Collaborative Local-Global Learning for Temporal Action Proposal

2021 ◽  
Vol 12 (5) ◽  
pp. 1-14
Author(s):  
Yisheng Zhu ◽  
Hu Han ◽  
Guangcan Liu ◽  
Qingshan Liu
Keyword(s):  
Short Term Memory ◽  
Background Suppression ◽  
Global Learning ◽  
Short Term ◽  
Term Memory ◽  
Temporal Intervals ◽  
Benchmark Datasets ◽  
Long Short Term Memory ◽  
Temporal Action

Temporal action proposal generation is an essential and challenging task in video understanding, which aims to locate the temporal intervals that likely contain the actions of interest. Although great progress has been made, the problem is still far from being well solved. In particular, prevalent methods can handle well only the local dependencies (i.e., short-term dependencies) among adjacent frames but are generally powerless in dealing with the global dependencies (i.e., long-term dependencies) between distant frames. To tackle this issue, we propose CLGNet, a novel Collaborative Local-Global Learning Network for temporal action proposal. The majority of CLGNet is an integration of Temporal Convolution Network and Bidirectional Long Short-Term Memory, in which Temporal Convolution Network is responsible for local dependencies while Bidirectional Long Short-Term Memory takes charge of handling the global dependencies. Furthermore, an attention mechanism called the background suppression module is designed to guide our model to focus more on the actions. Extensive experiments on two benchmark datasets, THUMOS’14 and ActivityNet-1.3, show that the proposed method can outperform state-of-the-art methods, demonstrating the strong capability of modeling the actions with varying temporal durations.

scholarly journals Long Short Term Memory Recurrent Network for Standard and Poor’s 500 Index Modelling

2018 ◽  
Vol 7 (4.15) ◽  
pp. 25 ◽  
Author(s):  
Said Jadid Abdulkadir ◽  
Hitham Alhussian ◽  
Muhammad Nazmi ◽  
Asim A Elsheikh
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Recurrent Neural Network ◽  
Time Series Data ◽  
Short Term Memory ◽  
Series Data ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory

Forecasting time-series data are imperative especially when planning is required through modelling using uncertain knowledge of future events. Recurrent neural network models have been applied in the industry and outperform standard artificial neural networks in forecasting, but fail in long term time-series forecasting due to the vanishing gradient problem. This study offers a robust solution that can be implemented for long-term forecasting using a special architecture of recurrent neural network known as Long Short Term Memory (LSTM) model to overcome the vanishing gradient problem. LSTM is specially designed to avoid the long-term dependency problem as their default behavior. Empirical analysis is performed using quantitative forecasting metrics and comparative model performance on the forecasted outputs. An evaluation analysis is performed to validate that the LSTM model provides better forecasted outputs on Standard & Poor’s 500 Index (S&P 500) in terms of error metrics as compared to other forecasting models.  

scholarly journals An LSTM-Based Method with Attention Mechanism for Travel Time Prediction

Sensors ◽  
2019 ◽  
Vol 19 (4) ◽  
pp. 861 ◽  
Author(s):  
Xiangdong Ran ◽  
Zhiguang Shan ◽  
Yufei Fang ◽  
Chuang Lin
Keyword(s):  
Short Term Memory ◽  
Attention Mechanism ◽  
Traffic Prediction ◽  
Travel Time Prediction ◽  
Short Term ◽  
Term Memory ◽  
Proposed Model ◽  
Departure Time ◽  
Long Short Term Memory

Traffic prediction is based on modeling the complex non-linear spatiotemporal traffic dynamics in road network. In recent years, Long Short-Term Memory has been applied to traffic prediction, achieving better performance. The existing Long Short-Term Memory methods for traffic prediction have two drawbacks: they do not use the departure time through the links for traffic prediction, and the way of modeling long-term dependence in time series is not direct in terms of traffic prediction. Attention mechanism is implemented by constructing a neural network according to its task and has recently demonstrated success in a wide range of tasks. In this paper, we propose an Long Short-Term Memory-based method with attention mechanism for travel time prediction. We present the proposed model in a tree structure. The proposed model substitutes a tree structure with attention mechanism for the unfold way of standard Long Short-Term Memory to construct the depth of Long Short-Term Memory and modeling long-term dependence. The attention mechanism is over the output layer of each Long Short-Term Memory unit. The departure time is used as the aspect of the attention mechanism and the attention mechanism integrates departure time into the proposed model. We use AdaGrad method for training the proposed model. Based on the datasets provided by Highways England, the experimental results show that the proposed model can achieve better accuracy than the Long Short-Term Memory and other baseline methods. The case study suggests that the departure time is effectively employed by using attention mechanism.

scholarly journals Long Short-Term Memory with Dynamic Skip Connections

2019 ◽  
Vol 33 ◽  
pp. 6481-6488 ◽  
Author(s):  
Tao Gui ◽  
Qi Zhang ◽  
Lujun Zhao ◽  
Yaosong Lin ◽  
Minlong Peng ◽  
...  
Keyword(s):  
Language Processing ◽  
Short Term Memory ◽  
Training Data ◽  
Sequential Data ◽  
Short Term ◽  
Term Memory ◽  
Transition Functions ◽  
Proposed Model ◽  
Long Short Term Memory

In recent years, long short-term memory (LSTM) has been successfully used to model sequential data of variable length. However, LSTM can still experience difficulty in capturing long-term dependencies. In this work, we tried to alleviate this problem by introducing a dynamic skip connection, which can learn to directly connect two dependent words. Since there is no dependency information in the training data, we propose a novel reinforcement learning-based method to model the dependency relationship and connect dependent words. The proposed model computes the recurrent transition functions based on the skip connections, which provides a dynamic skipping advantage over RNNs that always tackle entire sentences sequentially. Our experimental results on three natural language processing tasks demonstrate that the proposed method can achieve better performance than existing methods. In the number prediction experiment, the proposed model outperformed LSTM with respect to accuracy by nearly 20%.

scholarly journals DIANet: Dense-and-Implicit Attention Network

2020 ◽  
Vol 34 (04) ◽  
pp. 4206-4214
Author(s):  
Zhongzhan Huang ◽  
Senwei Liang ◽  
Mingfu Liang ◽  
Haizhao Yang
Keyword(s):  
Classification Accuracy ◽  
Short Term Memory ◽  
Residual Network ◽  
Short Term ◽  
Long Distance ◽  
Term Memory ◽  
Attention Networks ◽  
Network Layers ◽  
Benchmark Datasets ◽  
Long Short Term Memory

Attention networks have successfully boosted the performance in various vision problems. Previous works lay emphasis on designing a new attention module and individually plug them into the networks. Our paper proposes a novel-and-simple framework that shares an attention module throughout different network layers to encourage the integration of layer-wise information and this parameter-sharing module is referred to as Dense-and-Implicit-Attention (DIA) unit. Many choices of modules can be used in the DIA unit. Since Long Short Term Memory (LSTM) has a capacity of capturing long-distance dependency, we focus on the case when the DIA unit is the modified LSTM (called DIA-LSTM). Experiments on benchmark datasets show that the DIA-LSTM unit is capable of emphasizing layer-wise feature interrelation and leads to significant improvement of image classification accuracy. We further empirically show that the DIA-LSTM has a strong regularization ability on stabilizing the training of deep networks by the experiments with the removal of skip connections (He et al. 2016a) or Batch Normalization (Ioffe and Szegedy 2015) in the whole residual network.

scholarly journals Optimal Long Short Term Memory Networks for long-term forecasting of real wind characteristics

2020 ◽  
Vol 53 (1) ◽  
pp. 648-653
Author(s):  
Keerthi N Pujari ◽  
Srinivas S Miriyala ◽  
Prateek Mittal ◽  
Kishalay Mitra
Keyword(s):  
Short Term Memory ◽  
Short Term ◽  
Term Memory ◽  
Wind Characteristics ◽  
Long Short Term Memory ◽  
Long Term Forecasting

scholarly journals Mid-Long Term Inflow Forecasting Using the J-T Test and Long Short-Term Memory Considering Monthly factors

2021 ◽  
Author(s):  
shengli liao ◽  
yitong song ◽  
benxi liu ◽  
zhanwei liu ◽  
zhou fang
Keyword(s):  
Human Activities ◽  
Flood Control ◽  
Short Term Memory ◽  
T Test ◽  
Support Vector ◽  
Short Term ◽  
Term Memory ◽  
Inflow Forecasting ◽  
Long Short Term Memory

Mid-long term inflow forecasting plays an important supporting role in reservoir production planning, drought and flood control, comprehensive utilization and water resource management. Although the inflow data have some periodicity and predictability characteristics, the inflow sequence has complex nonlinearity due to the comprehensive influence of climate, underlying surfaces, human activities and other factors. Therefore, it is difficult to achieve accurate inflow forecasting. In this study, a new hybrid inflow forecast framework that uses previous inflows and monthly factors as inputs, and that adopts Long Short-Term Memory (LSTM) and the Jonckheere-Terpstra test (J-T test) is developed for mid-long term inflow forecasting. First, the J-T test can test whether the monthly average inflow sequence set exhibits significant differences due to climate, underlying surfaces, human activities and other factors to ensure the effectiveness of the framework. Second, the LSTM, which is good at determining the nonlinearity law of the time sequence and finding the best solution, is chosen as the framework algorithm. Finally, due to the periodicity of the inflow sequence, adding monthly factors into the framework can provide more information for the framework to improve the accuracy of the forecast. Xiaowan Hydropower Station in the Lancang River of China is selected as the research area. Six evaluation criteria are used to evaluate established framework using historical monthly inflow data (January 1954-December 2016). The performance of the framework is compared with that of the Back Propagation Neural Network (BPNN) and Support Vector Regression (SVR) models. The results show that the introduction of monthly factors greatly improves the accuracy of the inflow forecast studied, and the proposed method is also better than other frameworks.

Sign in / Sign up

Export Citation Format

Share Document