Adaptive Hidden Markov Model Research Articles

Forecasting corporate credit spreads: regime-switching in LSTM

A long short-term memory model (LSTM) which utilises regime-switching state information as a feature to predict the change of credit spreads is developed. Latent changes in the market are filtered out from observable credit spread time series. These hidden information of regime changes are incorporated into an LSTM, where the state probability is utilised as a feature for one-step ahead predictions of the credit spreads. Firstly, time series from corporate credit spreads are modelled through a Hidden Markov model (HMM) which is based on a discretised Ornstein-Uhlenbeck process. State-related information of the Markov chain, like the jump frequency and state occupation time hidden in the observed spreads are filtered out and adaptive HMM filters are built to estimate probabilities of hidden market states. The performance of the LSTM with regime-switching information is analysed and compared to the accuracy of a pure LSTM without state features. Furthermore, purely utilising the HMM forecast, the prediction of the credit spread is compared to the prediction within the LSTM. Beyond a simulations study, the HMM-LSTM model is calibrated on corporate credit spreads from three European countries between 2004 and 2019. The findings show that the LSTM forecast error is improved when regime information is added, mostly in cases with stronger market fluctuations.

An Approach for Automated Kannada Subtitle Generation from Kannada Video

This paper presents an automated Kannada subtitle generator from Kannada video which is implemented to assist people with auditory problems for watching videos. Henceforth the subtitle generation has become an important task for supporting such special people and it integrates an audio extraction and a speech recognition module. Three phases of the proposed technique were implemented, such as extracting audio from video, Recognition of Speech and Generation of Subtitle. An adaptive speech recognition module is implemented AMFCC for feature extraction which was an alternative to the most commonly used FFT. Hankel transform which was similar to FFT, but includes no elementary particles such as FFT. In addition to it, in the decoder acoustic module, such as Adaptive Hidden Markov Model using the Baum-Welch algorithm is utilized instead of a Viterbi algorithm to reduce the computational time and memory usage. The text file from the speech recognition module is rendered to synchronize the missing offset with the video using parallel processing by defining the start time, the end time, the delay time. Best outcomes are demonstrated by the experimental results of the proposed technique with 98.4% of accuracy compare with existing techniques. The proposed technique which gives 3.8% better accuracy performance compare with existing technique i.e. MFCC, DNN and CNN.

An adaptive HMM method to simulate and forecast ocean chemistry data in aquaculture

Due to the multifaceted influences of many factors and complex mechanisms, accurate predictions of ocean chemistry data are difficult to achieve. To address this problem, this paper proposes an adaptive hidden Markov model (AHMM) for predicting ocean chemistry data from aquaculture farms in eastern China. The periodic, trend and residual terms of the original sequence are obtained through Seasonal and Trend decomposition using Loess (STL), which effectively reveals the variation trend of the sequence. For the period and trend terms of the sequence, the autonomous search grasshopper optimization algorithm (ASGOA) is employed to train the parameters of the HMM, which avoids the problems of easily falling into local optima and the dependence on the initial value in the parameter training process. To predict ocean chemistry data, a sequence similarity measurement method is proposed to match similar sequences in the historical interval, and a linear regression equation is constructed for sequence similarity and the difference between the predicted value and the value at the previous time based on the HMM. The results have revealed that the MAE value decreases by 25.5%, MSE value by 45.9%, RMSE value by 27.4%, MAPE value by 39.5% and R2 increases by 6%. Compared with other prediction algorithms under the same conditions, the prediction method proposed in this paper is easily understandable and manages to achieve better prediction performance. This work can serve as a reference for predicting other ocean chemistry data and for the sustainable management of aquaculture ecosystems.

An Adaptive Hidden Markov Model with CCA for Privacy Preserving of Correlated Big Data

An Adaptive Hidden Markov Model with CCA for Privacy Preserving of Correlated Big Data

An Adaptive Hidden Markov Model with CCA for Privacy Preserving of Correlated Big Data

An Adaptive Hidden Markov Model with CCA for Privacy Preserving of Correlated Big Data

Segmentation of Ordinary Images and Medical Images with an Adaptive Hidden Markov Model and Viterbi Algorithm

Background: Image segmentation is an important problem in both image processing and computer vision. Given an image, the goal of image segmentation is to label each pixel in the image such that the pixels with the same label collectively represent an object. Materials and Methods: Due to the inherent complexity and noise that may exist in images, developing an algorithm that can generate excellent segmentation results for an arbitrary image is still a challenging problem. In this paper, a new adaptive Hidden Markov Model is developed to describe the spatial and semantic relationships among pixels in an image. Based on this statistical model, image segmentation can be efficiently performed with an adaptive Viterbi algorithm in linear time. Results: The algorithm is unsupervised and does not require being used along with any other approach in image segmentation. Testing results on synthetic and real images show that this algorithm is able to achieve excellent segmentation results in both ordinary images and medical images. Conclusion: An implementation of this algorithm in MATLAB is freely available upon request.

Intelligent Cache Pollution Attacks Detection for Edge Computing Enabled Mobile Social Networks

With the rapid advances of wireless technologies and popularization of smart mobile devices, edge-enabled mobile social networks (MSNs) have emerged as a promising network paradigm for mobile users to deliver, share, and exchange contents with each other. By leveraging edge caching technology, various content services can be provided to mobile users for improving their quality of experience (QoE). However, edge caching is vulnerable to cache pollution attacks (CPAttacks) with the result of disruptive content delivery. To tackle this problem, we propose a hidden Markov model (HMM) based CPAttack detection scheme in edge-enabled MSNs. Specifically, we first present the CPAttack model based on observations of attacking behaviors. According to the CPAttack model, the caching state of the edge device is characterized by two parameters-content request rate and cache missing rate. Then, with observation sequence constructed by caching states, we develop an HMM-based detection algorithm to distinguish the CPAttack in the approximately time-invariant content request process. To deal with the lack of training data and dynamic of caching states, an adaptive HMM (AHMM) based algorithm is designed to detect the CPAttack in the time-varying content request process. The simulation results demonstrate that the proposed scheme can efficiently improve edge devices’ abilities to sense the CPAttack.

Robust Lane-Change Recognition Based on An Adaptive Hidden Markov Model Using Measurement Uncertainty

Lane-changing of surrounding vehicles is a risky situation because vehicle accidents can be easily caused by driver’s unawareness of the surrounding vehicle. Much research has conducted on lane-change recognition (LCR) to avoid these vehicle accidents by warning drivers. LCR is a technology to estimate lane-changing behaviors of surrounding vehicles from observation data: position, velocity, and lane information. Since these observation data change continuously during lanechanging, most research for LCR has used time series data based on hidden Markov model (HMM). A challenging point of LCR is that HMM could make false positives in LCR when the observation data include uncertainties such as sensor noise and object detection error. Previous research has tried to process observation data by using Bayesian filter. However, the approach cannot remove all data uncertainties. This paper proposes a method for using observation uncertainty through an adaptive HMM for LCR. In the method, HMM models are modified in real time based on data covariance to filter data with high uncertainty. For evaluation of the algorithm, it was tested through 71 lane-changing cases in real driving situations. The results show that the proposed method enhanced the recognition accuracy by 25.3 % (63.3 % → 88.7 %) than a previous LCR method.

The Effect of Personalization in Longer-Term Robot Tutoring

The benefits of personalized social robots must be evaluated in real-world educational contexts over periods of time longer than a single session to understand their full potential to impact learning outcomes. In this work, we describe a personalization system designed for longer-term personalization that orders curriculum based on an adaptive Hidden Markov Model (HMM) that evaluates students’ skill proficiencies. We present a study investigating the effectiveness of this system in a five-session interaction with a robot tutor, taking place over the course of 2 weeks. Our system is evaluated in the context of native Spanish-speaking first-graders interacting with a social robot tutor while completing an English Language Learning educational task. Participants either received lessons: (1) ordered by our adaptive HMM personalization system which selects a lesson based on a skill that the individual participant needs more practice with (“personalized condition”) or (2) ordered randomly from among the lessons the participant had not yet seen (“non-personalized condition”). We found that participants who received personalized lessons from the robot tutor outperformed participants who received non-personalized lessons on a post-test by 2.0 standard deviations on average, corresponding to a mean learning gain in the 98th percentile.

Online Early-Late Fusion Based on Adaptive HMM for Sign Language Recognition

In sign language recognition (SLR) with multimodal data, a sign word can be represented by multiply features, for which there exist an intrinsic property and a mutually complementary relationship among them. To fully explore those relationships, we propose an online early-late fusion method based on the adaptive Hidden Markov Model (HMM). In terms of the intrinsic property, we discover that inherent latent change states of each sign are related not only to the number of key gestures and body poses but also to their translation relationships. We propose an adaptive HMM method to obtain the hidden state number of each sign by affinity propagation clustering. For the complementary relationship, we propose an online early-late fusion scheme. The early fusion (feature fusion) is dedicated to preserving useful information to achieve a better complementary score, while the late fusion (score fusion) uncovers the significance of those features and aggregates them in a weighting manner. Different from classical fusion methods, the fusion is query adaptive. For different queries, after feature selection (including the combined feature), the fusion weight is inversely proportional to the area under the curve of the normalized query score list for each selected feature. The whole fusion process is effective and efficient. Experiments verify the effectiveness on the signer-independent SLR with large vocabulary. Compared either on different dataset sizes or to different SLR models, our method demonstrates consistent and promising performance.

Adaptive hidden Markov model-based online learning framework for bearing faulty detection and performance degradation monitoring

This study proposes an adaptive-learning-based method for machine faulty detection and health degradation monitoring. The kernel of the proposed method is an “evolving” model that uses an unsupervised online learning scheme, in which an adaptive hidden Markov model (AHMM) is used for online learning the dynamic health changes of machines in their full life. A statistical index is developed for recognizing the new health states in the machines. Those new health states are then described online by adding of new hidden states in AHMM. Furthermore, the health degradations in machines are quantified online by an AHMM-based health index (HI) that measures the similarity between two density distributions that describe the historic and current health states, respectively. When necessary, the proposed method characterizes the distinct operating modes of the machine and can learn online both abrupt as well as gradual health changes. Our method overcomes some drawbacks of the HIs (e.g., relatively low comprehensibility and applicability) based on fixed monitoring models constructed in the offline phase. Results from its application in a bearing life test reveal that the proposed method is effective in online detection and adaptive assessment of machine health degradation. This study provides a useful guide for developing a condition-based maintenance (CBM) system that uses an online learning method without considerable human intervention.

A HMM text classification model with learning capacity

In this paper a method of classifying biomedical text documents based on Hidden Markov Model is proposed and evaluated. The method is integrated into a framework named BioClass. Bioclass is composed of intelligent text classification tools and facilitates the comparison between them because it has several views of the results. The main goal is to propose a more effective based-on content classifier than current methods in this environment To test the effectiveness of the classifier presented, a set of experiments performed on the OSHUMED corpus are preseted. Our model is tested adding it learning capacity and without it, and it is compared with other classification techniques. The results suggest that the adaptive HMM model is indeed more suitable for document classification.

An adaptive hidden Markov model-based gesture recognition approach using Kinect to simplify large-scale video data processing for humanoid robot imitation

Human gesture recognition to be a new type of natural user interface (NUI) using the person's gesture action for operating the device is attracting much attention nowadays. In this study, an adaptive hidden Markov model (HMM)-based gesture recognition method with user adaptation (UA) using the Kinect camera to simplify large-scale video processing is designed to be the NUI of a humanoid robot device. The popular Kinect camera is employed for acquiring the gesture signals made by the active user, and the gesture action from the user can then be recognized and used to be as the control command for driving the humanoid robot to imitate the user's actions. The large-scale video data can be reduced by the Kinect camera where the data from the Kinect camera for representing gesture signals includes the depth measurement information, and therefore only simple 3-axis coordinate information of the joints in a human skeleton is analyzed, categorized and managed in the developed system. By the presented scheme, the humanoid robot will imitate the human active gesture according to the content of the received gesture command. The well-known HMM pattern recognition method with the support of the Kinect device is explored to classify the human's active gestures where a user adaptation scheme of MAP+GoSSRT that enhances MAP by incorporating group of states shifted by referenced transfer (GoSSRT) is proposed for adjusting HMM parameters, which will further increase the recognition accuracy of HMM gesture recognition. Human gesture recognition experiments for controlling the activity of the humanoid robot were performed on the indicated 14 classes of human active gestures. Experimental results demonstrated the superiority of the NUI by presented HMM gesture recognition with user adaptation for humanoid robot imitation applications.

Adaptive HMM based Speech Recognition to Recognize Multi-lingual Sentence

Hidden Markov Models (HMMs) provides an effective framework for the modeling of time-varying sequence of spectral vector. An approach of mapping signal to discrete signal is to define it as a set of acoustic featured symbol over a minimal but constant time interval. The aim of proposing this paper is to recognize the speech sample using hidden markov model (HMM) with the use of cepstrum feature of our given speech sample within an adaptive interval of time for which pitch period is determined and divides the sample in accordance with this period. Secondly the phonetics or exact “pronunciation” of the word needs to be defined. These are established by associated rule probability where probability is done on word‟s pronunciation. General Terms Multilingual Speech Recognition

Real-Time Landmine Detection with Ground-Penetrating Radar Using Discriminative and Adaptive Hidden Markov Models

We propose a real-time software system for landmine detection using ground-penetrating radar (GPR). The system includes an efficient and adaptive preprocessing component; a hidden Markov model- (HMM-) based detector; a corrective training component; and an incremental update of the background model. The preprocessing is based on frequency-domain processing and performs ground-level alignment and background removal. The HMM detector is an improvement of a previously proposed system (baseline). It includes additional pre- and postprocessing steps to improve the time efficiency and enable real-time application. The corrective training component is used to adjust the initial model parameters to minimize the number of misclassification sequences. This component could be used offline, or online through feedback to adapt an initial model to specific sites and environments. The background update component adjusts the parameters of the background model to adapt it to each lane during testing. The proposed software system is applied to data acquired from three outdoor test sites at different geographic locations, using a state-of-the-art array GPR prototype. The first collection was used as training, and the other two (contain data from more than 1200 m of simulated dirt and gravel roads) for testing. Our results indicate that, on average, the corrective training can improve the performance by about 10% for each site. For individual lanes, the performance gain can reach 50%.

On adaptive HMM state estimation

New online adaptive hidden Markov model (HMM) state estimation schemes are developed, based on extended least squares (ELS) concepts and recursive prediction error (RPE) methods. The best of the new schemes exploit the idempotent nature of Markov chains and work with a least squares prediction error index, using a posterior estimates, more suited to Markov models than traditionally used in identification of linear systems. These new schemes learn the set of N Markov chain states, and the a posteriori probability of being in each of the states at each time instant. They are designed to achieve the strengths, in terms of computational effort and convergence rates, of each of the two classes of earlier proposed adaptive HMM schemes without the weaknesses of each in these areas. The computational effort is of order N. Implementation aspects of the proposed algorithms are discussed, and simulation studies are presented to illustrate convergence rates in comparison to earlier proposed online schemes.

An adaptive hidden Markov model approach to FM and M-ary DPSK demodulation in noisy fading channels

In this paper extended Kalman filtering (EKF) and hidden Markov model (HMM) signal processing techniques are coupled in order to demodulate frequency modulated signals in noisy fading channels. The demodulation scheme presented is applied to both digital M-ary differential phase shift keyed (MDPSK) and analog frequency modulated (FM) signals. Adaptive state-and-parameter estimation schemes are devised based on the assumption that the transmission channel introduces time-varying gain-and-phase changes, modelled by a stochastic linear system, and has additive Gaussian noise. An adaptive HMM approach is formulated which consists of a continuous state Kalman filter (KF) coupled with finite-discrete state HMM filters. The technique used is to represent MDPSK and FM signals with state space signal models for which the KF/HMM coupled filters are derived. A key to this approach is that complete information-states are used, instead of the maximum a posteriori estimates of the traditional matched filter approach, or maximum likelihood estimates of the Viterbi algorithm. The case of white observation noise is considered, as well as a generalisation to cope with coloured noise. Simulation studies are also presented.