Search

Electronic health records (EHRs) provide an abundance of data for clinical outcomes modeling. The prevalence of EHR data has enabled a number of studies using a variety of machine learning algorithms to predict potential adverse events. However, these studies do not account for the heterogeneity present in EHR data, for example, various lengths of stay and frequencies of vitals. Therefore, studies limit the types of features extracted or the domain considered to provide a more homogeneous training set to machine learning models. The heterogeneity in this data represents important risk differences in each patient. In this work, we examine such data in an intensive care unit (ICU) setting, where the length of stay and the frequency of data gathered may vary significantly based upon the severity of patient condition. Therefore, it is unreasonable to use the same model for patients first entering the ICU versus those that have been there for above average lengths of stay. Developing multiple individual models to account for the various EHR data may be tedious and not account for rare cases well. We address this challenge by developing a dynamic model, based upon meta- learning, to adapt to data heterogeneity and generate predictions of various outcomes across the different lengths of data. We compare this technique against a set of benchmarks on a publicly- available ICU dataset (MIMIC-III) and demonstrate improved model performance by accounting for data heterogeneity.

A chalenge in developing machine learning models for patient risk prediction involves addressing patient heterogeneity and interpreting the model outcome in clinical settings. Patient heterogeneity manifests as clinical differences among homogeneous patient subtypes in observational datasets. The discovery of such subtypes is helpful in precision medicine, where different risk factors from different patient would contribute differently to disease development and thus personalized treatment. In this paper, we use a Mixture-of-Experts (MoE) model and specifically couple it with a sparse gating network to handle patient heterogeneity for prediction and to aid interpretation of patient subtype separation. In experiment we show that with this sparsity we can improve the risk prediction. We therefore conduct empirical study to understand why and how the model learn to subtype patients from sparse training.

Deep recurrent neural networks perform well on sequence data and are the model of choice. However, it is a daunting task to decide the structure of the networks, i.e. the number of layers, especially considering different computational needs of a sequence. We propose a layer flexible recurrent neural network with adaptive computation time, and expand it to a sequence to sequence model. Different from the adaptive computation time model, our model has a dynamic number of transmission states which vary by step and sequence. We evaluate the model on a financial data set and Wikipedia language modeling. Experimental results show the performance improvement of 7 % to 12 % and indicate the model's ability to dynamically change the number of layers along with the computational steps.

Wearables with embedded electrodes and sensors are capable of continuously performing Electrocardiography (ECG), recording electrical activity of the heart, while estimating heart-rate of the wearer. Recent advances in wearable technology have generated comfortable flexible sensors that conform to the contours of the body and can measure heart-rate in the field by capturing QRS complexes of ECG signals. Due to various activities of daily living (ADL), skin deformation by means of lateral, rotational or skin stretching can cause changes in the current pathways of the sensor creating noisy data. The challenge is to disentangle the noise from the usable data in order to accurately detect QRS complexes of ECG signals. In this paper we design a framework to capture noise from a miniaturized flexible sensor, the Biostamp1 (with 4 leads), worn on the chest of 16 participants performing a set of structured ADL in a home setting, and a baseball player (pitcher). We present a machine-learning framework using a consensus fusion classifier comprising a Support Vector Machine and Neural Network learned model to remove noise while preserving neighboring R-peaks. We evaluate the model using Leave One Subject Out (LOSO) and yield an average of 83% F-measure on the 17 participants (including the pitcher). The low false negative rate provides accurate heart-rate detection on a finegrained (every 5 seconds) level, in the presence of intermittent stretching of the skin. Our results increase the reliability of detecting heart-rate in real-world and player settings, increasing the utility of flexible ECG-based sensors in the field.

This paper is based on the maximum power point tracking (MPPT) strategy for Direct Drive Permanent Magnet Synchronous Generator Wind turbine. The parts of generator is setted in d-p synchronous rotating reference frame. And the voltage space vector PWM(SVPWM) in field oriented control is adopted for generator sider converter and grid sider converter with back-to-back IGBT frequency converter. PI controller is widely used in industrial control scheme. However, Conventional PI controller doesn't appear smoothing and short time oscillation in complex structures. A novel control strategy has applied in this wind power generation system. Simulation and results show comparsion between tow kinds of controllers via Matlab/Simulink.