scholarly journals Efficient and Effective Learning of HMMs Based on Identification of Hidden States

2017 ◽  
Vol 2017 ◽  
pp. 1-26 ◽  
Author(s):  
Tingting Liu ◽  
Jan Lemeire

The predominant learning algorithm for Hidden Markov Models (HMMs) is local search heuristics, of which the Baum-Welch (BW) algorithm is mostly used. It is an iterative learning procedure starting with a predefined size of state spaces and randomly chosen initial parameters. However, wrongly chosen initial parameters may cause the risk of falling into a local optimum and a low convergence speed. To overcome these drawbacks, we propose to use a more suitable model initialization approach, a Segmentation-Clustering and Transient analysis (SCT) framework, to estimate the number of states and model parameters directly from the input data. Based on an analysis of the information flow through HMMs, we demystify the structure of models and show that high-impact states are directly identifiable from the properties of observation sequences. States having a high impact on the log-likelihood make HMMs highly specific. Experimental results show that even though the identification accuracy drops to 87.9% when random models are considered, the SCT method is around 50 to 260 times faster than the BW algorithm with 100% correct identification for highly specific models whose specificity is greater than 0.06.

Author(s):  
Yaping Li ◽  
Enrico Zio ◽  
Ershun Pan

Degradation is an unavoidable phenomenon in industrial systems. Hidden Markov models (HMMs) have been used for degradation modeling. In particular, segmental HMMs have been developed to model the explicit relationship between degradation signals and hidden states. However, existing segmental HMMs deal only with univariate cases, whereas in real systems, signals from various sensors are collected simultaneously, which makes it necessary to adapt the segmental HMMs to deal with multivariate processes. Also, to make full use of the information from the sensors, it is important to differentiate stable signals from deteriorating ones, but there is no good way for this, especially in multivariate processes. In this paper, the multivariate exponentially weighted moving average (MEWMA) control chart is employed to identify deteriorating multivariate signals. Specifically, the MEWMA statistic is used as a comprehensive indicator for differentiating multivariate observations. Likelihood Maximization is used to estimate the model parameters. To avoid underflow, the forward and backward probabilities are normalized. In order to assess degradation, joint probabilities are defined and derived. Further, the occurrence probability of each degradation state at the current time, as well as in the future, is derived. The Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset of NASA is employed for comparative analysis. In terms of degradation assessment and prediction, the proposed model performs very well in general. By sensitivity analysis, we show that in order to improve further the performance of the method, the weight of the chart should be set relatively small, whereas the method is not sensitive to the change of the in-control average run length (ARL).


2019 ◽  
Vol 46 (4) ◽  
pp. 296
Author(s):  
Victoria L. Goodall ◽  
Sam M. Ferreira ◽  
Paul J. Funston ◽  
Nkabeng Maruping-Mzileni

Context Direct observations of animals are the most reliable way to define their behavioural characteristics; however, to obtain these observations is costly and often logistically challenging. GPS tracking allows finer-scale interpretation of animal responses by measuring movement patterns; however, the true behaviour of the animal during the period of observation is seldom known. Aims The aim of our research was to draw behavioural inferences for a lioness with a hidden Markov model and to validate the predicted latent-state sequence with field observations of the lion pride. Methods We used hidden Markov models to model the movement of a lioness in the Kruger National Park, South Africa. A three-state log-normal model was selected as the most suitable model. The model outputs are related to collected data by using an observational model, such as, for example, a distribution for the average movement rate and/or direction of movement that depends on the underlying model states that are taken to represent behavioural states of the animal. These inferred behavioural states are validated against direct observation of the pride’s behaviour. Key results Average movement rate provided a useful alternative for the application of hidden Markov models to irregularly spaced GPS locations. The movement model predicted resting as the dominant activity throughout the day, with a peak in the afternoon. The local-movement state occurred consistently throughout the day, with a decreased proportion during the afternoon, when more resting takes place, and an increase towards the early evening. The relocating state had three peaks, namely, during mid-morning, early evening and about midnight. Because of the differences in timing of the direct observations and the GPS locations, we had to compare point observations of the true behaviour with an interval prediction of the modelled behavioural state. In 75% of the cases, the model-predicted behaviour and the field-observed behaviour overlapped. Conclusions Our data suggest that the hidden Markov modelling approach is successful at predicting a realistic behaviour of lions on the basis of the GPS location coordinates and the average movement rate between locations. The present study provided a unique opportunity to uncover the hidden states and compare the true behaviour with the inferred behaviour from the predicted state sequence. Implications Our results illustrated the potential of using hidden Markov models with movement rate as an input to understand carnivore behavioural patterns that could inform conservation management practices.


Sensors ◽  
2018 ◽  
Vol 18 (8) ◽  
pp. 2513 ◽  
Author(s):  
Zhenyi Gao ◽  
Bin Zhou ◽  
Bo Hou ◽  
Chao Li ◽  
Qi Wei ◽  
...  

Angle position sensors (APSs) usually require initial calibration to improve their accuracy. This article introduces a novel offline self-calibration scheme in which a signal flow network is employed to reduce the amplitude errors, direct-current (DC) offsets, and phase shift without requiring extra calibration instruments. In this approach, a signal flow network is firstly constructed to overcome the parametric coupling caused by the linearization model and to ensure the independence of the parameters. The model parameters are stored in the nodes of the network, and the intermediate variables are input into the optimization pipeline to overcome the local optimization problem. A deep learning algorithm is also used to improve the accuracy and speed of convergence to a global optimal solution. The results of simulations show that the proposed method can achieve a high identification accuracy with a relative parameter identification error less than 0.001‰. The practical effects were also verified by implementing the developed technique in a capacitive APS, and the experimental results demonstrate that the sensor error after signal calibration could be reduced to only 6.98%.


2013 ◽  
Vol 4 (1) ◽  
Author(s):  
Andrew Critch

This paper closely examines HMMs in which all the hidden random variables arebinary. Its main contributions are (1) a birational parametrization for every such HMM, with anexplicit inverse for recovering the hidden parameters in terms of observables, (2) a semialgebraicmodel membership test for every such HMM, and (3) minimal dening equations for the 4-nodefully binary model, comprising 21 quadrics and 29 cubics, which were computed using Grobnerbases in the cumulant coordinates of Sturmfels and Zwiernik. The new model parameters in (1) arerationally identiable in the sense of Sullivant, Garcia-Puente, and Spielvogel, and each model'sZariski closure is therefore a rational projective variety of dimension 5. Grobner basis computationsfor the model and its graph are found to be considerably faster using these parameters. In thecase of two hidden states, item (2) supersedes a previous algorithm of Schonhuth which is onlygenerically dened, and the dening equations (3) yield new invariants for HMMs of all lengths 4. Such invariants have been used successfully in model selection problems in phylogenetics, andone can hope for similar applications in the case of HMMs.


2017 ◽  
Vol 52 (14) ◽  
pp. 1947-1958 ◽  
Author(s):  
Sergio González ◽  
Gianluca Laera ◽  
Sotiris Koussios ◽  
Jaime Domínguez ◽  
Fernando A Lasagni

The simulation of long life behavior and environmental aging effects on composite materials are subjects of investigation for future aerospace applications (i.e. supersonic commercial aircrafts). Temperature variation in addition to matrix oxidation involves material degradation and loss of mechanical properties. Crack initiation and growth is the main damage mechanism. In this paper, an extended finite element analysis is proposed to simulate damage on carbon fiber reinforced polymer as a consequence of thermal fatigue between −50℃ and 150℃ under atmospheres with different oxygen content. The interphase effect on the degradation process is analyzed at a microscale level. Finally, results are correlated with the experimental data in terms of material stiffness and, hence, the most suitable model parameters are selected.


Entropy ◽  
2021 ◽  
Vol 23 (11) ◽  
pp. 1507
Author(s):  
Feiyu Zhang ◽  
Luyang Zhang ◽  
Hongxiang Chen ◽  
Jiangjian Xie

Deep convolutional neural networks (DCNNs) have achieved breakthrough performance on bird species identification using a spectrogram of bird vocalization. Aiming at the imbalance of the bird vocalization dataset, a single feature identification model (SFIM) with residual blocks and modified, weighted, cross-entropy function was proposed. To further improve the identification accuracy, two multi-channel fusion methods were built with three SFIMs. One of these fused the outputs of the feature extraction parts of three SFIMs (feature fusion mode), the other fused the outputs of the classifiers of three SFIMs (result fusion mode). The SFIMs were trained with three different kinds of spectrograms, which were calculated through short-time Fourier transform, mel-frequency cepstrum transform and chirplet transform, respectively. To overcome the shortage of the huge number of trainable model parameters, transfer learning was used in the multi-channel models. Using our own vocalization dataset as a sample set, it is found that the result fusion mode model outperforms the other proposed models, the best mean average precision (MAP) reaches 0.914. Choosing three durations of spectrograms, 100 ms, 300 ms and 500 ms for comparison, the results reveal that the 300 ms duration is the best for our own dataset. The duration is suggested to be determined based on the duration distribution of bird syllables. As for the performance with the training dataset of BirdCLEF2019, the highest classification mean average precision (cmAP) reached 0.135, which means the proposed model has certain generalization ability.


2020 ◽  
Author(s):  
Jan Münch ◽  
Fabian Paul ◽  
Ralf Schmauder ◽  
Klaus Benndorf

AbstractInferring the complex conformational dynamics of ion channels from ensemble currents is a daunting task due to limited information in the data leading to poorly determined model inference and selection. We address this problem with a parallelized Kalman filter for specifying Hidden Markov Models for current and fluorescence data. We demonstrate the flexibility of this Bayesian network by including different noises distributions. The accuracy of the parameter estimation is increased by tenfold compared to fitting Rate Equations. Furthermore, adding orthogonal fluorescence data increases the accuracy of the model parameters by up to two orders of magnitude. Additional prior information alleviates parameter unidenfiability for weakly informative data. We show that with Rate Equations a reliable detection of the true kinetic scheme requires cross validation. In contrast, our algorithm avoids overfitting by automatically switching of rates (continuous model expansion), by cross-validation, by applying the ‘widely applicable information criterion’ or variance-based model selection.


2020 ◽  
Vol 46 (1) ◽  
pp. 14-19
Author(s):  
Caroline Geraldi Pierozzi ◽  
Ricardo Toshio Fujihara ◽  
Efrain de Santana Souza ◽  
Marília Pizetta ◽  
Maria Márcia Pereira Sartori ◽  
...  

ABSTRACT Interactive keys are tools that aid research and technical work since identification of organisms has become increasingly present in the scientific and academic context. An interactive key was developed with the software Lucid v. 3.3 for the identification of eleven fungal species associated with onion, carrot, pepper and tomato seeds. It was based on a matrix composed of six features: crop, conidium, conidiophore, color of long conidiophore, color of mycelium and presence of setae, besides 21 character states. In addition, descriptions, illustrations and high-resolution photographs of the morphological characters and states were made available to aid in the correct identification of fungal species. Validation of the interactive key was performed by distinct groups of volunteers: (i) graduate students with prior knowledge and using the interactive key; (ii) undergraduate students with little prior knowledge and using the interactive key, and (iii) undergraduate students with little prior knowledge and using the conventional identification system such as the printed manuals used in seed pathology laboratories. We analyzed the time spent by each volunteer to evaluate 25 seeds infected with the fungal species in the key, as well as the percentage of success and the difficulty level for each participant. The high percentage of correct answers with the use of the interactive key and the ease of use by the volunteers confirmed its efficiency because there was an increase in the identification accuracy when compared to the conventional system. Furthermore, the rate of success and the difficulty level presented low variability within groups (i) and (ii). These results are a consequence of the interaction of the user with characteristics of the developed tool, such as high-resolution photographs, which faithfully reproduce the fungal characteristics observed in the seeds under a stereomicroscope. Thus, the interactive key presented here can aid in teaching, institutional and commercial research, inspection and certification of seeds, making diagnosis safer and more accurate. The key is available for free at https://keys.lucidcentral.org/keys/v3/seed_fungi/.


2018 ◽  
Vol 7 (5) ◽  
pp. 120
Author(s):  
T. H. M. Abouelmagd

A new version of the Lomax model is introduced andstudied. The major justification for the practicality of the new model isbased on the wider use of the Lomax model. We are also motivated tointroduce the new model since the density of the new distribution exhibitsvarious important shapes such as the unimodal, the right skewed and the leftskewed. The new model can be viewed as a mixture of the exponentiated Lomaxdistribution. It can also be considered as a suitable model for fitting thesymmetric, left skewed, right skewed, and unimodal data sets. The maximumlikelihood estimation method is used to estimate the model parameters. Weprove empirically the importance and flexibility of the new model inmodeling two types of aircraft windshield lifetime data sets. The proposedlifetime model is much better than gamma Lomax, exponentiated Lomax, Lomaxand beta Lomax models so the new distribution is a good alternative to thesemodels in modeling aircraft windshield data.


Sign in / Sign up

Export Citation Format

Share Document