Search results for: Markov%20parameters

Authors: Roengchai Tansuchat, Woraphon Yamaka, Paravee Maneejuk

Abstract:

From the past recent, the global financial crisis, economic instability, and large fluctuation in agricultural commodity price have led to increased concerns about the volatility transmission among them. The problem is further exacerbated by commodities volatility caused by other commodity price fluctuations, hence the decision on hedging strategy has become both costly and useless. Thus, this paper is conducted to analysis the volatility spillover effect among major agriculture including corn, soybeans, wheat and rice, to help the commodity suppliers hedge their portfolios, and manage the risk and co-volatility of them. We provide a switching regime approach to analyzing the issue of volatility spillovers in different economic conditions, namely upturn and downturn economic. In particular, we investigate relationships and volatility transmissions between these commodities in different economic conditions. We purposed a Copula-based multivariate Markov Switching GARCH model with two regimes that depend on an economic conditions and perform simulation study to check the accuracy of our proposed model. In this study, the correlation term in the cross-hedge ratio is obtained from six copula families – two elliptical copulas (Gaussian and Student-t) and four Archimedean copulas (Clayton, Gumbel, Frank, and Joe). We use one-step maximum likelihood estimation techniques to estimate our models and compare the performance of these copula using Akaike information criterion (AIC) and Bayesian information criteria (BIC). In the application study of agriculture commodities, the weekly data used are conducted from 4 January 2005 to 1 September 2016, covering 612 observations. The empirical results indicate that the volatility spillover effects among cereal futures are different, as response of different economic condition. In addition, the results of hedge effectiveness will also suggest the optimal cross hedge strategies in different economic condition especially upturn and downturn economic.

Keywords: agricultural commodity futures, cereal, cross-hedge, spillover effect, switching regime approach

Procedia PDF Downloads 169

Authors: Faiz N. K. Yusufi, Aquil Ahmed, Jamal Ahmad

Abstract:

Diabetes in India is growing at an alarming rate and the complications caused by it need to be controlled. Coronary heart disease (CHD) is one of the complications that will be discussed for prediction in this study. India has the second most number of diabetes patients in the world. To the best of our knowledge, there is no CHD risk score for Indian type 2 diabetes patients. Any form of CHD has been taken as the event of interest. A sample of 750 was determined and randomly collected from the Rajiv Gandhi Centre for Diabetes and Endocrinology, J.N.M.C., A.M.U., Aligarh, India. Collected variables include patients data such as sex, age, height, weight, body mass index (BMI), blood sugar fasting (BSF), post prandial sugar (PP), glycosylated haemoglobin (HbA1c), diastolic blood pressure (DBP), systolic blood pressure (SBP), smoking, alcohol habits, total cholesterol (TC), triglycerides (TG), high density lipoprotein (HDL), low density lipoprotein (LDL), very low density lipoprotein (VLDL), physical activity, duration of diabetes, diet control, history of antihypertensive drug treatment, family history of diabetes, waist circumference, hip circumference, medications, central obesity and history of CHD. Predictive risk scores of CHD events are designed by cox proportional hazard regression. Model calibration and discrimination is assessed from Hosmer Lemeshow and area under receiver operating characteristic (ROC) curve. Overfitting and underfitting of the model is checked by applying regularization techniques and best method is selected between ridge, lasso and elastic net regression. Youden’s index is used to choose the optimal cut off point from the scores. Five year probability of CHD is predicted by both survival function and Markov chain two state model and the better technique is concluded. The risk scores for CHD developed can be calculated by doctors and patients for self-control of diabetes. Furthermore, the five-year probabilities can be implemented as well to forecast and maintain the condition of patients.

Keywords: coronary heart disease, cox proportional hazard regression, ROC curve, type 2 diabetes Mellitus

Procedia PDF Downloads 189

Authors: Faiz N. K. Yusufi, Aquil Ahmed, Jamal Ahmad

Abstract:

Predicting the risk of diabetic retinopathy (DR) in Indian type 2 diabetes patients is immensely necessary. India, being the second largest country after China in terms of a number of diabetic patients, to the best of our knowledge not a single risk score for complications has ever been investigated. Diabetic retinopathy is a serious complication and is the topmost reason for visual impairment across countries. Any type or form of DR has been taken as the event of interest, be it mild, back, grade I, II, III, and IV DR. A sample was determined and randomly collected from the Rajiv Gandhi Centre for Diabetes and Endocrinology, J.N.M.C., A.M.U., Aligarh, India. Collected variables include patients data such as sex, age, height, weight, body mass index (BMI), blood sugar fasting (BSF), post prandial sugar (PP), glycosylated haemoglobin (HbA1c), diastolic blood pressure (DBP), systolic blood pressure (SBP), smoking, alcohol habits, total cholesterol (TC), triglycerides (TG), high density lipoprotein (HDL), low density lipoprotein (LDL), very low density lipoprotein (VLDL), physical activity, duration of diabetes, diet control, history of antihypertensive drug treatment, family history of diabetes, waist circumference, hip circumference, medications, central obesity and history of DR. Cox proportional hazard regression is used to design risk scores for the prediction of retinopathy. Model calibration and discrimination are assessed from Hosmer Lemeshow and area under receiver operating characteristic curve (ROC). Overfitting and underfitting of the model are checked by applying regularization techniques and best method is selected between ridge, lasso and elastic net regression. Optimal cut off point is chosen by Youden’s index. Five-year probability of DR is predicted by both survival function, and Markov chain two state model and the better technique is concluded. The risk scores developed can be applied by doctors and patients themselves for self evaluation. Furthermore, the five-year probabilities can be applied as well to forecast and maintain the condition of patients. This provides immense benefit in real application of DR prediction in T2DM.

Keywords: Cox proportional hazard regression, diabetic retinopathy, ROC curve, type 2 diabetes mellitus

Procedia PDF Downloads 148

Authors: Taylan Kabbani, Ekrem Duman

Abstract:

The design of adaptive systems that take advantage of financial markets while reducing the risk can bring more stagnant wealth into the global market. However, most efforts made to generate successful deals in trading financial assets rely on Supervised Learning (SL), which suffered from various limitations. Deep Reinforcement Learning (DRL) offers to solve these drawbacks of SL approaches by combining the financial assets price "prediction" step and the "allocation" step of the portfolio in one unified process to produce fully autonomous systems capable of interacting with its environment to make optimal decisions through trial and error. In this paper, a continuous action space approach is adopted to give the trading agent the ability to gradually adjust the portfolio's positions with each time step (dynamically re-allocate investments), resulting in better agent-environment interaction and faster convergence of the learning process. In addition, the approach supports the managing of a portfolio with several assets instead of a single one. This work represents a novel DRL model to generate profitable trades in the stock market, effectively overcoming the limitations of supervised learning approaches. We formulate the trading problem, or what is referred to as The Agent Environment as Partially observed Markov Decision Process (POMDP) model, considering the constraints imposed by the stock market, such as liquidity and transaction costs. More specifically, we design an environment that simulates the real-world trading process by augmenting the state representation with ten different technical indicators and sentiment analysis of news articles for each stock. We then solve the formulated POMDP problem using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, which can learn policies in high-dimensional and continuous action spaces like those typically found in the stock market environment. From the point of view of stock market forecasting and the intelligent decision-making mechanism, this paper demonstrates the superiority of deep reinforcement learning in financial markets over other types of machine learning such as supervised learning and proves its credibility and advantages of strategic decision-making.

Keywords: the stock market, deep reinforcement learning, MDP, twin delayed deep deterministic policy gradient, sentiment analysis, technical indicators, autonomous agent

Procedia PDF Downloads 149

Authors: Chen Yang, Jun Hu, Ping Li, Lili Xue

Abstract:

Most companies pay careful attention to consumer feedback collection, so it is popular to find the ‘feedback’ button of all kinds of mobile apps. Yet it is much more changeling to analyze these feedback texts and to catch the true feelings of a consumer regarding either a problem or a complimentary of consumers who hands out the feedback. Especially to the Chinese content, it is possible that; in one context the Chinese feedback expresses positive feedback, but in the other context, the same Chinese feedback may be a negative one. For example, in Chinese, the feedback 'operating with loudness' works well with both refrigerator and stereo system. Apparently, this feedback towards a refrigerator shows negative feedback; however, the same feedback is positive towards a stereo system. By introducing Bradley, M. and Lang, P.'s Affective Norms for English Text (ANET) theory and Bucci W.’s Referential Activity (RA) theory, we, usability researchers at Pingan, are able to decipher the feedback and to find the hidden feelings behind the content. We subtract 2 disciplines ‘valence’ and ‘dominance’ out of 3 of ANET and 2 disciplines ‘concreteness’ and ‘specificity’ out of 4 of RA to organize our own rating system with a scale of 1 to 5 points. This rating system enables us to judge the feelings/emotion behind each feedback, and it works well with both single word/phrase and a whole paragraph. The result of the rating reflects the strength of the feeling/emotion of the consumer when he/she is typing the feedback. In our daily work, we first require a consumer to answer the net promoter score (NPS) before writing the feedback, so we can determine the feedback is positive or negative. Secondly, we code the feedback content according to company problematic list, which contains 200 problematic items. In this way, we are able to collect the data that how many feedbacks left by the consumer belong to one typical problem. Thirdly, we rate each feedback based on the rating system mentioned above to illustrate the strength of the feeling/emotion when our consumer writes the feedback. In this way, we actually obtain two kinds of data 1) the portion, which means how many feedbacks are ascribed into one problematic item and 2) the severity, how strong the negative feeling/emotion is when the consumer is writing this feedback. By crossing these two, and introducing the portion into X-axis and severity into Y-axis, we are able to find which typical problem gets the high score in both portion and severity. The higher the score of a problem has, the more urgent a problem is supposed to be solved as it means more people write stronger negative feelings in feedbacks regarding this problem. Moreover, by introducing hidden Markov model to program our rating system, we are able to computerize the scoring system and are able to process thousands of feedback in a short period of time, which is efficient and accurate enough for the industrial purpose.

Keywords: computerized scoring system, feeling/emotion of consumer feedback, referential activity, text mining

Procedia PDF Downloads 139

Authors: Amani Tahat, Serdal Kirmizialtin

Abstract:

Trajectory fragment methods such as Markov State Models (MSM), Milestoning (MS) and Transition Path sampling are the prime choice of extending the timescale of all atom Molecular Dynamics simulations. In these approaches, a set of structures that covers the accessible phase space has to be chosen a priori using cluster analysis. Structural clustering serves to partition the conformational state into natural subgroups based on their similarity, an essential statistical methodology that is used for analyzing numerous sets of empirical data produced by Molecular Dynamics (MD) simulations. Local transition kernel among these clusters later used to connect the metastable states using a Markovian kinetic model in MSM and a non-Markovian model in MS. The choice of clustering approach in constructing such kernel is crucial since the high dimensionality of the biomolecular structures might easily confuse the identification of clusters when using the traditional hierarchical clustering methodology. Of particular interest, in the case of MS where the milestones are very close to each other, accurate determination of the milestone identity of the trajectory becomes a challenging issue. Throughout this work we present two cluster analysis methods applied to the cis–trans isomerism of dinucleotide AA. The choice of nucleic acids to commonly used proteins to study the cluster analysis is two fold: i) the energy landscape is rugged; hence transitions are more complex, enabling a more realistic model to study conformational transitions, ii) Nucleic acids conformational space is high dimensional. A diverse set of internal coordinates is necessary to describe the metastable states in nucleic acids, posing a challenge in studying the conformational transitions. Herein, we need improved clustering methods that accurately identify the AA structure in its metastable states in a robust way for a wide range of confused data conditions. The single linkage approach of the hierarchical clustering available in GROMACS MD-package is the first clustering methodology applied to our data. Self Organizing Map (SOM) neural network, that also known as a Kohonen network, is the second data clustering methodology. The performance comparison of the neural network as well as hierarchical clustering method is studied by means of computing the mean first passage times for the cis-trans conformational rates. Our hope is that this study provides insight into the complexities and need in determining the appropriate clustering algorithm for kinetic analysis. Our results can improve the effectiveness of decisions based on clustering confused empirical data in studying conformational transitions in biomolecules.

Keywords: milestoning, self organizing map, single linkage, structure clustering

Procedia PDF Downloads 194

Authors: M. Bahari Mehrabani, Hua-Peng Chen

Abstract:

Management and maintenance of coastal defence structures during the expected life cycle have become a real challenge for decision makers and engineers. Accurate evaluation of the current condition and future performance of flood defence structures is essential for effective practical maintenance strategies on the basis of available field inspection data. Moreover, as coastal defence structures age, it becomes more challenging to implement maintenance and management plans to avoid structural failure. Therefore, condition inspection data are essential for assessing damage and forecasting deterioration of ageing flood defence structures in order to keep the structures in an acceptable condition. The inspection data for flood defence structures are often collected using discrete visual condition rating schemes. In order to evaluate future condition of the structure, a probabilistic deterioration model needs to be utilised. However, existing deterioration models may not provide a reliable prediction of performance deterioration for a long period due to uncertainties. To tackle the limitation, a time-dependent condition-based model associated with a transition probability needs to be developed on the basis of condition grade scheme for flood defences. This paper presents a probabilistic method for predicting future performance deterioration of coastal flood defence structures based on condition grading inspection data and deterioration curves estimated by expert judgement. In condition-based deterioration modelling, the main task is to estimate transition probability matrices. The deterioration process of the structure related to the transition states is modelled according to Markov chain process, and a reliability-based approach is used to estimate the probability of structural failure. Visual inspection data according to the United Kingdom Condition Assessment Manual are used to obtain the initial condition grade curve of the coastal flood defences. The initial curves then modified in order to develop transition probabilities through non-linear regression based optimisation algorithms. The Monte Carlo simulations are then used to evaluate the future performance of the structure on the basis of the estimated transition probabilities. Finally, a case study is given to demonstrate the applicability of the proposed method under no-maintenance and medium-maintenance scenarios. Results show that the proposed method can provide an effective predictive model for various situations in terms of available condition grading data. The proposed model also provides useful information on time-dependent probability of failure in coastal flood defences.

Keywords: condition grading, flood defense, performance assessment, stochastic deterioration modelling

Procedia PDF Downloads 207

Authors: Guntur Bagus Pamungkas

Abstract:

As part of a tropical country that is normally rich in forest land areas, Indonesia has always been in the world's spotlight due to its significantly increasing process of deforestation. In one hand, it is related to the mainstay for maintaining the sustainability of the earth's ecosystem functions. On the other hand, they also cover the various potential sources of the global economy. Therefore, it can always be the target of different scale of investors to excessively exploit them. No wonder the emergence of disasters in various characteristics always comes up. In fact, the deforestation phenomenon does not only occur in various forest land areas in the main islands of Indonesia but also includes Java Island, the most densely populated areas in the world. This island only remains the forest land of about 9.8% of the total forest land in Indonesia due to its long history of it, especially in Central Java Province, the most densely populated area in Java. Again, not surprisingly, this province belongs to the area with the highest frequency of disasters because of it, landslides in particular. One of the areas that often experience it is Banjarnegara District, especially in mountainous areas that lies in the range from 1000 to 3000 meters above sea level, where the remains of land forest area can easyly still be found. Even among them still leaves less untouchable tropical rain forest whose area also covers part of a neighboring district, Pekalongan, which is considered to be the rest of the world's little paradise on Earth. The district's landscape is indeed beautiful, especially in the Dieng area, a major tourist destination in Central Java Province after Borobudur Temple. However, annually hazardous always threatens this district due to this landslide disaster. Even, there was a tragic event that was buried with its inhabitants a few decades ago. This research aims to find part of the concept of effective forest management through monitoring the presence of remaining forest areas in this area. The research implemented monitoring of deforestation rates using the Stochastic Cellular Automata-Markov Chain (SCA-MC) method, which serves to provide a spatial simulation of land use and cover changes (LULCC). This geospatial process uses the Landsat-8 OLI image product with Thermal Infra-Red Sensors (TIRS) Band 10 in 2020 and Landsat 5 TM with TIRS Band 6 in 2010. Then it is also integrated with physical and social geography issues using the QGIS 2.18.11 application with the Mollusce Plugin, which serves to clarify and calculate the area of land use and cover, especially in forest areas—using the LULCC method, which calculates the rate of forest area reduction in 2010-2020 in Banjarnegara District. Since the dependence of this area on the use of forest land is quite high, concepts and preventive actions are needed, such as rehabilitation and reforestation of critical lands through providing proper monitoring and targeted forest management to restore its ecosystem in the future.

Keywords: deforestation, populous area, LULCC method, proper control and effective forest management

Procedia PDF Downloads 107

Authors: Christopher Thornton, Niina Kolehmainen, Kianoush Nazarpour

Abstract:

Background: Accelerometers are widely used to measure physical activity behavior, including in children. The traditional method for processing acceleration data uses cut points, relying on calibration studies that relate the quantity of acceleration to energy expenditure. As these relationships do not generalise across diverse populations, they must be parametrised for each subpopulation, including different age groups, which is costly and makes studies across diverse populations difficult. A data-driven approach that allows physical activity intensity states to emerge from the data under study without relying on parameters derived from external populations offers a new perspective on this problem and potentially improved results. We evaluated the data-driven approach in a diverse population with a range of rapidly evolving physical and mental capabilities, namely very young children (9-38 months old), where this new approach may be particularly appropriate. Methods: We applied an unsupervised machine learning approach (a hidden semi-Markov model - HSMM) to segment and cluster the accelerometer data recorded from 275 children with a diverse range of physical and cognitive abilities. The HSMM was configured to identify a maximum of six physical activity intensity states and the output of the model was the time spent by each child in each of the states. For comparison, we also processed the accelerometer data using published cut points with available thresholds for the population. This provided us with time estimates for each child’s sedentary (SED), light physical activity (LPA), and moderate-to-vigorous physical activity (MVPA). Data on the children’s physical and cognitive abilities were collected using the Paediatric Evaluation of Disability Inventory (PEDI-CAT). Results: The HSMM identified two inactive states (INS, comparable to SED), two lightly active long duration states (LAS, comparable to LPA), and two short-duration high-intensity states (HIS, comparable to MVPA). Overall, the children spent on average 237/392 minutes per day in INS/SED, 211/129 minutes per day in LAS/LPA, and 178/168 minutes in HIS/MVPA. We found that INS overlapped with 53% of SED, LAS overlapped with 37% of LPA and HIS overlapped with 60% of MVPA. We also looked at the correlation between the time spent by a child in either HIS or MVPA and their physical and cognitive abilities. We found that HIS was more strongly correlated with physical mobility (R²HIS =0.5, R²MVPA= 0.28), cognitive ability (R²HIS =0.31, R²MVPA= 0.15), and age (R²HIS =0.15, R²MVPA= 0.09), indicating increased sensitivity to key attributes associated with a child’s mobility. Conclusion: An unsupervised machine learning technique can segment and cluster accelerometer data according to the intensity of movement at a given time. It provides a potentially more sensitive, appropriate, and cost-effective approach to analysing physical activity behavior in diverse populations, compared to the current cut points approach. This, in turn, supports research that is more inclusive across diverse populations.

Keywords: physical activity, machine learning, under 5s, disability, accelerometer

Procedia PDF Downloads 176

Authors: Valentin Père, Mathieu Milhé, Fabien Baillon, Jean-Louis Dirion

Abstract:

Microgrids offer potential solutions to meet the need for local grid stability and increase isolated networks autonomy with the integration of intermittent renewable energy production and storage facilities. In such a context, sizing production and storage for a given network is a complex task, highly depending on input data such as power load profile and renewable resource availability. This work aims at developing an operating cost computation methodology for different microgrid designs based on the use of deep reinforcement learning (RL) algorithms to tackle the optimal operation problem in stochastic environments. RL is a data-based sequential decision control method based on Markov decision processes that enable the consideration of random variables for control at a chosen time scale. Agents trained via RL constitute a promising class of Energy Management Systems (EMS) for the operation of microgrids with energy storage. Microgrid sizing (or design) is generally performed by minimizing investment costs and operational costs arising from the EMS behavior. The latter might include economic aspects (power purchase, facilities aging), social aspects (load curtailment), and ecological aspects (carbon emissions). Sizing variables are related to major constraints on the optimal operation of the network by the EMS. In this work, an islanded mode microgrid is considered. Renewable generation is done with photovoltaic panels; an electrochemical battery ensures short-term electricity storage. The controllable unit is a hydrogen tank that is used as a long-term storage unit. The proposed approach focus on the transfer of agent learning for the near-optimal operating cost approximation with deep RL for each microgrid size. Like most data-based algorithms, the training step in RL leads to important computer time. The objective of this work is thus to study the potential of Batch-Constrained Q-learning (BCQ) for the optimal sizing of microgrids and especially to reduce the computation time of operating cost estimation in several microgrid configurations. BCQ is an off-line RL algorithm that is known to be data efficient and can learn better policies than on-line RL algorithms on the same buffer. The general idea is to use the learned policy of agents trained in similar environments to constitute a buffer. The latter is used to train BCQ, and thus the agent learning can be performed without update during interaction sampling. A comparison between online RL and the presented method is performed based on the score by environment and on the computation time.

Keywords: batch-constrained reinforcement learning, control, design, optimal

Procedia PDF Downloads 91

Authors: John G. Skelly, Peter K. Dearden, Thomas W. R. Harrop, Sarah N. Inwood, Joseph Guhlin

Abstract:

The Argentine stem weevil (ASW; L. bonariensis) is an economically important pasture pest in New Zealand, which causes about $200 million of damage per annum. Microctonus hyperodae (Mh), a parasite of the ASW in its natural range in South America, was introduced into New Zealand to curb the pasture damage caused by the ASW. Mh is an endoparasitic wasp that lays its eggs in the ASW halting its reproduction. Mh was initially successful at preventing ASW proliferation and reducing pasture damage. The effectiveness of Mh has since declined due to decreased parasitism rates and has resulted in increased pasture damage. Although the mechanism through which ASW has developed resistance to Mh has not been discovered, it has been proposed to be due to the different reproductive modes used by Mh and the ASW in New Zealand. The ASW reproduces sexually, whereas Mh reproduces asexually, which has been hypothesised to have allowed the ASW to ‘out evolve’ Mh. Other species within the Microctonus genus reproduce both sexually and asexually. Strains of Microctonus aethiopoides (Ma), a species closely related to Mh, reproduce either by sexual or asexual reproduction. Comparing the genomes of sexual and asexual Microctonus may allow for the identification of the mechanism of asexual reproduction and other characteristics that may improve Mh as a biocontrol agent. The genomes of Mh and three strains of Ma, two of which reproduce sexually and one reproduces asexually, have been sequenced and annotated. The French (MaFR) and Moroccan (MaMO) reproduce sexually, whereas the Irish strain (MaIR) reproduces asexually. Like Mh, The Ma strains are also used as biocontrol agents, but for different weevil species. The genomes of Mh and MaIR were subsequently upgraded using Hi-C, resulting in a set of high quality, highly contiguous genomes. A subset of the genes involved in mitosis and meiosis, which have been identified though the use of Hidden Markov Models generated from genes involved in these processes in other Hymenoptera, have been catalogued in Mh and the strains of Ma. Meiosis and mitosis genes were broadly conserved in both sexual and asexual Microctonus species. This implies that either the asexual species have retained a subset of the molecular components required for sexual reproduction or that the molecular mechanisms of mitosis and meiosis are different or differently regulated in Microctonus to other insect species in which these mechanisms are more broadly characterised. Bioinformatic analysis of the chemoreceptor compliment in Microctonus has revealed some variation in the number of olfactory receptors, which may be related to host preference. Phylogenetic analysis of olfactory receptors highlights variation, which may be able to explain different host range preferences in the Microctonus. Hi-C clustering implies that Mh has 12 chromosomes, and MaIR has 8. Hence there may be variation in gene regulation between species. Genome alignment of Mh and MaIR implies that there may be large scale genome structural variation. Greater insight into the genetics of these agriculturally important group of parasitic wasps may be beneficial in restoring or maintaining their biocontrol efficacy.

Keywords: argentine stem weevil, asexual, genomics, Microctonus hyperodae

Procedia PDF Downloads 127

Authors: Elena Kokkoni, Prasanna Kannappan, Ashkan Zehfroosh, Effrosyni Mavroudi, Kristina Strother-Garcia, James C. Galloway, Jeffrey Heinz, Rene Vidal, Herbert G. Tanner

Abstract:

Early immobility affects the motor, cognitive, and social development. Current pediatric rehabilitation lacks the technology that will provide the dosage needed to promote mobility for young children at risk. The addition of socially assistive robots in early interventions may help increase the mobility dosage. The aim of this study is to examine the feasibility of an early intervention paradigm where non-walking infants experience independent mobility while socially interacting with robots. A dynamic environment is developed where both the child and the robot interact and learn from each other. The environment involves: 1) a range of physical activities that are goal-oriented, age-appropriate, and ability-matched for the child to perform, 2) the automatic functions that perceive the child’s actions through novel activity recognition algorithms, and decide appropriate actions for the robot, and 3) a networked visual data acquisition system that enables real-time assessment and provides the means to connect child behavior with robot decision-making in real-time. The environment was tested by bringing a two-year old boy with Down syndrome for eight sessions. The child presented delays throughout his motor development with the current being on the acquisition of walking. During the sessions, the child performed physical activities that required complex motor actions (e.g. climbing an inclined platform and/or staircase). During these activities, a (wheeled or humanoid) robot was either performing the action or was at its end point 'signaling' for interaction. From these sessions, information was gathered to develop algorithms to automate the perception of activities which the robot bases its actions on. A Markov Decision Process (MDP) is used to model the intentions of the child. A 'smoothing' technique is used to help identify the model’s parameters which are a critical step when dealing with small data sets such in this paradigm. The child engaged in all activities and socially interacted with the robot across sessions. With time, the child’s mobility was increased, and the frequency and duration of complex and independent motor actions were also increased (e.g. taking independent steps). Simulation results on the combination of the MDP and smoothing support the use of this model in human-robot interaction. Smoothing facilitates learning MDP parameters from small data sets. This paradigm is feasible and provides an insight on how social interaction may elicit mobility actions suggesting a new early intervention paradigm for very young children with motor disabilities. Acknowledgment: This work has been supported by NIH under grant #5R01HD87133.

Keywords: activity recognition, human-robot interaction, machine learning, pediatric rehabilitation

Procedia PDF Downloads 269

Authors: B. R. Campomanes-Alvarez, P. Quiros, B. Fernandez

Abstract:

Automatic Speech Recognition (ASR) is a machine-based process of decoding and transcribing oral speech. A typical ASR system receives acoustic input from a speaker or an audio file, analyzes it using algorithms, and produces an output in the form of a text. Some speech recognition systems use Hidden Markov Models (HMMs) to deal with the temporal variability of speech and Gaussian Mixture Models (GMMs) to determine how well each state of each HMM fits a short window of frames of coefficients that represents the acoustic input. Another way to evaluate the fit is to use a feed-forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output. Deep neural networks (DNNs) that have many hidden layers and are trained using new methods have been shown to outperform GMMs on a variety of speech recognition systems. Acoustic models for state-of-the-art ASR systems are usually training on massive amounts of data. However, audio files with their corresponding transcriptions can be difficult to obtain, especially in the Spanish language. Hence, in the case of these low-resource scenarios, building an ASR model is considered as a complex task due to the lack of labeled data, resulting in an under-trained system. Semi-supervised learning approaches arise as necessary tasks given the high cost of transcribing audio data. The main goal of this proposal is to develop a procedure based on acoustic semi-supervised learning for Spanish ASR systems by using DNNs. This semi-supervised learning approach consists of: (a) Training a seed ASR model with a DNN using a set of audios and their respective transcriptions. A DNN with a one-hidden-layer network was initialized; increasing the number of hidden layers in training, to a five. A refinement, which consisted of the weight matrix plus bias term and a Stochastic Gradient Descent (SGD) training were also performed. The objective function was the cross-entropy criterion. (b) Decoding/testing a set of unlabeled data with the obtained seed model. (c) Selecting a suitable subset of the validated data to retrain the seed model, thereby improving its performance on the target test set. To choose the most precise transcriptions, three confidence scores or metrics, regarding the lattice concept (based on the graph cost, the acoustic cost and a combination of both), was performed as selection technique. The performance of the ASR system will be calculated by means of the Word Error Rate (WER). The test dataset was renewed in order to extract the new transcriptions added to the training dataset. Some experiments were carried out in order to select the best ASR results. A comparison between a GMM-based model without retraining and the DNN proposed system was also made under the same conditions. Results showed that the semi-supervised ASR-model based on DNNs outperformed the GMM-model, in terms of WER, in all tested cases. The best result obtained an improvement of 6% relative WER. Hence, these promising results suggest that the proposed technique could be suitable for building ASR models in low-resource environments.

Keywords: automatic speech recognition, deep neural networks, machine learning, semi-supervised learning

Procedia PDF Downloads 316