Hidden Markov Model Training Research Articles

Maximum margin hidden Markov models for sequence classification

Discriminative learning methods are known to work well in pattern classification tasks and often show benefits compared to generative learning. This is particularly true in case of model mismatch, i.e. the model cannot represent the true data distribution. In this paper, we derive discriminative maximum margin learning for hidden Markov models (HMMs) with emission probabilities represented by Gaussian mixture models (GMMs). The focus is on single-label sequence classification where the margin objective is specified by the probabilistic gap between the true class and the most competing class. In particular, we use the extended Baum–Welch (EBW) framework to optimize this probabilistic margin embedded in a hinge loss function. Approximations of the margin objective and the derivatives are necessary. In the experiments, we compare maximum margin HMMs to generative maximum likelihood and discriminative conditional log-likelihood (CLL) HMM training. We present results of classifying trajectories of handwritten characters, Australian sign language data, digits of speech data and UCR time-series data. Maximum margin HMMs outperform in many cases CLL-HMMs. Furthermore, maximum margin HMMs achieve a significantly better performance than generative maximum likelihood HMMs.

An intelligent performance degradation assessment method for bearings

Bearings are one of the most frequently used components in the rotatory machinery, so the performance degradation assessment of bearings plays an important role in the prognostics and health management of systems. Hidden Markov model (HMM) is a widely applied data-driven model used for bearing performance degradation assessment and has many successful applications. A normal HMM needs to be trained in advance, which has close relationship with the evaluation system. However, the trained HMM is quite influenced by many issues, such as the data integrity and the feature space. In this paper, an intelligent bearing performance degradation assessment method based on HMM and nuisance attribute projection (NAP) is proposed. The proposed method can combine the information from the experimental data and the real-time data effectively and assess the performance since the beginning of the monitoring. The effectiveness of the proposed method is verified through an accelerated life test of rolling element bearings.

A comparison of static, dynamic, and hybrid analysis for malware detection

In this research, we compare malware detection techniques based on static, dynamic, and hybrid analysis. Specifically, we train Hidden Markov Models (HMMs ) on both static and dynamic feature sets and compare the resulting detection rates over a substantial number of malware families. We also consider hybrid cases, where dynamic analysis is used in the training phase, with static techniques used in the detection phase, and vice versa. In our experiments, a fully dynamic approach generally yields the best detection rates. We discuss the implications of this research for malware detection based on hybrid techniques.

Detection and classification of seismic events with progressive multi-channel correlation and hidden Markov models

The progressive multi-channel correlation (PMCC) detector and a statistical time series classifier based on hidden Markov models (HMMs) are associated in order to detect and classify seismic events at an array station. We classify the signal patterns that have triggered PMCC detections thanks to HMMs. The signal is transformed into sequences of feature vectors, which are used to train HMMs. The features include outputs from the PMCC detector describing the seismic wave propagation. Our method is applied to the automatic classification of regional seismic event, teleseismic event, and noise. The experimental data set comprises records from the Songino array station in Mongolia. Results on test events show an average classification performance of 84%.

Combining intrinsic disorder prediction and augmented training of hidden Markov models improves discriminative motif discovery

Identifying short linear motifs (SLiMs) usually suffers from lack of sufficient sequences. SLiMs with the same functional site class are typically characterized by similar motif patterns, which makes them hard to distinguish by generative motif discovery methods. A discriminative method based on maximal mutual information estimation (MMIE) of hidden Markov models (HMMs) is proposed. It masks ordered regions to improve signal to noise ratio and augments the training set to diminish the impact of the lack of sequences. Experimental results on a dataset selected from the Eukaryotic Linear Motif (ELM) resource show that the proposed method is effective and practical.

HMM 기반 TTS와 MusicXML을 이용한 노래음 합성

노래음 합성이란 주어진 가사와 악보를 이용하여 컴퓨터에서 노래음을 생성하는 것이다. 텍스트/음성 변환기에 널리 사용된 HMM 기반 음성합성기는 최근 노래음 합성에도 적용되고 있다. 그러나 기존의 구현방법에는 대용량의 노래음 데이터베이스 수집과 학습이 필요하여 구현에 어려움이 있다. 또한 기존의 상용 노래음 합성시스템은 피아노 롤 방식의 악보 표현방식을 사용하고 있어 일반인에게는 익숙하지 않으므로 읽기 쉬운 표준 악보형식의 사용자 인터페이스를 지원하여 노래 학습의 편의성을 향상시킬 필요가 있다. 이 문제를 해결하기 위하여 본 논문에서는 기존 낭독형 음성합성기의 HMM 모델을 이용하고 노래음에 적합한 피치값과 지속시간 제어방법을 적용하여 HMM 모델 파라미터 값을 변화시킴으로서 노래음을 생성하는 방법을 제안한다. 그리고 음표와 가사를 입력하기 위한 MusicXML 기반의 악보편집기를 전단으로, HMM 기반의 텍스트/음성 변환 합성기를 합성기 후단으로서 사용하여 노래음 합성시스템을 구현하는 방법을 제안한다. 본 논문에서 제안하는 방법을 이용하여 합성된 노래음을 평가하였으며 평가결과 활용 가능성을 확인하였다. Singing voice synthesis is the generation of a song using a computer given its lyrics and musical notes. Hidden Markov models (HMM) have been proved to be the models of choice for text to speech synthesis. HMMs have also been used for singing voice synthesis research, however, a huge database is needed for the training of HMMs for singing voice synthesis. And commercially available singing voice synthesis systems which use the piano roll music notation, needs to adopt the easy to read standard music notation which make it suitable for singing learning applications. To overcome this problem, we use a speech database for training context dependent HMMs, to be used for singing voice synthesis. Pitch and duration control methods have been devised to modify the parameters of the HMMs trained on speech, to be used as the synthesis units for the singing voice. This work describes a singing voice synthesis system which uses a MusicXML based music score editor as the front-end interface for entry of the notes and lyrics to be synthesized and a hidden Markov model based text to speech synthesis system as the back-end synthesizer. A perceptual test shows the feasibility of our proposed system.

Discovering short linear protein motif based on selective training of profile hidden Markov models

Short linear motifs (SLiMs) in proteins are relatively conservative sequence patterns within disordered regions of proteins, typically 3–10 amino acids in length. They play an important role in mediating protein–protein interactions. Discovering SLiMs by computational methods has attracted more and more attention, most of which were based on regular expressions and profiles. In this paper, a de novo motif discovery method was proposed based on profile hidden Markov models (HMMs), which can not only provide the emission probabilities of amino acids in the defined positions of SLiMs, but also model the undefined positions. We adopted the ordered region masking and the relative local conservation (RLC) masking to improve the signal to noise ratio of the query sequences while applying evolutionary weighting to make the important sequences in evolutionary process get more attention by the selective training of profile HMMs. The experimental results show that our method and the profile-based method returned different subsets within a SLiMs dataset, and the performance of the two approaches are equivalent on a more realistic discovery dataset. Profile HMM-based motif discovery methods complement the existing methods and provide another way for SLiMs analysis.

Subsequent application of self-organizing map and hidden Markov models infer community states of stream benthic macroinvertebrates

Because an ecological community consists of diverse species that vary nonlinearly with environmental variability, its dynamics are complex and difficult to analyze. To investigate temporal variations of benthic macroinvertebrate com munity, we used the community data that were collected at the sampling site in Baenae Stream near Busan, Korea, which is a clean stream with minimum pollution, from July 2006 to July 2013. First, we used a self-organizing map (SOM) to heuristically derive the states that characterizes the biotic condition of the benthic macroinvertebrate communities in forms of time series data. Next, we applied the hidden Markov model (HMM) to fine-tune the states objectively and to obtain the transition probabilities between the states and the emission probabilities that show the connection of the states with observable events such as the number of species, the diversity measured by Shannon entropy, and the biological water quality index (BMWP). While the number of species apparently addressed the state of the community, the diversity reflected the state changes after the HMM training along with seasonal variations in cyclic manners. The BMWP showed clear characterization of events that correspond to the different states based on the emission probabilities. The environmental factors such as temperature and precipitation also indicated the seasonal and cyclic changes according to the HMM. Though the usage of the HMM alone can guarantee the convergence of the training or the precision of the derived states based on field data in this study, the derivation of the states by the SOM that followed the fine-tuning by the HMM well elucidated the states of the community and could serve as an alternative reference system to reveal the ecological structures in stream communities.

Fault Detection, Identification, and Location in Smart Grid Based on Data-Driven Computational Methods

A fault detection, identification, and location approach is proposed and studied in this paper. This approach is based on matching pursuit decomposition (MPD) using Gaussian atom dictionary, hidden Markov model (HMM) of real-time frequency and voltage variation features, and fault contour maps generated by machine learning algorithms in smart grid (SG) systems. Specifically, the time-frequency features are extracted by MPD from the frequency and voltage signals, which are sampled by the frequency disturbance recorders in SG. A hybrid clustering algorithm is then developed and used to cluster the frequency and voltage signal features into various symbols. Using the symbols, two detection HMMs are trained for fault detection to distinguish between normal and abnormal SG operation conditions. Also, several identification HMMs are trained under different system fault scenarios, and if a fault occurs, the trained identification HMMs are used to identify different fault types. In the meantime, if the fault is detected by the detection HMMs, a fault contour map will be generated using the feature extracted by the MPD from the voltage signals and topology information of SG. The numerical results demonstrate the feasibility, effectiveness, and accuracy of the proposed approach for the diagnosis of various types of faults with different measurement signal-to-noise ratios in SG systems.

Hybrid Metaheuristic Approaches to the Expectation Maximization for Estimation of the Hidden Markov Model for Signal Modeling

The expectation maximization (EM) is the standard training algorithm for hidden Markov model (HMM). However, EM faces a local convergence problem in HMM estimation. This paper attempts to overcome this problem of EM and proposes hybrid metaheuristic approaches to EM for HMM. In our earlier research, a hybrid of a constraint-based evolutionary learning approach to EM (CEL-EM) improved HMM estimation. In this paper, we propose a hybrid simulated annealing stochastic version of EM (SASEM) that combines simulated annealing (SA) with EM. The novelty of our approach is that we develop a mathematical reformulation of HMM estimation by introducing a stochastic step between the EM steps and combine SA with EM to provide better control over the acceptance of stochastic and EM steps for better HMM estimation. We also extend our earlier work and propose a second hybrid which is a combination of an EA and the proposed SASEM, (EA-SASEM). The proposed EA-SASEM uses the best constraint-based EA strategies from CEL-EM and stochastic reformulation of HMM. The complementary properties of EA and SA and stochastic reformulation of HMM of SASEM provide EA-SASEM with sufficient potential to find better estimation for HMM. To the best of our knowledge, this type of hybridization and mathematical reformulation have not been explored in the context of EM and HMM training. The proposed approaches have been evaluated through comprehensive experiments to justify their effectiveness in signal modeling using the speech corpus: TIMIT. Experimental results show that proposed approaches obtain higher recognition accuracies than the EM algorithm and CEL-EM as well.

Extracting bibliographical data for PDF documents with HMM and external resources

Purpose – The purpose of this paper is to propose an automatic metadata extraction and retrieval system to extract bibliographical information from digital academic documents in portable document formats (PDFs). Design/methodology/approach – The authors use PDFBox to extract text and font size information, a rule-based method to identify titles, and an Hidden Markov Model (HMM) to extract the titles and authors. Finally, the extracted titles and authors (possibly incorrect or incomplete) are sent as query strings to digital libraries (e.g. ACM, IEEE, CiteSeerX, SDOS, and Google Scholar) to retrieve the rest of metadata. Findings – Four experiments are conducted to examine the feasibility of the proposed system. The first experiment compares two different HMM models: multi-state model and one state model (the proposed model). The result shows that one state model can have a comparable performance with multi-state model, but is more suitable to deal with real-world unknown states. The second experiment shows that our proposed model (without the aid of online query) can achieve as good performance as other researcher's model on Cora paper header dataset. In the third experiment the paper examines the performance of our system on a small dataset of 43 real PDF research papers. The result shows that our proposed system (with online query) can perform pretty well on bibliographical data extraction and even outperform the free citation management tool Zotero 3.0. Finally, the paper conducts the fourth experiment with a larger dataset of 103 papers to compare our system with Zotero 4.0. The result shows that our system significantly outperforms Zotero 4.0. The feasibility of the proposed model is thus justified. Research limitations/implications – For academic implication, the system is unique in two folds: first, the system only uses Cora header set for HMM training, without using other tagged datasets or gazetteers resources, which means the system is light and scalable. Second, the system is workable and can be applied to extracting metadata of real-world PDF files. The extracted bibliographical data can then be imported into citation software such as endnote or refworks to increase researchers’ productivity. Practical implications – For practical implication, the system can outperform the existing tool, Zotero v4.0. This provides practitioners good chances to develop similar products in real applications; though it might require some knowledge about HMM implementation. Originality/value – The HMM implementation is not novel. What is innovative is that it actually combines two HMM models. The main model is adapted from Freitag and Mccallum (1999) and the authors add word features of the Nymble HMM (Bikel et al, 1997) to it. The system is workable even without manually tagging the datasets before training the model (the authors just use cora dataset to train and test on real-world PDF papers), as this is significantly different from what other works have done so far. The experimental results have shown sufficient evidence about the feasibility of our proposed method in this aspect.

Investigations on an EM-Style Optimization Algorithm for Discriminative Training of HMMs

Today's speech recognition systems are based on hidden Markov models (HMMs) with Gaussian mixture models whose parameters are estimated using a discriminative training criterion such as Maximum Mutual Information (MMI) or Minimum Phone Error (MPE). Currently, the optimization is almost always done with (empirical variants of) Extended Baum-Welch (EBW). This type of optimization requires sophisticated update schemes for the step sizes and a considerable amount of parameter tuning, and only little is known about its convergence behavior. In this paper, we derive an EM-style algorithm for discriminative training of HMMs. Like Expectation-Maximization (EM) for the generative training of HMMs, the proposed algorithm improves the training criterion on each iteration, converges to a local optimum, and is completely parameter-free. We investigate the feasibility of the proposed EM-style algorithm for discriminative training of two tasks, namely grapheme-to-phoneme conversion and spoken digit string recognition.

Multiple Observations HMM Learning by Aggregating Ensemble Models

Hidden Markov Model (HMM) is a widespread statistical model used in cases where the system involves not fully observable data sequences such as temporal pattern recognition and signal processing. The most difficult problem in dealing with HMMs is the training procedure, or parameter learning, for which several approaches has been proposed. Nevertheless, these methods suffer from trapping in local maxima and still no tractable algorithm is present to overcome this problem. On the other hand, good performances of ensemble methods, where multiple models are employed to obtain the target model, lead to considering ensemble learning in the HMM training problem. Until now, just a few ensemble methods have been proposed for HMMs which lack strong theoretical background, or do not involve all the basic models to construct the final model. Hence in this paper a new ensemble learning method for HMMs is proposed which takes advantage of information theory measures, specifically Rényi entropy, and addresses the mentioned problems of previous methods. In agreement with this claim, the results show superiority of the proposed method over other compared methods for both synthetic and real-world datasets. Besides, the proposed ensemble method succeeded to meet the performance of other methods with much lower required training samples.

Fusion of HMM Classifiers, Based on x, y and (x, y) Signals, for the Recognition of Online Farsi Handwriting: a Large Lexicon Approach

Online handwriting is basically formed by a combination of horizontal and vertical trajectories. We employed decision fusion of semi-continuous hidden Markov model (HMM), classifiers, designed for x, y and (x, y) signals separately, to increase the recognition rate of online Farsi subwords. Because of large lexicon and few training samples for some classes, an embedded method was used to train HMM models of subwords. The result of segmentation of (x, y) signal was exploited to segment x and y signals to letters or sub-letters in the training phase. Different decision fusions of the three classifiers were investigated. The best result was achieved by applying the product rule to combine y and (x, y) classifiers.

Reasoning-Based Framework for Driving Safety Monitoring Using Driving Event Recognition

With the growing concern for driving safety, many driving-assistance systems have been developed. In this paper, we develop a reasoning-based framework for the monitoring of driving safety. The main objective is to present drivers with an intuitively understood green/yellow/red indicator of their danger level. Because the danger level may change owing to the interaction of the host vehicle and the environment, the proposed framework involves two stages of danger-level alerts. The first stage collects lane bias, the distance to the front car, longitudinal and lateral accelerations, and speed data from sensors installed in a real vehicle. All data were recorded in a normal driving environment for the training of hidden Markov models of driving events, including normal driving, acceleration, deceleration, changing to the left or right lanes, zigzag driving, and approaching the car in front. In addition to recognizing these driving events, the degree of each event is estimated according to its character. In the second stage, the danger-level indicator, which warns the driver of a dangerous situation, is inferred by fuzzy logic rules that address the recognized driving events and their degrees. A hierarchical decision strategy is also designed to reduce the number of rules that are triggered. The proposed framework was successfully implemented on a TI DM3730-based embedded platform and was fully evaluated in a real road environment. The experimental results achieved a detection ratio of 99 % for event recognition, compared with that achieved by four conventional methods.

The Learning Algorithms of Coupled Discrete Hidden Markov Models

A hidden Markov model (HMM) encompasses a large class of stochastic process models and has been successfully applied to a number of scientific and engineering problems, including speech and other pattern recognition problems, and biological sequence analysis. A major restriction is found, however, in conventional HMM, i.e., it is ill-suited to capture the interactions among different models. A variety of coupled hidden Markov models (CHMMs) have recently been proposed as extensions of HMM to better characterize multiple interdependent sequences. The resulting models have multiple state variables that are temporally coupled via matrices of conditional probabilities. This paper study is focused on the coupled discrete HMM, there are two state variables in the network. By generalizing forward-backward algorithm, Viterbi algorithm and Baum-Welch algorithm commonly used in conventional HMM to accommodate two state variables, several new formulae solving the 2-chain coupled discrete HMM probability evaluation, decoding and training problem are theoretically derived.

Optimizing multiple sequence alignments using a genetic algorithm based on three objectives: structural information, non-gaps percentage and totally conserved columns

Multiple sequence alignments (MSAs) are widely used approaches in bioinformatics to carry out other tasks such as structure predictions, biological function analyses or phylogenetic modeling. However, current tools usually provide partially optimal alignments, as each one is focused on specific biological features. Thus, the same set of sequences can produce different alignments, above all when sequences are less similar. Consequently, researchers and biologists do not agree about which is the most suitable way to evaluate MSAs. Recent evaluations tend to use more complex scores including further biological features. Among them, 3D structures are increasingly being used to evaluate alignments. Because structures are more conserved in proteins than sequences, scores with structural information are better suited to evaluate more distant relationships between sequences. The proposed multiobjective algorithm, based on the non-dominated sorting genetic algorithm, aims to jointly optimize three objectives: STRIKE score, non-gaps percentage and totally conserved columns. It was significantly assessed on the BAliBASE benchmark according to the Kruskal-Wallis test (P < 0.01). This algorithm also outperforms other aligners, such as ClustalW, Multiple Sequence Alignment Genetic Algorithm (MSA-GA), PRRP, DIALIGN, Hidden Markov Model Training (HMMT), Pattern-Induced Multi-sequence Alignment (PIMA), MULTIALIGN, Sequence Alignment Genetic Algorithm (SAGA), PILEUP, Rubber Band Technique Genetic Algorithm (RBT-GA) and Vertical Decomposition Genetic Algorithm (VDGA), according to the Wilcoxon signed-rank test (P < 0.05), whereas it shows results not significantly different to 3D-COFFEE (P > 0.05) with the advantage of being able to use less structures. Structural information is included within the objective function to evaluate more accurately the obtained alignments. The source code is available at http://www.ugr.es/~fortuno/MOSAStrE/MO-SAStrE.zip.

Classification of Multiple Seizure-Like States in Three Different Rodent Models of Epileptogenesis

Epilepsy is a dynamical disease and its effects are evident in over fifty million people worldwide. This study focused on objective classification of the multiple states involved in the brain's epileptiform activity. Four datasets from three different rodent hippocampal preparations were explored, wherein seizure-like-events (SLE) were induced by the perfusion of a low - Mg(2+) /high-K(+) solution or 4-Aminopyridine. Local field potentials were recorded from CA3 pyramidal neurons and interneurons and modeled as Markov processes. Specifically, hidden Markov models (HMM) were used to determine the nature of the states present. Properties of the Hilbert transform were used to construct the feature spaces for HMM training. By sequentially applying the HMM training algorithm, multiple states were identified both in episodes of SLE and nonSLE activity. Specifically, preSLE and postSLE states were differentiated and multiple inner SLE states were identified. This was accomplished using features extracted from the lower frequencies (1-4 Hz, 4-8 Hz) alongside those of both the low- (40-100 Hz) and high-gamma (100-200 Hz) of the recorded electrical activity. The learning paradigm of this HMM-based system eliminates the inherent bias associated with other learning algorithms that depend on predetermined state segmentation and renders it an appropriate candidate for SLE classification.

HMM-Based Human Fall Detection and Prediction Method Using Tri-Axial Accelerometer

Falls in the elderly have always been a serious medical and social problem. To detect and predict falls, a hidden Markov model (HMM)-based method using tri-axial accelerations of human body is proposed. A wearable motion detection device using tri-axial accelerometer is designed and realized, which can detect and predict falls based on tri-axial acceleration of human upper trunk. The acceleration time series (ATS) extracted from human motion processes are used to describe human motion features, and the ATS extracted from human fall courses but before the collision are used to train HMM so as to build a random process mathematical model. Thus, the outputs of HMM, which express the marching degrees of input ATS and HMM, can be used to evaluate the risks to fall. The experiment results show that fall events can be predicted 200–400 ms ahead the occurrence of collisions, and distinguished from other daily life activities with an accuracy of 100%.

Thai sign language translation using Scale Invariant Feature Transform and Hidden Markov Models

Visual communication is important for a deft and/or mute person. It is also one of the tools for the communication between human and machines. In this paper, we develop an automatic Thai sign language translation system that is able to translate sign language that is not finger-spelling sign language. In particular, we utilize Scale Invariant Feature Transform (SIFT) to match a test frame with observation symbols from keypoint descriptors collected in the signature library. These keypoint descriptors are computed from several keyframes that are recorded at different times of day for several days from five subjects. Hidden Markov Models (HMMs) are then used to translate observation sequences to words. We also collect Thai sign language videos from 20 subjects for testing. The system achieves approximately 86–95% for signer-dependent on the average, 79.75% for signer-semi-independent (same subjects used in the HMM training only) on the average and 76.56% for signer-independent on the average. These results are from the constrained system in which each signer wears a shirt with long sleeves in front of dark background. The unconstrained system in which each signer does not wear a long-sleeve shirt in front of various natural backgrounds yields a good result of around 74% on the average on the signer-independent experiment. The important feature of the proposed system is the consideration of shapes and positions of fingers, in addition to hand information. This feature provides the system ability to recognize the hand sign words that have similar gestures.