Добірка наукової літератури з теми "Speech prediction EEG"

Autoregressive processes play a major role in speech processing (linear prediction), seismic signal processing, biological signal processing, and many other applications. We consider the quantity defined by Shannon in 1948, the entropy rate power, and show that the log ratio of entropy powers equals the difference in the differential entropy of the two processes. Furthermore, we use the log ratio of entropy powers to analyze the change in mutual information as the model order is increased for autoregressive processes. We examine when we can substitute the minimum mean squared prediction error for the entropy power in the log ratio of entropy powers, thus greatly simplifying the calculations to obtain the differential entropy and the change in mutual information and therefore increasing the utility of the approach. Applications to speech processing and coding are given and potential applications to seismic signal processing, EEG classification, and ECG classification are described.

Human perception is shaped by past experience on multiple timescales. Sudden and dramatic changes in perception occur when prior knowledge or expectations match stimulus content. These immediate effects contrast with the longer-term, more gradual improvements that are characteristic of perceptual learning. Despite extensive investigation of these two experience-dependent phenomena, there is considerable debate about whether they result from common or dissociable neural mechanisms. Here we test single- and dual-mechanism accounts of experience-dependent changes in perception using concurrent magnetoencephalographic and EEG recordings of neural responses evoked by degraded speech. When speech clarity was enhanced by prior knowledge obtained from matching text, we observed reduced neural activity in a peri-auditory region of the superior temporal gyrus (STG). Critically, longer-term improvements in the accuracy of speech recognition following perceptual learning resulted in reduced activity in a nearly identical STG region. Moreover, short-term neural changes caused by prior knowledge and longer-term neural changes arising from perceptual learning were correlated across subjects with the magnitude of learning-induced changes in recognition accuracy. These experience-dependent effects on neural processing could be dissociated from the neural effect of hearing physically clearer speech, which similarly enhanced perception but increased rather than decreased STG responses. Hence, the observed neural effects of prior knowledge and perceptual learning cannot be attributed to epiphenomenal changes in listening effort that accompany enhanced perception. Instead, our results support a predictive coding account of speech perception; computational simulations show how a single mechanism, minimization of prediction error, can drive immediate perceptual effects of prior knowledge and longer-term perceptual learning of degraded speech.

The efficacy of audiovisual (AV) integration is reflected in the degree of cross-modal suppression of the auditory event-related potentials (ERPs, P1-N1-P2), while stronger semantic encoding is reflected in enhanced late ERP negativities (e.g., N450). We hypothesized that increasing visual stimulus reliability should lead to more robust AV-integration and enhanced semantic prediction, reflected in suppression of auditory ERPs and enhanced N450, respectively. EEG was acquired while individuals watched and listened to clear and blurred videos of a speaker uttering intact or highly-intelligible degraded (vocoded) words and made binary judgments about word meaning (animate or inanimate). We found that intact speech evoked larger negativity between 280–527-ms than vocoded speech, suggestive of more robust semantic prediction for the intact signal. For visual reliability, we found that greater cross-modal ERP suppression occurred for clear than blurred videos prior to sound onset and for the P2 ERP. Additionally, the later semantic-related negativity tended to be larger for clear than blurred videos. These results suggest that the cross-modal effect is largely confined to suppression of early auditory networks with weak effect on networks associated with semantic prediction. However, the semantic-related visual effect on the late negativity may have been tempered by the vocoded signal’s high-reliability.

A brain computer interface is a communication system that translates brain activities into commands for a computer. For physically disabled people, who cannot express their needs through verbal mode (such as thirst, appetite etc), a brain-computer interface (BCI) is the only feasible channel for communicating with others. This technology has the capability of providing substantial independence and hence, a greatly improved quality of life for the physically disabled persons. The BCI technique utilizes electrical brain potentials to directly communicate to devices such as a personal computer system. Cerebral electric activity is recorded via the electroencephalogram (EEG) electrodes attached to the scalp measure the electric signals of the brain. These signals are transmitted to the computer, which transforms them into device control commands. The efficiency of the BCI techniques lies in the extraction of suitable features from EEG signals followed by the classification scheme. This paper focuses on development of brain-computer interface model for motor imagery tasks such as movement of left hand, right hand etc. Several time domain features namely, spike rhythmicity, autoregressive method by Burgs, auto regression with exogenous input, autoregressive method based on Levinson are used by varying the prediction order. Frequency domain method involving estimation of power spectral density using Welch and Burg’s method are applied. A binary classification based on recurrent neural network is used. An optimal classification of the imagery tasks with an overall accuracy of 100% is achieved based on configuring the neural network model and varying the extracted feature and EEG channels optimally. A device command translator finally converts these tasks into speech thereby providing the practical usage of this model for real-time BCI application.

Speech comprehension requires rapid online processing of a continuous acoustic signal to extract structure and meaning. Previous studies on sentence comprehension have found neural correlates of the predictability of a word given its context, as well as of the precision of such a prediction. However, they have focused on single sentences and on particular words in those sentences. Moreover, they compared neural responses to words with low and high predictability, as well as with low and high precision. However, in speech comprehension, a listener hears many successive words whose predictability and precision vary over a large range. Here, we show that cortical activity in different frequency bands tracks word surprisal in continuous natural speech and that this tracking is modulated by precision. We obtain these results through quantifying surprisal and precision from naturalistic speech using a deep neural network and through relating these speech features to EEG responses of human volunteers acquired during auditory story comprehension. We find significant cortical tracking of surprisal at low frequencies, including the delta band as well as in the higher frequency beta and gamma bands, and observe that the tracking is modulated by the precision. Our results pave the way to further investigate the neurobiology of natural speech comprehension.

Semantically ambiguous words challenge speech comprehension, particularly when listeners must select a less frequent (subordinate) meaning at disambiguation. Using combined magnetoencephalography (MEG) and EEG, we measured neural responses associated with distinct cognitive operations during semantic ambiguity resolution in spoken sentences: (i) initial activation and selection of meanings in response to an ambiguous word and (ii) sentence reinterpretation in response to subsequent disambiguation to a subordinate meaning. Ambiguous words elicited an increased neural response approximately 400–800 msec after their acoustic offset compared with unambiguous control words in left frontotemporal MEG sensors, corresponding to sources in bilateral frontotemporal brain regions. This response may reflect increased demands on processes by which multiple alternative meanings are activated and maintained until later selection. Disambiguating words heard after an ambiguous word were associated with marginally increased neural activity over bilateral temporal MEG sensors and a central cluster of EEG electrodes, which localized to similar bilateral frontal and left temporal regions. This later neural response may reflect effortful semantic integration or elicitation of prediction errors that guide reinterpretation of previously selected word meanings. Across participants, the amplitude of the ambiguity response showed a marginal positive correlation with comprehension scores, suggesting that sentence comprehension benefits from additional processing around the time of an ambiguous word. Better comprehenders may have increased availability of subordinate meanings, perhaps due to higher quality lexical representations and reflected in a positive correlation between vocabulary size and comprehension success.

The process of categorizing sounds into distinct phonetic categories is known as categorical perception (CP). Response times (RTs) provide a measure of perceptual difficulty during labeling decisions (i.e., categorization). The RT is quasi-stochastic in nature due to individuality and variations in perceptual tasks. To identify the source of RT variation in CP, we have built models to decode the brain regions and frequency bands driving fast, medium and slow response decision speeds. In particular, we implemented a parameter optimized convolutional neural network (CNN) to classify listeners’ behavioral RTs from their neural EEG data. We adopted visual interpretation of model response using Guided-GradCAM to identify spatial-spectral correlates of RT. Our framework includes (but is not limited to): (i) a data augmentation technique designed to reduce noise and control the overall variance of EEG dataset; (ii) bandpower topomaps to learn the spatial-spectral representation using CNN; (iii) large-scale Bayesian hyper-parameter optimization to find best performing CNN model; (iv) ANOVA and posthoc analysis on Guided-GradCAM activation values to measure the effect of neural regions and frequency bands on behavioral responses. Using this framework, we observe that α−β (10–20 Hz) activity over left frontal, right prefrontal/frontal, and right cerebellar regions are correlated with RT variation. Our results indicate that attention, template matching, temporal prediction of acoustics, motor control, and decision uncertainty are the most probable factors in RT variation.

The electroencephalogram (EEG) has great attraction in emotion recognition studies due to its resistance to deceptive actions of humans. This is one of the most significant advantages of brain signals in comparison to visual or speech signals in the emotion recognition context. A major challenge in EEG-based emotion recognition is that EEG recordings exhibit varying distributions for different people as well as for the same person at different time instances. This nonstationary nature of EEG limits the accuracy of it when subject independency is the priority. The aim of this study is to increase the subject-independent recognition accuracy by exploiting pretrained state-of-the-art Convolutional Neural Network (CNN) architectures. Unlike similar studies that extract spectral band power features from the EEG readings, raw EEG data is used in our study after applying windowing, pre-adjustments and normalization. Removing manual feature extraction from the training system overcomes the risk of eliminating hidden features in the raw data and helps leverage the deep neural network’s power in uncovering unknown features. To improve the classification accuracy further, a median filter is used to eliminate the false detections along a prediction interval of emotions. This method yields a mean cross-subject accuracy of 86.56% and 78.34% on the Shanghai Jiao Tong University Emotion EEG Dataset (SEED) for two and three emotion classes, respectively. It also yields a mean cross-subject accuracy of 72.81% on the Database for Emotion Analysis using Physiological Signals (DEAP) and 81.8% on the Loughborough University Multimodal Emotion Dataset (LUMED) for two emotion classes. Furthermore, the recognition model that has been trained using the SEED dataset was tested with the DEAP dataset, which yields a mean prediction accuracy of 58.1% across all subjects and emotion classes. Results show that in terms of classification accuracy, the proposed approach is superior to, or on par with, the reference subject-independent EEG emotion recognition studies identified in literature and has limited complexity due to the elimination of the need for feature extraction.

Under adverse listening conditions, speech comprehension profits from the expectancies that listeners derive from the semantic context. However, the neurocognitive mechanisms of this semantic benefit are unclear: How are expectancies formed from context and adjusted as a sentence unfolds over time under various degrees of acoustic degradation? In an EEG study, we modified auditory signal degradation by applying noise-vocoding (severely degraded: four-band, moderately degraded: eight-band, and clear speech). Orthogonal to that, we manipulated the extent of expectancy: strong or weak semantic context (±con) and context-based typicality of the sentence-last word (high or low: ±typ). This allowed calculation of two distinct effects of expectancy on the N400 component of the evoked potential. The sentence-final N400 effect was taken as an index of the neural effort of automatic word-into-context integration; it varied in peak amplitude and latency with signal degradation and was not reliably observed in response to severely degraded speech. Under clear speech conditions in a strong context, typical and untypical sentence completions seemed to fulfill the neural prediction, as indicated by N400 reductions. In response to moderately degraded signal quality, however, the formed expectancies appeared more specific: Only typical (+con +typ), but not the less typical (+con −typ) context–word combinations led to a decrease in the N400 amplitude. The results show that adverse listening “narrows,” rather than broadens, the expectancies about the perceived speech signal: limiting the perceptual evidence forces the neural system to rely on signal-driven expectancies, rather than more abstract expectancies, while a sentence unfolds over time.

This thesis explores how predictions about upcoming linguistic stimuli are generated during real-time language comprehension in younger and older adults. Previous research has shown humans' ability to use rich contextual information to compute linguistic prediction during real-time language comprehension. Research in the modulating factors of prediction has shown, first, that predictions are informed by our experience with language and second, that these predictions are modulated by cognitive factors such as working memory and processing speed. However, little is known about how these factors interact in aging in which verbal intelligence remains stable or even increases, whereas processing speed, working memory, and inhibitory control decline with age. Experience-driven models of language learning argue that learning occurs across the life span instead of terminating once representations are learned well enough to approximate a stable state. In relation to aging, these models predict that older adults are likely to possess stronger learned associations, such that the predictions they generate during on-line processing may be stronger. At the same time, however, processing speed, working memory, and inhibitory control decline as a function of age, and age-related declines in these processes may reduce the degree to which older adults can predict. Here, I explored the interplay between language and cognitive factors in the generation of predictions and hypothesized that older adults will show stronger predictability effects than younger adults likely because of their language experience. In this thesis, I provide evidence from reading eye-movements, event-related potentials (ERPs), and EEG phase synchronization, for the role of language experience and cognitive decline in prediction in younger and older English speakers. I demonstrated that the eye-movement record is influenced by linguistic factors, which produce greater predictability effects as linguistic experience advances, and cognitive factors, which produce smaller predictability effects as they decline. Similarly, the N400, an ERP response that is modulated by a word's predictability, was also moderated by cognitive factors. Most importantly, older adults were able to use context efficiently to facilitate upcoming words in the ERP study, contrary to younger adults. Further, I provide initial evidence that coherence analysis may be used as a measure of cognitive effort to illustrate the facilitation that prediction confers to language comprehenders. The results indicate that for a comprehensive account of predictive processing research needs to take into account the role of experience acquired through lifetime and the declines that aging brings.

La pharmaco-résistance est une complication évolutive fréquente de l'épisode dépressif majeur. La rTMS est une technique de stimulation cérébrale innovante dont l'efficacité antidépressive est maintenant établie.Le premier objectif de notre travail de thèse a été de caractériser les substrats fonctionnels de la dépression pharmaco-résistante à l'aide de la technique TEMP, afin d'identifier des patterns d'anomalies cérébrales qui leur sont propres. Dans un second temps, et sur la base des travaux existant sur les mécanismes d'action de la rTMS, nous avons étudié la valeur prédictive de marqueurs fonctionnels en neuro-imagerie TEMP et par EEG. Enfin, nous avons relié l'effet cérébral de la rTMS révélé par la neuro-imagerie fonctionnelle à son efficacité antidépressive, et de façon plus globale à la qualité de vie, comme recommandé actuellement.Nos résultats montrent l'existence d'un pattern de perfusion cérébrale commun aux patients pharmaco-résistants quel que soit le type de dépression, impliquant les régions fronto-temporales et le cervelet. L'étude TEMP de la perfusion cérébrale et de l'activité cérébrale en l'EEG dans sa bande alpha semble pouvoir prédire de façon satisfaisante, avant traitement, l'amélioration clinique individuelle des patients dépressifs pharmaco-résistants traités par rTMS. L'efficacité antidépressive de la rTMS apparait équivalente quel que soit le côté stimulé, entrainant des modifications de perfusion cérébrale comparables. Enfin, nos résultats ont permis d'identifier des régions cérébrales dysfonctionnelles distinctes et confirment l'interet d'une approche complémentaire de la dépression, par l'évaluation de la qualité de vie
Treatment-resistance is a common outcome of a major depressive episode. Repetitive transcranial magnetic stimulation has been put forward as a new technique to treat this debilitating illness. The first objective of our thesis was to characterize the functional substrates of treatment-resistant depression (TRD) using SPECT technique, in order to identify specific patterns of brain abnormalities. In a second part, based on existing work on the antidepressant mechanisms of rTMS, we investigated the predictive value of two neurofunctional biomarkers: SPECT and EEG. Finally, we studied brain SPECT perfusion changes underlying therapeutic efficiency and improvement of quality of life, as currently recommended. Our results showed the existence of a common pattern of brain perfusion in treatment-resistant patients involving the fronto-temporal regions and the cerebellum, regardless the type of depression. At baseline, SPECT brain perfusion and alpha EEG band power could predict individual clinical improvement in TRD-patients treated with rTMS. Regardless the stimulated side, the antidepressant efficacy of rTMS consisted in similar changes in cerebral perfusion. Finally, our results have identified distinct dysfunctional brain regions and confirm the interest of a complementary approach to depression, by assessing quality of life

AbstractIn operational weather forecasting standard environmental parameters, such as air temperature and humidity, are traditionally used to predict thermal conditions in the future. These parameters, however, are not enough to describe the thermal stress induced by the outdoor environment to the human body as they neglect the human heat budget and personal characteristics (e.g. clothing). The Universal Thermal Climate Index (UTCI) overcomes these limitations by using an advanced thermo-physiological model coupled with a state-of-the-art clothing model. Several systems have been recently developed to operationally forecast human biometeorological conditions via the UTCI, i.e. by computing UTCI from the forecasts of air temperature, humidity, wind speed and radiation as provided by numerical weather prediction models. Here we describe the UTCI-based forecasting systems developed in Czech Republic, Italy, Poland, Portugal and at the pan-European scale. Their characteristics are illustrated and their potential as warning systems for thermal hazards discussed.

Emotion awareness is one of the most important subjects in the field of affective computing. Using nonverbal behavioral methods such as recognition of facial expression, verbal behavioral method, recognition of speech emotion, or physiological signals-based methods such as recognition of emotions based on electroencephalogram (EEG) can predict human emotion. However, it is notable that data obtained from either nonverbal or verbal behaviors are indirect emotional signals suggesting brain activity. Unlike the nonverbal or verbal actions, EEG signals are reported directly from the human brain cortex and thus may be more effective in representing the inner emotional states of the brain. Consequently, when used to measure human emotion, the use of EEG data can be more accurate than data on behavior. For this reason, the identification of human emotion from EEG signals has become a very important research subject in current emotional brain-computer interfaces (BCIs) aimed at inferring human emotional states based on the EEG signals recorded. In this paper, a hybrid deep learning approach has proposed using CNN and a long short-term memory (LSTM) algorithm is investigated for the purpose of automatic classification of epileptic disease from EEG signals. The signals have been processed by CNN for feature extraction from runtime environment while LSTM has used for classification of entire data. Finally, system demonstrates each EEG data file as normal or epileptic disease. In this research to describes a state of art for effective epileptic disease detection prediction and classification using hybrid deep learning algorithms. This research demonstrates a collaboration of CNN and LSTM for entire classification of EEG signals in numerous existing systems.

This chapter surveys two basic tools for bypassing the most severe language and speech deficits: augmentative and alternative communication (AAC) tools and assistive tools. It discusses the various forms of AAC, particularly high-tech speech generating devices. It reviews specific high-tech AAC programs, from those with simpler interfaces that allow only limited communication, to those with complex interfaces that allow longer sentences with grammatical features. It discusses two different strategies for optimizing user interface—category-based strategies (e.g., the Proloquo2Go programs) and motor-planning-based strategies (e.g., the LAMP programs)—and their various pros and cons. It then turns to what the efficacy data shows about boosting communicative skills and behaviors. It then reviews various assistive tools: first, those that transition users from AAC to typing; then, those that only support users through word prediction. It concludes with caveats about how AAC and assistive tools should not de-incentive linguistic instruction or undermine independent communication.

The LabVIEW platform with graphical programming environment, will help to integrate the human machine interface controller with the software like MATLAB, Python etc. This platform plays the vital role in many pioneering areas like speech signal processing, bio medical signals like Electrocardiogram (ECG) and Electroencephalogram (EEG) processing, fault analysis in analog electronic circuits, Cognitive Radio(CR), Software Defined Radio (SDR), flexible and wearable electronics. Nowadays most engineering colleges redesign their laboratory curricula for the students to enhance the potential inclusion of remote based laboratory to facilitate and encourage the students to access the laboratory anywhere and anytime. This would help every young learner to bolster their innovation, if the laboratory environment is within the reach of their hand. LabVIEW is widely recognized for its flexibility and adaptability. Due to the versatile nature of LabVIEW in the Input- Output systems, it has find its broad applications in integrated systems. It can provide a smart assistance to deaf and dumb people for interpreting the sign language by gesture recognition using flex sensors, monitor the health condition of elderly people by predicting the abnormalities in the heart beat through remote access, and identify the stage of breast cancer from the Computed tomography (CT) and Magnetic resonance imaging (MRI) scans using image processing techniques. In this chapter, the previous work of authors who have extensively incorporated LabVIEW in the field of electronics and communication are discussed in detail.

Despite the increasing development of wind energy (WE) which has notably become the leading renewable energy now, there are still persistent barriers to wider implementation related to technology, cost, and environmental and social impacts, e.g., super large-scale (≥10 megawatt) wind turbine (WT) technology, WT deployment at low wind speed and/or residential-closed areas, and offshore wind farm under severe wind-wave conditions. As one of the emerging technologies, artificial intelligence (AI) could take some of these challenges and assist in achieving the goal of further lowering the levelized cost of energy from wind. Recently AI has been tremendously investigated in an extremely wide range from financial data analytics and industrial production design to novel biometric and forensic applications and various renewable energy development. This chapter provides the fundamental introduction of four categories of state-of-the-art AI methods including (i) neural learning methods, (ii) statistical learning methods, (iii) evolutionary learning methods, and (iv) hybrid learning methods in a lucid way, then carry out a comprehensive survey of various AI methods used in WE including (i) wind speed prediction, (ii) wind power estimation, (iii) WT design and optimization, and (iv) WT condition monitoring.

In order to facilitate an empirical comparison of the performance of Evolution Strategies, Evolutionary Programming, and Genetic Algorithms, a test environment for these algorithms must be provided in the form of several objective functions f : IRn → IR. Finding an appropriate and representative set of test problems is not an easy task, since any particular combination of properties represented by a test function does not allow for generalized performance statements. However, there is evidence from a vast number of applications that Evolutionary Algorithms are robust in the sense that they give reasonable performance over a wide range of different topologies. Here, a set of test functions that are completely artificial and simple is used, i.e., they are stated in a closed, analytical form and have no direct background from any practical application. Instead, they allow for a detailed analysis of certain special characteristics of the topology, e.g. unimodality or multimodality, continuous or discontinuous cases, and others. If any prediction is drawn up for the behavior of Evolutionary Algorithms depending on such strong topological characteristics, the appropriate idealized test function provides a good instrument to test such hypotheses. Furthermore, since many known test sets have some functions in common, at least a minimal level of comparability of results is often guaranteed. Finally, before we can expect an algorithm to be successful in the case of hard problems, it has to demonstrate that it does not fail to work on simple problems. On the other hand, the (public relations) effect of using artificial topologies is vanishingly small, since the test functions used are of no industrial relevance. This way, researchers working with such test functions can never rest on their industrial laurels. A more legitimate objection against artificial topologies may be that they are possibly not representative of the “average complexity” of real-world problems, and that some regularity features of their topology may inadmissibly speed up the search. However, most test function sets incorporate even multimodal functions of remarkable complexity, such that only the regularity argument counts against using an artifical function set.

A cohort study is one in which the outcome (usually disease status) is ascertained for groups of individuals defined on the basis of their exposure. At the time exposure status is determined, all must be free of the disease. All eligible participants are then followed up over time. Since exposure status is determined before the occurrence of the outcome, a cohort study can clarify the temporal sequence between exposure and outcome, with minimal information bias. The historical and the population cohort study (Box 9.1) are efficient variants of the classical cohort study described above, which nevertheless retain the essential components of the cohort study design. The exposure can be dichotomous [i.e. exposed (to obstetric complications at birth) vs. not exposed], or graded as degrees of exposure (e.g. no recent life events, one to two life events, three or more life events). The use of grades of exposure strengthens the results of a cohort study by supporting or refuting the hypothesis that the incidence of the disease increases with increasing exposure to the risk factor; a so-called dose–response relationship. The essential features of a cohort study are: ♦ participants are defined by their exposure status rather than by outcome (as in case–control design); ♦ it is a longitudinal design: exposure status must be ascertained before outcome is known. The classical cohort study In a classical cohort study participants are selected for study on the basis of a single exposure of interest. This might be exposure to a relatively rare occupational exposure, such as ionizing radiation (through working in the nuclear power industry). Care must be taken in selecting the unexposed cohort; perhaps those working in similar industries, but without any exposure to radiation. The outcome in this case might be leukaemia. All those in the exposed and unexposed cohorts would need to be free of leukaemia (hence ‘at risk’) on recruitment into the study. The two cohorts would then be followed up for (say) 10 years and rates at which they develop leukaemia compared directly. Classical cohort studies are rare in psychiatric epidemiology. This may be in part because this type of study is especially suited to occupational exposures, which have previously been relatively little studied as causes of mental illness. However, this may change as the high prevalence of mental disorders in the workplace and their negative impact upon productivity are increasingly recognized. The UK Gulf War Study could be taken as one rather unusual example of the genre (Unwin et al. 1999). Health outcomes, including mental health status, were compared between those who were deployed in the Persian Gulf War in 1990–91, those who were later deployed in Bosnia, and an ‘era control group’ who were serving at the time of the Gulf war but were not deployed. There are two main variations on this classical cohort study design: they are popular as they can, depending on circumstances, be more efficient than the classical cohort design. The population cohort study In the classical cohort study, participants are selected on the basis of exposure, and the hypothesis relates to the effect of this single exposure on a health outcome. However, a large cohort or panel of subjects are sometimes recruited and followed up, often over many years, to study multiple exposures and outcomes. No separate comparison group is required as the comparison group is generally an unexposed sub-group of the panel. Examples include the British Doctor's Study in which over 30,000 British doctors were followed up for over 20 years to study the effects of smoking and other exposures on health (Doll et al. 1994), and the Framingham Heart Study, in which residents of a town in Massachusetts, USA have been followed up for 50 years to study risk factors for coronary heart disease (Wolf et al. 1988). The Whitehall and Whitehall II studies in the UK (Fuhrer et al. 1999; Stansfeld et al. 2002) were based again on an occupationally defined cohort, and have led to important findings concerning workplace conditions and both physical and psychiatric morbidity. Birth cohort studies, in which everyone born within a certain chronological interval are recruited, are another example of this type of study. In birth cohorts, participants are commonly followed up at intervals of 5–10 years. Many recent panel studies in the UK and elsewhere have been funded on condition that investigators archive the data for public access, in order that the dataset might be more fully exploited by the wider academic community. Population cohort studies can test multiple hypotheses, and are far more common than any other type of cohort study. The scope of the study can readily be extended to include mental health outcomes. Thus, both the British Doctor's Study (Doll et al. 2000) and the Framingham Heart Study (Seshadri et al. 2002) have gone on to report on aetiological factors for dementia and Alzheimer's Disease as the cohorts passed into the age groups most at risk for these disorders. A variant of the population cohort study is one in which those who are prevalent cases of the outcome of interest at baseline are also followed up effectively as a separate cohort in order (a) to study the natural history of the disorder by estimating its maintenance (or recovery) rate, and (b) studying risk factors for maintenance (non-recovery) over the follow-up period (Prince et al. 1998). Historical cohort studies In the classical cohort study outcome is ascertained prospectively. Thus, new cases are ascertained over a follow-up period, after the exposure status has been determined. However, it is possible to ascertain both outcome and exposure retrospectively. This variant is referred to as a historical cohort study (Fig. 9.1). A good example is the work of David Barker in testing his low birth weight hypothesis (Barker et al. 1990; Hales et al. 1991). Barker hypothesized that risk for midlife vascular and endocrine disorders would be determined to some extent by the ‘programming’ of the hypothalamo-pituitary axis through foetal growth in utero. Thus ‘small for dates’ babies would have higher blood pressure levels in adult life, and greater risk for type II diabetes (through insulin resistance). A prospective cohort study would have recruited participants at birth, when exposure (birth weight) would be recorded. They would then be followed up over four or five decades to examine the effect of birth weight on the development of hypertension and type II diabetes. Barker took the more elegant (and feasible) approach of identifying hospitals in the UK where several decades previously birth records were meticulously recorded. He then traced the babies as adults (where they still lived in the same area) and measured directly their status with respect to outcome. The ‘prospective’ element of such studies is that exposure was recorded well before outcome even though both were ascertained retrospectively with respect to the timing of the study. The historical cohort study has also proved useful in psychiatric epidemiology where it has been used in particular to test the neurodevelopmental hypothesis for schizophrenia (Jones et al. 1994; Isohanni et al. 2001). Jones et al. studied associations between adult-onset schizophrenia and childhood sociodemographic, neurodevelopmental, cognitive, and behavioural factors in the UK 1946 birth cohort; 5362 people born in the week 3–9 March 1946, and followed up intermittently since then. Subsequent onsets of schizophrenia were identified in three ways: (a) routine data: cohort members were linked to the register of the Mental Health Enquiry for England in which mental health service contacts between 1974 and 1986 were recorded; (b) cohort data: hospital and GP contacts (and the reasons for these contacts) were routinely reported at the intermittent resurveys of the cohort; (c) all cohort participants identified as possible cases of schizophrenia were given a detailed clinical interview (Present State examination) at age 36. Milestones of motor development were reached later in cases than in non-cases, particularly walking. Cases also had more speech problems than had noncases. Low educational test scores at ages 8,11, and 15 years were a risk factor. A preference for solitary play at ages 4 and 6 years predicted schizophrenia. A health visitor's rating of the mother as having below average mothering skills and understanding of her child at age 4 years was a predictor of schizophrenia in that child. Jones concluded ‘differences between children destined to develop schizophrenia as adults and the general population were found across a range of developmental domains. As with some other adult illnesses, the origins of schizophrenia may be found in early life’. Jones' findings were largely confirmed in a very similar historical cohort study in Finland (Isohanni et al. 2001); a 31 year follow-up of the 1966 North Finland birth cohort (n = 12,058). Onsets of schizophrenia were ascertained from a national hospital discharge register. The ages at learning to stand, walk and become potty-trained were each related to subsequent incidence of schizophrenia and other psychoses. Earlier milestones reduced, and later milestones increased, the risk in a linear manner. These developmental effects were not seen for non-psychotic outcomes. The findings support hypotheses regarding psychosis as having a developmental dimension with precursors apparent in early life. There are many conveniences to this approach for the contemporary investigator. ♦ The exposure data has already been collected for you. ♦ The follow-up period has already elapsed. ♦ The design maintains the essential feature of the cohort study, namely that information bias with respect to the assessment of the exposure should not be a problem. ♦ As with the Barker hypothesis example, historical cohort studies are particularly useful for investigating associations across the life course, when there is a long latency between hypothesized exposure and outcome. Despite these important advantages, such retrospective studies are often limited by reliance on historical data that was collected routinely for other purposes; often these data will be inaccurate or incomplete. Also information about possible confounders, such as smoking or diet, may be inadequate.

In a structured prediction problem, we need to learn a predictor that can produce a structured output given a structured input (e.g., part-of-speech tagging). The key learning and inference challenge is due to the exponential size of the structured output space. This paper makes four contributions towards the goal of a computationally-efficient inference and training approach for structured prediction that allows to employ complex models and to optimize for non-decomposable loss functions. First, we define a simple class of randomized greedy search (RGS) based inference procedures that leverage classification algorithms for simple outputs. Second, we develop a RGS specific learning approach for amortized inference that can quickly produce high-quality outputs for a given set of structured inputs. Third, we plug our amortized RGS inference solver inside the inner loop of parameter-learning algorithms (e.g., structured SVM) to improve the speed of training. Fourth, we perform extensive experiments on diverse structured prediction tasks. Results show that our proposed approach is competitive or better than many state-of-the-art approaches in spite of its simplicity.

Deep learning is revolutionizing predictive healthcare, including recommending medications to patients with complex health conditions. Existing approaches focus on predicting all medications for the current visit, which often overlaps with medications from previous visits. A more clinically relevant task is to identify medication changes. In this paper, we propose a new recurrent residual networks, named MICRON, for medication change prediction. MICRON takes the changes in patient health records as input and learns to update a hid- den medication vector and the medication set recurrently with a reconstruction design. The medication vector is like the memory cell that encodes longitudinal information of medications. Unlike traditional methods that require the entire patient history for prediction, MICRON has a residual-based inference that allows for sequential updating based only on new patient features (e.g., new diagnoses in the recent visit), which is efficient. We evaluated MICRON on real inpatient and outpatient datasets. MICRON achieves 3.5% and 7.8% relative improvements over the best baseline in F1 score, respectively. MICRON also requires fewer parameters, which significantly reduces the training time to 38.3s per epoch with 1.5× speed-up.

Abstract Predicting internal combustion (IC) engine variables such as the combustion phasing and duration are essential to zero-dimensional (0D) single-zone engine simulations (e.g., for the Wiebe function combustion model). This paper investigated the use of random forest machine learning models to predict these engine combustion parameters as a modality to reduce expensive engine dynamometer tests. A single-cylinder four-stroke heavy-duty spark-ignition engine fueled with methane was operated at different engine speeds and loads to provide the data for training, validation, and testing the proposed correlated model. Key engine operating variables such as spark timing, mixture equivalence ratio, and engine speed were the model inputs. The performance of the models was validated by comparing the prediction dataset with the experimentally measured results. Results showed that the prediction error of the random forest machine learning algorithm was acceptable, suggesting that it can be used to predict the combustion parameters of interest with acceptable accuracy.

Abstract This work develops a comprehensive optimal speed control framework for connected and autonomous electric vehicles considering both battery aging effects and regenerative braking limits. With the battery aging consideration, the energy benefits can be achieved together with a satisfactory battery life. The regenerative braking limits ensure the optimal control law is realistic and can be implemented in practice (e.g. there should be no regen when battery is almost full). The target vehicle intelligently controls the vehicle speed and car-following distance based on predicted traffic conditions using real-time information enabled by connectivity. The traffic prediction is based on a traffic flow model and can be implemented in a mixed-traffic scenario where both connected vehicles and non-connected vehicles share the road. The optimal control problem is formulated, simplified and discretized with minimal computational burden. It can be solved in real-time using an efficient nonlinear programming solver and is implemented in the model predictive control (MPC) fashion. The average computational time of the optimization is 0.54 seconds for a 15-second prediction horizon. A representative traffic scenario is evaluated in simulation where the target vehicle follows a vehicle platoon with 50% penetration rate of connectivity to pass a signalized roadway. The results show that 9.1% energy benefits can be obtained. The performance is satisfactory compared to 14.3% benefits with perfect traffic prediction.

Denoising diffusion probabilistic models (DDPMs) have recently achieved leading performances in many generative tasks. However, the inherited iterative sampling process costs hindered their applications to speech synthesis. This paper proposes FastDiff, a fast conditional diffusion model for high-quality speech synthesis. FastDiff employs a stack of time-aware location-variable convolutions of diverse receptive field patterns to efficiently model long-term time dependencies with adaptive conditions. A noise schedule predictor is also adopted to reduce the sampling steps without sacrificing the generation quality. Based on FastDiff, we design an end-to-end text-to-speech synthesizer, FastDiff-TTS, which generates high-fidelity speech waveforms without any intermediate feature (e.g., Mel-spectrogram). Our evaluation of FastDiff demonstrates the state-of-the-art results with higher-quality (MOS 4.28) speech samples. Also, FastDiff enables a sampling speed of 58x faster than real-time on a V100 GPU, making diffusion models practically applicable to speech synthesis deployment for the first time. We further show that FastDiff generalized well to the mel-spectrogram inversion of unseen speakers, and FastDiff-TTS outperformed other competing methods in end-to-end text-to-speech synthesis. Audio samples are available at https://FastDiff.github.io/.

Computational Fluid Dynamics (CFD) has been widely used for compressor design, yet the prediction of performance and stage matching for multi-stage, high-speed machines remain challenging. This paper presents the authors’ effort to improve the reliability of CFD in multistage compressor simulations. The endwall features (e.g. blade fillet and shape of the platform edge) are meshed with minimal approximations. Turbulence models with linear and non-linear eddy viscosity models are assessed. The non-linear eddy viscosity model predicts a higher production of turbulent kinetic energy in the passages, especially close to the endwall region. This results in a more accurate prediction of the choked mass flow and the shape of total pressure profiles close to the hub. The non-linear viscosity model generally shows an improvement on its linear counterparts based on the comparisons with the rig data. For geometrical details, truncated fillet leads to thicker boundary layer on the fillet and reduced mass flow and efficiency. Shroud cavities are found to be essential to predict the right blockage and the flow details close to the hub. At the part speed the computations without the shroud cavities fail to predict the major flow features in the passage and this leads to inaccurate predictions of massflow and shapes of the compressor characteristic. The paper demonstrates that an accurate representation of the endwall geometry and an effective turbulence model, together with a good quality and sufficiently refined grid result in a credible prediction of compressor matching and performance with steady state mixing planes.

Employing numerical prediction models can be a powerful tool for forecasting air quality and visibility hazards related to dust events. However, these numerical models are sensitive to surface conditions. Roughness features (e.g., rocks, vegetation, furrows, etc.) that shelter or attenuate wind flow over the soil surface affect the magnitude and spatial distribution of dust emission. To aide in simulating the emission phase of dust transport, we used a previously published albedo-based drag partition parameterization to better represent the component of wind friction speed affecting the immediate soil sur-face. This report serves as a guide for integrating this parameterization into the Weather Research and Forecasting with Chemistry (WRF-Chem) model. We include the procedure for preprocessing the required input data, as well as the code modifications for the Air Force Weather Agency (AFWA) dust emission module. In addition, we provide an example demonstration of output data from a simulation of a dust event that occurred in the Southwestern United States, which incorporates use of the drag partition.