There have been many new algorithms proposed over the last five years for solving time series classification (TSC) problems. A recent experimental comparison of the leading TSC algorithms has demonstrated that one approach is significantly more accurate than all others over 85 datasets. That approach, the Flat Collective of Transformation-based Ensembles (Flat-COTE), achieves superior accuracy through combining predictions of 35 individual classifiers built on four representations of the data into a flat hierarchy. Outside of TSC, deep learning approaches such as convolutional neural networks (CNN) have seen a recent surge in popularity and are now state of the art in many fields. An obvious question is whether CNNs could be equally transformative in the field of TSC. To test this, we implement a common CNN structure and compare performance to Flat-COTE and a recently proposed time series-specific CNN implementation. We find that Flat-COTE is significantly more accurate than both deep learning approaches on 85 datasets. These results are impressive, but Flat-COTE is not without deficiencies. We improve the collective by adding new components and proposing a modular hierarchical structure with a probabilistic voting scheme that allows us to encapsulate the classifiers built on each transformation. We add two new modules representing dictionary and interval-based classifiers, and significantly improve upon the existing frequency domain classifiers with a novel spectral ensemble. The resulting classifier, the Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE) is significantly more accurate than Flat-COTE and represents a new state of the art for TSC. HIVE-COTE captures more sources of possible discriminatory features in time series and has a more modular, intuitive structure.

Fig. 1. A machine learning approach is used to learn a regression function mapping phoneme labels to speech animation. Our approach generates continuous, natural-looking speech animation for a reference face parameterization that can be retargeted to the face of any computer generated character.We introduce a simple and efective deep learning approach to automatically generate natural looking speech animation that synchronizes to input speech. Our approach uses a sliding window predictor that learns arbitrary nonlinear mappings from phoneme label input sequences to mouth movements in a way that accurately captures natural motion and visual coarticulation efects. Our deep learning approach enjoys several attractive properties: it runs in real-time, requires minimal parameter tuning, generalizes well to novel input speech sequences, is easily edited to create stylized and emotional speech, and is compatible with existing animation retargeting approaches. One important focus of our work is to develop an efective approach for speech animation that can be easily integrated into existing production pipelines. We provide a detailed description of our end-to-end approach, including machine learning design decisions. Generalized speech animation results are demonstrated over a wide range of animation clips on a variety of characters and voices, including singing and foreign language input. Our approach can also generate on-demand speech animation in real-time from user speech input.

BackgroundMovement behaviours performed over a finite period such as a 24 h day are compositional data. Compositional data exist in a constrained simplex geometry that is incongruent with traditional multivariate analytical techniques. However, the expression of compositional data as log-ratio co-ordinate systems transfers them to the unconstrained real space, where standard multivariate statistics can be used. This study aimed to use a compositional data analysis approach to examine the adiposity and cardiorespiratory fitness predictions of time reallocations between children’s daily movement behaviours.MethodsThis study used cross-sectional data from the Active Schools: Skelmersdale study, which involved Year 5 children from a low-income community in northwest England (n = 169). Measures included accelerometer-derived 24 h activity (sedentary time [ST], light physical activity [LPA], moderate-to-vigorous physical activity [MVPA], and sleep), cardiorespiratory fitness determined by the 20 m shuttle run test, objectively measured height, weight and waist circumference (from which zBMI and percent waist circumference-to-height ratio (%WHtR) were derived) and sociodemographic covariates. Log-ratio multiple linear regression models were used to predict adiposity and fitness for the mean movement behaviour composition, and for new compositions where fixed durations of time had been reallocated from one behaviour to another, while the remaining behaviours were unchanged. Predictions were also made for reallocations of fixed durations of time using the mean composition of three different weight status categories (underweight, normal-weight, and overweight/obese) as the starting point.ResultsReplacing MVPA with any other movement behaviour around the mean movement composition predicted higher adiposity and lower CRF. The log-ratio model predictions were asymmetrical: when time was reallocated to MVPA from sleep, ST, or LPA, the estimated detriments to fitness and adiposity were larger in magnitude than the estimated benefits of time reallocation from MVPA to sleep, ST or LPA. The greatest differences in fitness and fatness for reallocation of fixed duration of MVPA were predicted at the mean composition of overweight/obese children.ConclusionsFindings reinforce the key role of MVPA for children’s health. Reallocating time from ST and LPA to MVPA in children is advocated in school, home, and community settings.Electronic supplementary materialThe online version of this article (doi:10.1186/s12966-017-0521-z) contains supplementary material, which is available to authorized users.

The version presented here may differ from the published version or from the version of the record. Please see the repository URL above for details on accessing the published version and note that access may require a subscription.

Regular physical activity (PA) is associated with numerous physical and psychological health benefits. Adolescents, specifically girls, are at risk of physical inactivity. To date, there is limited research on PA interventions involving peers, which could encourage more adolescent girls to engage in PA. The investigation aimed to evaluate the feasibility of a novel school three-tier peer-led mentoring model designed to improve PA levels and reduce sedentary time (ST) of adolescent girls. Two-hundred and forty-nine Year 9 adolescent girls (13–15 years old) from three UK secondary schools were invited to participate in a peer-led mentoring intervention (Girls Peer Activity (G-PACT) project). The peer-led mentoring model was delivered in all three schools. Two of the schools received an additional after-school PA component. PA and ST were assessed through wrist-worn accelerometry. Girls who received an exercise class after-school component significantly increased their whole day moderate-to-vigorous PA (MVPA) (3.2 min, p = 0.009, d = 0.33). Girls who received no after-school component significantly decreased their MVPA (3.5 min, p = 0.016, d = 0.36) and increased their ST (17.2 min, p = 0.006, d = 0.43). The G-PACT intervention demonstrated feasibility of recruitment and data collection procedures for adolescent girls. The peer-led mentoring model shows promise for impacting girls’ MVPA levels when combined with an after-school club PA opportunity.