructed using the flow velocity and pressure data, and these are used to detect the positions of the hydrofoil and surrounding objects. The model based on a long short-term memory network, which is capable of learning order dependence in sequence prediction problems, outperforms the other models. The number of sensors is then optimized using feature selection techniques. This sensor optimization leads to a new object localization model that achieves impressive accuracy in predicting the locations of the hydrofoil and objects with only 40% of the sensors used in the original model.

PredRANN: The spatiotemporal attention Convolution Recurrent Neural Network for precipitation nowcasting

Precipitation nowcasting is an important task in the fields of transportation, traffic, agriculture, and tourism. One of the main challenges is radar echo maps forecasting. It is regarded as a spatiotemporal sequence prediction problem. The prevailing approaches including the state-of-the-art methods are all based on the ConvRNN which combines the Convolution Neural Network (CNN) and Recurrent Neural Network (RNN). However, the feature flow delivered in multi-layer CNNs and RNN usually accompanies the information loss. Therefore, these algorithms fail to model the long-term dependency and the heavy rainfalls tend to be underestimated. In addition, they cannot predict the increasing intensity trend of heavy rainfalls. In this paper, we propose a PredRANN model by embedding the Temporal Attention Module (TAM) and Layer Attention Module (LAM) into the prediction unit to preserve more representation from temporal and spatial dimensions respectively. The extensive experimental results on both synthetic data sets and real world data sets demonstrate the effectiveness and superiority of the proposed method over state-of-the-art methods. Ablation studies also validate the developed TAM and LAM components. To reproduce the results, we release the source code at: https://github.com/luochuyao/PredRANN.

COVID-19 ChatBot

Abstract: Chat bots are software applications that help users to communicate with the machine and get the required result, this is where Natural Language Processing (NLP) comes into the picture. Natural language processing is based on deep learning that enables computers to acquire meaning from inputs given by the users. Natural language processing techniques can make possible the use of natural language to express ideas, thus drastically increasing accessibility. NLP engines rely on the elements of intent, utterance, entity, context, and session. Here in this project, we will be using Deep learning techniques which will be trained on the dataset which contains categories, patterns, and responses. Long Short-Term Memory (LSTM) is a Recurrent Neural Network that is capable of learning order dependence in sequence prediction problems. One of the most popular RNN approaches is LSTM to identify and control a dynamic system. We use an RNN to classify the category user’s message belongs to and then will give a response from the list of responses. Keywords: NLP – Natural Language Processing, LSTM – Long Short Term Memory, RNN – Recurrent Neural Networks.

Rule Ensemble Learning Using Hierarchical Kernels in Structured Output Spaces

The goal in Rule Ensemble Learning (REL) is simultaneous discovery of a small set of simple rules and their optimal weights that lead to good generalization. Rules are assumed to be conjunctions of basic propositions concerning the values taken by the input features. It has been shown that rule ensembles for classification can be learnt optimally and efficiently using hierarchical kernel learning approaches that explore the exponentially large space of conjunctions by exploiting its hierarchical structure. The regularizer employed penalizes large features and thereby selects a small set of short features. In this paper, we generalize the rule ensemble learning using hierarchical kernels (RELHKL) framework to multi class structured output spaces. We build on the StructSVM model for sequence prediction problems and employ a ρ-norm hierarchical regularizer for observation features and a conventional 2-norm regularizer for state transition features. The exponentially large feature space is searched using an active set algorithm and the exponentially large set of constraints are handled using a cutting plane algorithm. The approach can be easily extended to other structured output problems. We perform experiments on activity recognition datasets which are prone to noise, sparseness and skewness. We demonstrate that our approach outperforms other approaches.

Neural Sequence-to-grid Module for Learning Symbolic Rules

Logical reasoning tasks over symbols, such as learning arithmetic operations and computer program evaluations, have become challenges to deep learning. In particular, even state-of-the-art neural networks fail to achieve \textit{out-of-distribution} (OOD) generalization of symbolic reasoning tasks, whereas humans can easily extend learned symbolic rules. To resolve this difficulty, we propose a neural sequence-to-grid (seq2grid) module, an input preprocessor that automatically segments and aligns an input sequence into a grid. As our module outputs a grid via a novel differentiable mapping, any neural network structure taking a grid input, such as ResNet or TextCNN, can be jointly trained with our module in an end-to-end fashion. Extensive experiments show that neural networks having our module as an input preprocessor achieve OOD generalization on various arithmetic and algorithmic problems including number sequence prediction problems, algebraic word problems, and computer program evaluation problems while other state-of-the-art sequence transduction models cannot. Moreover, we verify that our module enhances TextCNN to solve the bAbI QA tasks without external memory.

Touch Modality Classification Using Recurrent Neural Networks

Recurrent Neural Networks (RNNs) are mainly designed to deal with sequence prediction problems and they show their effectiveness in processing data originally represented as time series. This paper investigates the time series characteristics of RNNs to classify touch modalities represented as spatio temporal 3D tensor data. Different approaches are followed in order to propose efficient RNN models aimed at tactile data classification. The main idea is to capture long-term dependence from data that can be used to deal with long sequences represented by employing Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) architectures. Moreover, a case specific approach to dataset organization of the 3D tensor data is presented. The target is to provide efficient hardware-friendly touch modality classification approaches suitable for embedded applications. To this end, the proposed work achieves effective performance in terms of hardware complexity by reducing the FLOPS by 99.98% and the memory storage by 98.34%, with respect to the state-of-art solutions on the same benchmark dataset. This directly affects the time latency and energy consumption of the embedded hardware. Besides, the implemented models shows a classification accuracy higher than those state-of-art solutions. Results demonstrate that the proposed computing architecture is scalable showing acceptable complexity when the system is scaled up in terms of input matrix size and number of classes to be recognized.

User-centric Activity Recognition and Prediction Model using Machine Learning Algorithms

Human Activity Recognition has been a dynamic research area in recent years. Various methods of collecting data and analyzing them to detect activity have been well investigated. Some machine learning algorithms have shown excellent performance in activity recognition, based on which many applications and systems are being developed. Unlike this, the prediction of the next activity is an emerging field of study. This work proposes a conceptual model that uses machine learning algorithms to detect activity from sensor data and predict the next activity from the previously seen activity sequence. We created our activity recognition dataset and used six machine learning algorithms to evaluate the recognition task. We have proposed a method for the next activity prediction from the sequence of activities by converting a sequence prediction problem into a supervised learning problem using the windowing technique. Three classification algorithms were used to evaluate the next activity prediction task. Gradient Boosting performs best for activity recognition, yielding 87.8% accuracy for the next activity prediction over a 16-day timeframe. We have also measured the performance of an LSTM sequence prediction model for predicting the next activity, where the optimum accuracy is 70.90%.