Long Short-term Memory Recurrent Neural Network Model Research Articles

Predicting Major Bleeding in Patients with Venous Thromboembolism on Extended Anticoagulation Therapy Using Follow-up Data and Long Short-Term Memory Based Recurrent Neural Network

Background: It is often challenging for physicians to decide on the duration of anticoagulation treatment for patients with venous thromboembolism (VTE), as they need to weigh the risk of recurrent thrombosis and bleeding at the same time. Several clinical models have been developed to help physicians identify patients with high risk of bleeding. However, these tools use only the baseline clinical information and are not able to incorporate the clinical conditions and events that occur over time that may influence the risk of bleeding. Therefore, capturing the patterns and relationships in the continuously changing time series of clinical data could be a more powerful approach to developing predictor models than just relying on the baseline clinical information. Nonetheless, creating predictive models from follow-up clinical information is challenging given that time series clinical data are non-uniform, high-dimensional, multivariate observations. Here, we present the first attempt at creating a model that uses the time series follow-up information to predict major bleeding over time using a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) cells. Methods: 2542 patients diagnosed with VTE were enrolled in a prospective cohort study over 8 years. In addition to recording their clinical information at baseline, 6-month follow-up interviews were conducted using a standard script to monitor bleeding status and record clinical information. Major bleeding was defined by the International Society on Thrombosis and Haemostasis, with suspected bleeding events classified by an independent adjudication committee. Overall, 118 patients had major bleeding - a 4.6% incidence rate. The median and mode of the clinical variables were used to impute missing numerical and categorical values, respectively, and for patients who had no follow-up information, for whom bleeding occurred before the first follow-up, an artificial follow-up data point was generated from their corresponding baseline data. Thereafter, the data was divided into two stratified sets: 70% for training, and 30% for testing. Five supervised neural network-based machine learning models with different architectures were trained on the baseline dataset, or the follow-up dataset, or both to predict major bleeding. After training, these machine learning models were tested on the testing set and compared to the conventional clinical models, modified to make them compatible with the available predictor variables in our dataset, including the CHAP, the HAS-BLED, the VTE-BLEED, the RIETE, the ACCP, and the OBRI, which only use the baseline information. Results: Overall, the models that used the follow-up information had a higher area under the Receiver Operating Curve (AUROC) or c-statistic compared to the other models that only relied on the baseline dataset. In particular, the LSTM RNN model was able to achieve AUROC of 81.3% that is more than 10% higher compared to the best performing clinical model. We discovered that the LSTM RNN model mostly relied on features such as number of concomitant medications, years since baseline visit, use of specific antibiotics or antiplatelet agents, and presence of new hypertension to predict bleeding from the follow-up dataset. Furthermore, half of the bleeding events occurred within the first year after patients' baseline visits - a trend reflected in the predictions made by LSTM RNN model. Finally, the models that used both the baseline and the follow-up datasets showed different results depending on their architectures; that is, the simpler ensemble model achieved AUROC of 82.5% while the more complex model had AUROC of 70.8% due to overfitting. Conclusion: We have shown that using time series follow-up data can improve bleeding risk prediction in patients with VTE who are on extended anticoagulant therapy compared to just using the baseline data, and clinicians might benefit from using such an approach. Furthermore, our results indicate that LSTM RNN is a suitable architecture to model routine clinical follow-up data. Finally, we believe using time series data could improve the performance of the other clinical models that are currently based on one-time baseline measurements.

Lithium battery life prediction based on deep learning

Li-ion batteries have the advantages of high efficiency, high energy density and long life, having developed rapidly in recent years. However, it is tough to accurately predict the decline trend of Li-ion battery capacity, which limits the further improvement of their service life and safety. In this study, a method of using wavelet denoising to preprocess the data and predicting the life of Li-ion battery based on whale optimization algorithm combined with long short-term memory network (WOA-LSTM) is proposed. In this paper, two sets of data sets B0005 and B0006 of NASA 's public data set are used. The original battery capacity data is subjected to wavelet transform and noise reduction to remove noise and redundant information. The calculation results of SNR and RMSE are 48.1119 and 0.006225, respectively. Then LSTM (Long short-term memory), RNN (Recurrent Neural Network), GRU (Gated Recurrent Unit) and WOA-LSTM are used to predict the remaining useful life of the data set RUL (Remaining Useful Lifetime, RUL), and the data error is compared, showing that in the results of MAE, RMSE and MAPE three prediction error indicators. The prediction results of WOA-LSTM in B0005 and B0006 show the minimum prediction errors, which are 0.0563,0.0710,0.0415 in B0005 data set and 0.0583,0.0831,0.0454 in B0006 data set. Compared with the standard LSTM model, RNN model and GRU model, the error indexes of the model are decreased, which are 7 %, 4 % and 3 % respectively, which has great advantages. This method can provide a reliable predictive analysis method for battery design and fault diagnosis.

HH model based smart deep brain stimulator to detect, predict and control epilepsy using machine learning algorithm

BackgroundEpilepsy is the most common neurological disorder in the world. To control epilepsy, deep brain stimulation is one of the widely accepted treatment techniques. However, conventional deep brain stimulation technique provides continuous stimulation without optimizing the stimulation parameters, resulting in adverse side effects and unexpected death. Hence, understanding the dynamic behavior of brain neural networks at a cellular level is required for patient-specific epilepsy treatment. Considering the underlying mechanism of a single neuronal shift in the brain neural network, computational model-based techniques have a new face for healthcare, which aims to develop effective medical devices for preclinical investigations. New methodThis paper discusses the design of a Smart Deep Brain Stimulator (SDBS) using the Hodgkin-Huxley (HH) conductance-based cellular model of brain neurons to automatically detect, predict and regulate epilepsy against patient-specific conditions. Epileptic activity is simulated as a spike train of action potential due to sodium and potassium channel conductance variations in the single-neuron HH model. The proposed SDBS consists of three components:- i) seizure detection using bagging and boosting-based ensemble machine learning classifiers, ii) channel conductance prediction using Long Short Term Memory-Recurrent Neural Network (LSTM-RNN) based Deep Neural Network (DNN) for updating model parameters of brain neuron, and iii) model-based intelligent control of epileptic seizure with Nonlinear Autoregressive Moving Average-L2 (NARMA-L2) Controller and Nonlinear Model Predictive Controller (NMPC). ResultsFor effective treatment, improving the overall accuracy and efficiency of SDBS is essential. For epilepsy detection, the ensemble bagging machine learning algorithm provides better accuracy of 92.7% compared to the ensemble boosting algorithm. LSTM-RNN deep neural network model with four layers predicts the variations in channel conductance with Root Mean Square Error (RMSE) of 0.00568 and 0.009081 for sodium and potassium channel conductance, respectively. From the closed-loop performances of SDBS with an intelligent control scheme, it is observed that SDBS with NMPC provides efficient and accurate stimulation with minimum energy consumption. From a stability point of view, SDBS with NMPC provides better stability than SDBS with NARMA-L2 Controller. Comparison with existing methodThe proposed SDBS is designed to generate accurate stimulation pulses for epilepsy patients with specific conditions depending on the neuronal activity of a single neuron. Moreover, it will also adapt to the dynamic condition of epilepsy patients. The existing deep brain stimulator continuously provides stimulation pulses without adapting to the patient’s conditions. ConclusionThe proposed SDBS could provide patient-specific treatment based on sodium/potassium channel conductance variations of brain neurons. It will help increase the use of deep brain stimulation techniques and reduce sudden death. Furthermore, the proposed technique will be extended to neural network models with larger neuronal populations to improve the practical feasibility.

Trajectory Propagation of Symmetrical Quasi-classical Dynamics with Meyer-Miller Mapping Hamiltonian Using Machine Learning

The long short-term memory recurrent neural network (LSTM-RNN) approach is applied to realize the trajectory-based nonadiabatic dynamics within the framework of the symmetrical quasi-classical dynamics method based on the Meyer-Miller mapping Hamiltonian (MM-SQC). After construction, the LSTM-RNN model allows us to propagate the entire trajectory evolutions of all involved degrees of freedoms (DOFs) from initial conditions. The proposed idea is proven to be reliable and accurate in the simulations of the dynamics of several site-exciton electron-phonon coupling models and three Tully's scattering models. It indicates that the LSTM-RNN model perfectly captures the dynamical information on the trajectory evolution in the MM-SQC dynamics. Our work proposes a novel machine learning approach in the simulation of trajectory-based nonadiabatic dynamic of complex systems with a large number of DOFs.

Deep Learning LSTM Recurrent Neural Network Model for Prediction of Electric Vehicle Charging Demand

The immense growth and penetration of electric vehicles has become a major component of smart transport systems; thereby decreasing the greenhouse gas emissions that pollute the environment. With the increased volumes of electric vehicles (EV) in the past few years, the charging demand of these vehicles has also become an immediate requirement. Due to which, the prediction of the demand of electric vehicle charging is of key importance so that it minimizes the burden on the electric grids and also offers reduced costs of charging. In this research study, an attempt is made to develop a novel deep learning (DL)-based long-short term memory (LSTM) recurrent neural network predictor model to carry out the forecasting of electric vehicle charging demand. The parameters of the new deep long-short term memory (DLSTM) neural predictor model are tuned for its optimal values using the classic arithmetic optimization algorithm (AOA) and the input time series data are decomposed so as to maintain their features using the empirical mode decomposition (EMD). The novel EMD—AOA—DLSTM neural predictor modeled in this study overcomes the vanishing and exploding gradients of basic recurrent neural learning and is tested for its superiority on the EV charging dataset of Georgia Tech, Atlanta, USA. At the time of simulation, the best results of 97.14% prediction accuracy with a mean absolute error of 0.1083 and a root mean square error of 2.0628 × 10−5 are attained. Furthermore, the mean absolute error was evaluated to be 0.1083 and the mean square error pertaining to 4.25516 × 10−10. The results prove the efficacy of the prediction metrics computed with the novel deep learning LSTM neural predictor for the considered dataset in comparison with the previous techniques from existing works.

Prediction of water turbidity in a marine environment using machine learning: A case study of Hong Kong

The water quality measurement of marine water is a key research topic for environmental and ocean modelers in the past several decades. Marine water quality is mainly described by its chemical, physical and biological properties. According to the Environmental Protection Department (EPD), Hong Kong, the chemical and physical measurements are integrative parameters for water quality measurements. The existence of undesirable components in marine water degrades the water quality. The increase of natural and anthropogenic events in coastal regions poses the marine ecosystem in danger day by day. Therefore, the prediction of marine water quality indicators is essential. Turbidity is a key indicator of marine water quality, whose prediction is difficult due to its non-linear time series behavior. Therefore, the objective of the present study is to predict turbidity in the marine environment of Hong Kong using Machine Learning (ML) approach. The artificial neural network (ANN), support vector regression (SVR), and Long short-term memory recurrent neural network (LSTM-RNN) have been used as ML tools in the marine environment. The ML model prediction results were compared and the obtained results suggest that the LSTM-RNN model outperforms the ANN and SVR model results with an accuracy of 88.45%. The neural network-based approaches give better prediction accuracy over the SVR model in this study. Thus, it can be concluded that the neural network-based algorithm results can be used to monitor water quality parameters in the marine environment.

Spatiotemporal Dynamic Forecasting and Analysis of Regional Traffic Flow in Urban Road Networks Using Deep Learning Convolutional Neural Network

The purpose is to explore the spatial-temporal dynamic prediction performance of urban road network traffic flow based on convolutional neural networks (CNN) of deep learning. A dynamic prediction model of road network traffic flow based on STGCN-BiLSTM (spatial-temporal graph convolution network Bi-directional Long Short-Term Memory) is designed in view of the complex and highly nonlinear traffic data in the actual environment. Finally, the simulation experiment is conducted on the constructed model to verify its performance which is compared with that of the LSTM (Long Short-Term Memory) model, CNN model, RNN (Recurrent Neural Network) model, AlexNet model and STGCN model. The results show that the root mean square error, mean absolute error and mean absolute percentage error of the proposed algorithm model are 4.60%, 5.46% and 7.73%, respectively. The training time is stable at about 45s, the test time is stable at about 33s, and the delay is close to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.86\times 10 ^{-4}\text{s}$ </tex-math></inline-formula> . Additionally, the traffic flow of the test section in the next 15 min, 30 min, 1 h and 2 h is further predicted. Compared with other algorithms, the predicted value of the proposed algorithm is the closest to the actual value, along with the best prediction effect. Therefore, the constructed dynamic prediction model of road network traffic flow can achieve high accuracy and better robustness under the premise of low error, which can provide experimental basis for the spatial-temporal dynamic digital development of transportation in smart cities.

Identification of Oscillatory Modes in Power System Using Deep Learning Approach

The increase in electric power demand pushes the modern power system for more interconnected networks. It leads to a lack of inertia and creates more critical disturbances in the power system. When this oscillation isn’t damped out, it results in cascade tripping. Immediate detection of low-frequency oscillatory modes and their parameters will help the power system operator to act on a particular event without consuming much time. This research paper proposes novel strategies for identifying low-frequency modes using deep learning techniques, and the model can predict the LFO modes in different topologies. This work presents the Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) approach to predict the instantaneous mode oscillatory parameters in the power system. Once the LSTM-RNN model is trained for different power disturbance situations, it can be used for any events associated with the system. Simulation results are verified using two area Kundur systems at various disturbance conditions. The simulations are performed using MATLAB software and python tensor flow library. The results are validated using statistical methods, and it confirms the superior viability and adaptability of the proposed approach in predicting the instantaneous mode parameters.

An Interpretation of Long Short-Term Memory Recurrent Neural Network for Approximating Roots of Polynomials

This paper aims to present a flexible method for interpreting the Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) for the relational structure between the roots and the coefficients of a polynomial. A database is first developed for randomly selected inputs based on the degrees of the univariate polynomial which is then used to approximate the polynomial roots through the proposed LSTM-RNN model. Furthermore, an adaptive learning optimization algorithm is used specifically to update the network weights iteratively based on training deep neural networks data. Thus, the method can exploit the ability to find the individual learning rates for each variable through adaptive learning rate strategies to effectively prevent the weights from fluctuating in a wide spectrum. Finally, several experimental results are performed which shows that the proposed LSTM-RNN model can be used as an alternative approach to compute an approximation of each root for a given polynomial. Furthermore, the results are compared with the conventional feedforward neural network based artificial neural network model. The results clearly demonstrate the superiority of the proposed LSTM-RNN model for roots approximation in terms of accuracy, mean square error and faster convergence.

GAMPAL: an anomaly detection mechanism for Internet backbone traffic by flow size prediction with LSTM-RNN

This paper proposes a general-purpose anomaly detection mechanism for Internet backbone traffic named GAMPAL (General-purpose Anomaly detection Mechanism using Prefix Aggregate without Labeled data). GAMPAL does not require labeled data to achieve general-purpose anomaly detection. For scalability to the number of entries in the BGP RIB (Border Gateway Protocol Routing Information Base), GAMPAL introduces prefix aggregate. The BGP RIB entries are classified into prefix aggregates, each of which is identified with the first three AS (Autonomous System) numbers in the AS_PATH attribute. GAMPAL establishes a prediction model for traffic sizes based on past traffic sizes. It adopts a LSTM-RNN (Long Short-Term Memory Recurrent Neural Network) model that focuses on the periodicity of the Internet traffic patterns at a weekly scale. The validity of GAMPAL is evaluated using real traffic information, BGP RIBs exported from the WIDE backbone network (AS2500), a nationwide backbone network for research and educational organizations in Japan, and the dataset of an ISP (Internet Service Provider) in Spain. As a result, GAMPAL successfully detects anomalies such as increased traffic due to an event, DDoS (Distributed Denial of Service) attacks targeted at a stub organization, a connection failure, an SSH (Secure Shell) scan attack, and anomaly spam.

A LSTM-RNN-Based Fiber Optic Gyroscope Drift Compensation

Fiber optic gyroscope (FOG) inertial measurement unit (IMU) containing a three-orthogonal gyroscope and three-orthogonal accelerometer has been widely utilized in positioning and navigation of military and aerospace fields, due to its simple structure, small size, and high accuracy. However, noise such as temperature drift will reduce the accuracy of FOG, which will affect the resolution accuracy of IMU. In order to reduce the FOG drift and improve the navigation accuracy, a long short-term memory recurrent neural network (LSTM-RNN) model is established, and a real-time acquisition method of the temperature change rate based on moving average is proposed. In addition, for comparative analysis, backpropagation (BP) neural network model, CART-Bagging, classification and regression tree (CART) model, and online support vector machine regression (Online-SVR) model are established to filter FOG outputs. Numerical simulation based on field test data in the range of -20°C to 50°C is employed to verify the effectiveness and superiority of the LSTM-RNN model. The results indicate that the LSTM-RNN model has better compensation accuracy and stability, which is suitable for online compensation. This proposed solution can be applied in military and aerospace fields.

Option hedging using LSTM-RNN: an empirical analysis

This paper proposes an optimal hedging strategy in the presence of market frictions using the Long Short Term Memory Recurrent Neural Network (LSTM-RNN) method, which is a modification of the method proposed in Buehler et al. (Deep hedging. Quant. Finance, 2019, 19(8), 1271–1291). The market frictions are transaction costs, liquidity constraints, trading limits and cost of funds. The loss function is a spectral risk measure. We first make an empirical analysis of the LSTM-RNN model of real option markets, which are the Asian market (domestic market 50 ETF option, Hong Kong Hang Seng Index Option, Nikkei Index Option), the North American market (S&P 500 Index Option) and the European market (FTSE 100 Index Option). The benchmark models are from Leland (Option pricing and replication with transaction costs. J. Finance., 1985, 40(5), 1283–1301), Boyle and Vorst (Option replication in discrete time with transaction costs. J. Finance, 1992, 47(1), 271–293) and Whalley and Wilmott (A hedging strategy and option valuation model with transaction costs. OCIAM Working Paper, Mathematical Institute, Oxford, 1993). Finally, we compare the results from the LSTM-RNN model with benchmark models involving transaction costs for both simulated market data generated by Geometric Brownian Motion (GBM) and the Heston model and real market data. The results show that the LSTM-RNN model outperforms benchmark models for low or medium volatility (<0.8), OTM moneyness and under a certain risk level (<80%) in the GBM market setting. For the 50ETF option market, the LSTM-RNN model outperforms benchmark models for ATM and under a certain risk level (<15%). For the HSI option market, the LSTM-RNN model outperforms benchmark models when transaction costs are smaller than 1.5%. For the Nikkei and S&P 500 option markets, the LSTM-RNN model always outperforms benchmark models. For the FTSE option market, the LSTM-RNN model outperforms benchmark models when moneyness is not too deeply ITM.

Adaptive Anomaly Detection Framework Model Objects in Cyberspace.

Telecommunication has registered strong and rapid growth in the past decade. Accordingly, the monitoring of computers and networks is too complicated for network administrators. Hence, network security represents one of the biggest serious challenges that can be faced by network security communities. Taking into consideration the fact that e-banking, e-commerce, and business data will be shared on the computer network, these data may face a threat from intrusion. The purpose of this research is to propose a methodology that will lead to a high level and sustainable protection against cyberattacks. In particular, an adaptive anomaly detection framework model was developed using deep and machine learning algorithms to manage automatically-configured application-level firewalls. The standard network datasets were used to evaluate the proposed model which is designed for improving the cybersecurity system. The deep learning based on Long-Short Term Memory Recurrent Neural Network (LSTM-RNN) and machine learning algorithms namely Support Vector Machine (SVM), K-Nearest Neighbor (K-NN) algorithms were implemented to classify the Denial-of-Service attack (DoS) and Distributed Denial-of-Service (DDoS) attacks. The information gain method was applied to select the relevant features from the network dataset. These network features were significant to improve the classification algorithm. The system was used to classify DoS and DDoS attacks in four stand datasets namely KDD cup 199, NSL-KDD, ISCX, and ICI-ID2017. The empirical results indicate that the deep learning based on the LSTM-RNN algorithm has obtained the highest accuracy. The proposed system based on the LSTM-RNN algorithm produced the highest testing accuracy rate of 99.51% and 99.91% with respect to KDD Cup'99, NSL-KDD, ISCX, and ICI-Id2017 datasets, respectively. A comparative result analysis between the machine learning algorithms, namely SVM and KNN, and the deep learning algorithms based on the LSTM-RNN model is presented. Finally, it is concluded that the LSTM-RNN model is efficient and effective to improve the cybersecurity system for detecting anomaly-based cybersecurity.

Predicting Vacant Parking Space Availability: A Long Short-Term Memory Approach

The accurate prediction of vacant parking space availability is becoming increasingly essential for assisting drivers to determine where to park in advance. It helps ease traffic pressure and reduce gas emission and pollution. This article proposes a novel multistep long short-term memory recurrent neural network (LSTM-NN) model to predict the number of the vacant parking spaces on the basis of historical parking availability information. The key parameters of the model are deeply optimized. The prediction model is fully benchmarked with five well-known machine learning models&#x2014;i.e. a gated recurrent units neural network, a stacked autoencoder, a support vector regression, a back propagation neural network, and a <i>k</i>-nearest neighbor algorithm&#x2014;whose key parameters are sufficiently optimized as well. Adequate experiments with practical data collected from two parking lots with various capacities and traffic flows were conducted to evaluate the models&#x2019; performances on short- and long-term predictions. Experimental results show that the proposed multistep LSTM-NN model outperforms all the benchmark models, especially in a commercial parking lot with heavy traffic flow.

A day-ahead PV power forecasting method based on LSTM-RNN model and time correlation modification under partial daily pattern prediction framework

Photovoltaic (PV) power generation is an effective means to realize solar energy utilization. Due to the natural characteristics of random fluctuations in solar energy, the applications of PV power such as grid-connected PV power plant, distributed PVs, and building integrated PVs will introduce new characteristics to the generation and load side of the power grid. Therefore, accurate day-ahead PV power forecasting is of great significance for enabling grid manager to achieve PV power output data in advance and mitigate the influence of random fluctuations. To tackle the deficiencies of conventional artificial intelligence (AI) modeling methods such as overfitting problem and insufficient generalization ability to complex nonlinear modeling, a day-ahead PV power forecasting model assembled by fusing deep learning modeling and time correlation principles under a partial daily pattern prediction (PDPP) framework is proposed. First, an independent day-ahead PV power forecasting model based on long-short-term memory recurrent neural network (LSTM-RNN) is established. Second, a modification method is proposed to update the forecasting results of LSTM-RNN model based on time correlation principles regarding different patterns of PV power in the forecasting day. Third, a partial daily pattern prediction (PDPP) framework is proposed to provide accurate daily pattern prediction information of particular days, which is used to guide the modification parameters. Simulation results show that the proposed forecasting method with time correlation modification (TCM) is more accurate than the individual LSTM-RNN model, and the performance of the forecasting model can be further improved for those days with accurate daily pattern predictions under the proposed PDPP framework.

Long short-term memory (LSTM) recurrent neural network for low-flow hydrological time series forecasting

This article explores the suitability of a long short-term memory recurrent neural network (LSTM-RNN) and artificial intelligence (AI) method for low-flow time series forecasting. The long short-term memory works on the sequential framework which considers all of the predecessor data. This forecasting method used daily discharged data collected from the Basantapur gauging station located on the Mahanadi River basin, India. Different metrics [root-mean-square error (RMSE), Nash–Sutcliffe efficiency (ENS), correlation coefficient (R) and mean absolute error] were selected to assess the performance of the model. Additionally, recurrent neural network (RNN) model is also used to compare the adaptability of LSTM-RNN over RNN and naive method. The results conclude that the LSTM-RNN model (R = 0.943, ENS = 0.878, RMSE = 0.487) outperformed RNN model (R = 0.935, ENS = 0.843, RMSE = 0.516) and naive method (R = 0.866, ENS = 0.704, RMSE = 0.793). The finding of this research concludes that LSTM-RNN can be used as new reliable AI technique for low-flow forecasting.

Modeling Speech Acts in Asynchronous Conversations: A Neural-CRF Approach

Participants in an asynchronous conversation (e.g., forum, e-mail) interact with each other at different times, performing certain communicative acts, called speech acts (e.g., question, request). In this article, we propose a hybrid approach to speech act recognition in asynchronous conversations. Our approach works in two main steps: a long short-term memory recurrent neural network (LSTM-RNN) first encodes each sentence separately into a task-specific distributed representation, and this is then used in a conditional random field (CRF) model to capture the conversational dependencies between sentences. The LSTM-RNN model uses pretrained word embeddings learned from a large conversational corpus and is trained to classify sentences into speech act types. The CRF model can consider arbitrary graph structures to model conversational dependencies in an asynchronous conversation. In addition, to mitigate the problem of limited annotated data in the asynchronous domains, we adapt the LSTM-RNN model to learn from synchronous conversations (e.g., meetings), using domain adversarial training of neural networks. Empirical evaluation shows the effectiveness of our approach over existing ones: (i) LSTM-RNNs provide better task-specific representations, (ii) conversational word embeddings benefit the LSTM-RNNs more than the off-the-shelf ones, (iii) adversarial training gives better domain-invariant representations, and (iv) the global CRF model improves over local models.