Prediction of activity type in preschool children using machine learning techniques

Objectives Recent research has shown that machine learning techniques can accurately predict activity classes from accelerometer data in adolescents and adults. The purpose of this study is to develop and test machine learning models for predicting activity type in preschool-aged children. Design Participants completed 12 standardised activity trials (TV, reading, tablet game, quiet play, art, treasure hunt, cleaning up, active game, obstacle course, bicycle riding) over two laboratory visits. Methods Eleven children aged 3–6 years (mean age = 4.8 ± 0.87; 55% girls) completed the activity trials while wearing an ActiGraph GT3X+ accelerometer on the right hip. Activities were categorised into five activity classes: sedentary activities, light activities, moderate to vigorous activities, walking, and running. A standard feed-forward Artificial Neural Network and a Deep Learning Ensemble Network were trained on features in the accelerometer data used in previous investigations (10th, 25th, 50th, 75th and 90th percentiles and the lag-one autocorrelation). Results Overall recognition accuracy for the standard feed forward Artificial Neural Network was 69.7%. Recognition accuracy for sedentary activities, light activities and games, moderate-to-vigorous activities, walking, and running was 82%, 79%, 64%, 36% and 46%, respectively. In comparison, overall recognition accuracy for the Deep Learning Ensemble Network was 82.6%. For sedentary activities, light activities and games, moderate-to-vigorous activities, walking, and running recognition accuracy was 84%, 91%, 79%, 73% and 73%, respectively. Conclusions Ensemble machine learning approaches such as Deep Learning Ensemble Network can accurately predict activity type from accelerometer data in preschool children.

The BAD project: data mining, database and prediction of protein adsorption on surfaces

Protein adsorption at solid-liquid interfaces is critical to many applications, including biomaterials, protein microarrays and lab-on-a-chip devices. Despite this general interest, and a large amount of research in the last half a century, protein adsorption cannot be predicted with an engineering level, design-orientated accuracy. Here we describe a Biomolecular Adsorption Database (BAD), freely available online, which archives the published protein adsorption data. Piecewise linear regression with breakpoint applied to the data in the BAD suggests that the input variables to protein adsorption, i.e., protein concentration in solution; protein descriptors derived from primary structure (number of residues, global protein hydrophobicity and range of amino acid hydrophobicity, isoelectric point); surface descriptors (contact angle); and fluid environment descriptors (pH, ionic strength), correlate well with the output variable-the protein concentration on the surface. Furthermore, neural network analysis revealed that the size of the BAD makes it sufficiently representative, with a neural network-based predictive error of 5% or less. Interestingly, a consistently better fit is obtained if the BAD is divided in two separate sub-sets representing protein adsorption on hydrophilic and hydrophobic surfaces, respectively. Based on these findings, selected entries from the BAD have been used to construct neural network-based estimation routines, which predict the amount of adsorbed protein, the thickness of the adsorbed layer and the surface tension of the protein-covered surface. While the BAD is of general interest, the prediction of the thickness and the surface tension of the protein-covered layers are of particular relevance to the design of microfluidics devices.

Learning the learning rate for prediction with expert advice

Most standard algorithms for prediction with expert advice depend on a parameter called the learning rate. This learning rate needs to be large enough to fit the data well, but small enough to prevent overfitting. For the exponential weights algorithm, a sequence of prior work has established theoretical guarantees for higher and higher data-dependent tunings of the learning rate, which allow for increasingly aggressive learning. But in practice such theoretical tunings often still perform worse (as measured by their regret) than ad hoc tuning with an even higher learning rate. To close the gap between theory and practice we introduce an approach to learn the learning rate. Up to a factor that is at most (poly)logarithmic in the number of experts and the inverse of the learning rate, our method performs as well as if we would know the empirically best learning rate from a large range that includes both conservative small values and values that are much higher than those for which formal guarantees were previously available. Our method employs a grid of learning rates, yet runs in linear time regardless of the size of the grid.

Superpixel-based appearance change prediction for long-term navigation across seasons

Changing environments pose a serious problem to current robotic systems aiming at long term operation under varying seasons or local weather conditions. This paper is built on our previous work where we propose to learn to predict the changes in an environment. Our key insight is that the occurring scene changes are in part systematic, repeatable and therefore predictable. The goal of our work is to support existing approaches to place recognition by learning how the visual appearance of an environment changes over time and by using this learned knowledge to predict its appearance under different environmental conditions. We describe the general idea of appearance change prediction (ACP) and investigate properties of our novel implementation based on vocabularies of superpixels (SP-ACP). Our previous work showed that the proposed approach significantly improves the performance of SeqSLAM and BRIEF-Gist for place recognition on a subset of the Nordland dataset under extremely different environmental conditions in summer and winter. This paper deepens the understanding of the proposed SP-ACP system and evaluates the influence of its parameters. We present the results of a large-scale experiment on the complete 10 h Nordland dataset and appearance change predictions between different combinations of seasons.

Evaluation of intra- and extra-epithelial secretory IgA in chlamydial infections

IgA is an important mucosal antibody that can neutralize mucosal pathogens by either preventing attachment to epithelia (immune exclusion) or alternatively inhibit intraepithelial replication following transcytosis by the polymeric immunoglobulin receptor (pIgR). Chlamydia trachomatis is a major human pathogen that initially targets the endocervical or urethral epithelium in women and men, respectively. As both tissues contain abundant SIgA we assessed the protection afforded by IgA targeting different chlamydial antigens expressed during the extra and intraepithelial stages of infection. We developed an in vitro model utilizing polarizing cells expressing the murine pIgR together with antigen-specific mouse IgA, and an in vivo model utilizing pIgR-/- mice. SIgA targeting the extraepithelial chlamydial antigen, the major outer membrane protein (MOMP), significantly reduced infection in vitro by 24 % and in vivo by 44 %. Conversely, pIgR-mediated delivery of IgA targeting the intraepithelial inclusion membrane protein A (IncA) bound to the inclusion but did not reduce infection in vitro or in vivo. Similarly, intraepithelial IgA targeting the secreted protease Chlamydia protease-like activity factor (CPAF) also failed to reduce infection. Together, these data suggest the importance of pIgR-mediated delivery of IgA targeting extra but not intraepithelial chlamydial antigens for protection against a genital tract infection.

Prediction of GMA welding characteristic parameter by artificial neural network system

Nowadays, demand for automated Gas metal arc welding (GMAW) is growing and consequently need for intelligent systems is increased to ensure the accuracy of the procedure. To date, welding pool geometry has been the most used factor in quality assessment of intelligent welding systems. But, it has recently been found that Mahalanobis Distance (MD) not only can be used for this purpose but also is more efficient. In the present paper, Artificial Neural Networks (ANN) has been used for prediction of MD parameter. However, advantages and disadvantages of other methods have been discussed. The Levenberg–Marquardt algorithm was found to be the most effective algorithm for GMAW process. It is known that the number of neurons plays an important role in optimal network design. In this work, using trial and error method, it has been found that 30 is the optimal number of neurons. The model has been investigated with different number of layers in Multilayer Perceptron (MLP) architecture and has been shown that for the aim of this work the optimal result is obtained when using MLP with one layer. Robustness of the system has been evaluated by adding noise into the input data and studying the effect of the noise in prediction capability of the network. The experiments for this study were conducted in an automated GMAW setup that was integrated with data acquisition system and prepared in a laboratory for welding of steel plate with 12 mm in thickness. The accuracy of the network was evaluated by Root Mean Squared (RMS) error between the measured and the estimated values. The low error value (about 0.008) reflects the good accuracy of the model. Also the comparison of the predicted results by ANN and the test data set showed very good agreement that reveals the predictive power of the model. Therefore, the ANN model offered in here for GMA welding process can be used effectively for prediction goals.

Remaining useful life prediction of rotating equipment using covariate-based hazard models : industry applications

The ability to estimate the expected Remaining Useful Life (RUL) is critical to reduce maintenance costs, operational downtime and safety hazards. In most industries, reliability analysis is based on the Reliability Centred Maintenance (RCM) and lifetime distribution models. In these models, the lifetime of an asset is estimated using failure time data; however, statistically sufficient failure time data are often difficult to attain in practice due to the fixed time-based replacement and the small population of identical assets. When condition indicator data are available in addition to failure time data, one of the alternate approaches to the traditional reliability models is the Condition-Based Maintenance (CBM). The covariate-based hazard modelling is one of CBM approaches. There are a number of covariate-based hazard models; however, little study has been conducted to evaluate the performance of these models in asset life prediction using various condition indicators and data availability. This paper reviews two covariate-based hazard models, Proportional Hazard Model (PHM) and Proportional Covariate Model (PCM). To assess these models’ performance, the expected RUL is compared to the actual RUL. Outcomes demonstrate that both models achieve convincingly good results in RUL prediction; however, PCM has smaller absolute prediction error. In addition, PHM shows over-smoothing tendency compared to PCM in sudden changes of condition data. Moreover, the case studies show PCM is not being biased in the case of small sample size.

Prediction of shrinkage and porosity during drying: Considering both material properties and process conditions

Drying of food materials offers a significant increase in the shelf life of food materials, along with the modification of quality attributes due to simultaneous heat and mass transfer. Shrinkage and variations in porosity are the common micro and microstructural changes that take place during the drying of mostly the food materials. Although extensive research has been carried out on the prediction of shrinkage and porosity over the time of drying, no single model exists which consider both material properties and process condition in the same model. In this study, an attempt has been made to develop and validate shrinkage and porosity models of food materials during drying considering both process parameters and sample properties. The stored energy within the sample, elastic potential energy, glass transition temperature and physical properties of the sample such as initial porosity, particle density, bulk density and moisture content have been taken into consideration. Physical properties and validation have been made by using a universal testing machine ( Instron 2kN), a profilometer (Nanovea) and a pycnometer. Apart from these, COMSOL Multiphysics 4.4 has been used to solve heat and mass transfer physics. Results obtained from models of shrinkage and porosity is quite consistent with the experimental data. Successful implementation of these models would ensure the use of optimum energy in the course of drying and better quality retention of dried foods.

Increased intra-cortical porosity reduces bone stiffness and strength in pediatric patients with osteogenesis imperfecta

Osteogenesis imperfecta (OI) is a heritable disease occurring in one out of every 20,000 births. Although it is known that Type I collagen mutation in OI leads to increased bone fragility, the mechanism of this increased susceptibility to fracture is not clear. The aim of this study was to assess the microstructure of cortical bone fragments from patients with osteogenesis imperfecta (OI) using polarized light microscopy, and to correlate microstructural observations with the results of previously performed mechanical compression tests on bone from the same source. Specimens of cortical bone were harvested from the lower limbs of three (3) OI patients at the time of surgery, and were divided into two groups. Group 1 had been subjected to previous micro-mechanical compression testing, while Group 2 had not been subjected to any prior testing. Polarized light microscopy revealed disorganized bone collagen architecture as has been previously observed, as well as a large increase in the areal porosity of the bone compared to typical values for healthy cortical bone, with large (several hundred micron sized), asymmetrical pores. Importantly, the areal porosity of the OI bone samples in Group 1 appears to correlate strongly with their previously measured apparent Young's modulus and compressive strength. Taken together with prior nanoindentation studies on OI bone tissue, the results of this study suggest that increased intra-cortical porosity is responsible for the reduction in macroscopic mechanical properties of OI cortical bone, and therefore that in vivo imaging modalities with resolutions of ~ 100 μm or less could potentially be used to non-invasively assess bone strength in OI patients. Although the number of subjects in this study is small, these results highlight the importance of further studies in OI bone by groups with access to human OI tissue in order to clarify the relationship between increased porosity and reduced macroscopic mechanical integrity.

Prediction of maximal surface electromyographically based voluntary contractions of erector spinae muscles from sonographic measurements during isometric contractions

Objectives Currently, there are no studies combining electromyography (EMG) and sonography to estimate the absolute and relative strength values of erector spinae (ES) muscles in healthy individuals. The purpose of this study was to establish whether the maximum voluntary contraction (MVC) of the ES during isometric contractions could be predicted from the changes in surface EMG as well as in fiber pennation and thickness as measured by sonography. Methods Thirty healthy adults performed 3 isometric extensions at 45° from the vertical to calculate the MVC force. Contractions at 33% and 100% of the MVC force were then used during sonographic and EMG recordings. These measurements were used to observe the architecture and function of the muscles during contraction. Statistical analysis was performed using bivariate regression and regression equations. Results The slope for each regression equation was statistically significant (P < .001) with R2 values of 0.837 and 0.986 for the right and left ES, respectively. The standard error estimate between the sonographic measurements and the regression-estimated pennation angles for the right and left ES were 0.10 and 0.02, respectively. Conclusions Erector spinae muscle activation can be predicted from the changes in fiber pennation during isometric contractions at 33% and 100% of the MVC force. These findings could be essential for developing a regression equation that could estimate the level of muscle activation from changes in the muscle architecture.