An On-Device Learning System for Estimating Liquid Consumption from Consumer-Grade Water Bottles and Its Evaluation.

A lightweight on-device liquid consumption estimation system involving an energy-aware machine learning algorithm is developed in this work. This system consists of two separate on-device neural network models that carry out liquid consumption estimation with the result of two tasks: the detection of sip from gestures with which the bottle is handled by its user and the detection of first sips after a bottle refill. This predictive volume estimation framework incorporates a self-correction mechanism that can minimize the error after each bottle fill-up cycle, which makes the system robust to errors from the sip classification module. In this paper, a detailed characterization of sip detection is performed to understand the accuracy-complexity tradeoffs by developing and implementing a variety of different ML models with varying complexities. The maximum energy consumed by the entire framework is around 119 mJ during a maximum computation time of 300 µs. The energy consumption and computation times of the proposed framework is suitable for implementation in low-power embedded hardware that can be incorporated in consumer grade water bottles.

GazeBase, a large-scale, multi-stimulus, longitudinal eye movement dataset.

This manuscript presents GazeBase, a large-scale longitudinal dataset containing 12,334 monocular eye-movement recordings captured from 322 college-aged participants. Participants completed a battery of seven tasks in two contiguous sessions during each round of recording, including a - (1) fixation task, (2) horizontal saccade task, (3) random oblique saccade task, (4) reading task, (5/6) free viewing of cinematic video task, and (7) gaze-driven gaming task. Nine rounds of recording were conducted over a 37 month period, with participants in each subsequent round recruited exclusively from prior rounds. All data was collected using an EyeLink 1000 eye tracker at a 1,000 Hz sampling rate, with a calibration and validation protocol performed before each task to ensure data quality. Due to its large number of participants and longitudinal nature, GazeBase is well suited for exploring research hypotheses in eye movement biometrics, along with other applications applying machine learning to eye movement signal analysis. Classification labels produced by the instrument's real-time parser are provided for a subset of GazeBase, along with pupil area.

A Container-Attachable Inertial Sensor for Real-Time Hydration Tracking.

Various sensors have been proposed to address the negative health ramifications of inadequate fluid consumption. Amongst these solutions, motion-based sensors estimate fluid intake using the characteristics of drinking kinematics. This sensing approach is complicated due to the mutual influence of both the drink volume and the current fill level on the resulting motion pattern, along with differences in biomechanics across individuals. While motion-based strategies are a promising approach due to the proliferation of inertial sensors, previous studies have been characterized by limited accuracy and substantial variability in performance across subjects. This research seeks to address these limitations for a container-attachable triaxial accelerometer sensor. Drink volume is computed using support vector machine regression models with hand-engineered features describing the container's estimated inclination. Results are presented for a large-scale data collection consisting of 1908 drinks consumed from a refillable bottle by 84 individuals. Per-drink mean absolute percentage error is reduced by 11.05% versus previous state-of-the-art results for a single wrist-wearable inertial measurement unit (IMU) sensor assessed using a similar experimental protocol. Estimates of aggregate consumption are also improved versus previously reported results for an attachable sensor architecture. An alternative tracking approach using the fill level from which a drink is consumed is also explored herein. Fill level regression models are shown to exhibit improved accuracy and reduced inter-subject variability versus volume estimators. A technique for segmenting the entire drink motion sequence into transport and sip phases is also assessed, along with a multi-target framework for addressing the known interdependence of volume and fill level on the resulting drink motion signature.

Home-based upper extremity rehabilitation support using a contactless ultrasonic sensor.

Home-based rehabilitation protocols have been shown to improve outcomes amongst individuals with limited upper-extremity (UE) functionality. While approaches employing both video conferencing technologies and gaming platforms have been successfully demonstrated for such applications, concerns regarding patient privacy and technological complexity may limit further adoption. As an alternative solution for assessing adherence to prescribed UE rehabilitation protocols, the Echolocation Activity Detector, a linear array of first-reflection ultrasonic distance sensors, is proposed herein. To demonstrate its utility for home-based rehabilitation, a controlled experiment exploring the ability of the system to distinguish between various parameters of UE motion, including motion plane, range, and speed, was conducted for five participants. Activity classification is accomplished using a quadratic support vector machine classifier using time-domain features which exploit the known geometric relationships between the patient and the device, along with the ideal kinematics of the activities of interest. Average classification accuracy for the five classes of UE motion considered herein exceeds 91%.

Office activity classification using first-reflection ultrasonic echolocation.

Excessive sedentary time poses considerable health risks for individuals predominately engaged in desk-bound work. To empower interventions aimed at addressing this problem, reliable technologies for continuous activity monitoring within an office environment are required. As an alternative to existing solutions, we propose the Echolocation-based Activity Detector, a contactless sensor array of four first-reflection ultrasonic distance sensors. The research described herein demonstrates the capacity of the sensor to distinguish between common activities performed at a workstation within an office environment, including sedentary sitting, typing, writing, and standing. Cubic support vector machine classifiers are developed using dispersion-related features computed from the time-series array outputs. Average classification accuracy for sedentary activities exceeds 85%, while classification accuracy for the entire activity set exceeds 80% for a controlled experiment conducted with six participants.

Contactless on-bed activity sensing using first-reflection echolocation.

A novel system capable of detecting on-bed activity during sleep using first-reflection ultrasonic echolocation is described herein. As is employed in many existing solutions, such activity detection may be utilized in the assessment of sleep quality. Compared to current approaches using either wearable devices or sensors collocated on the surface of the bed, the proposed architecture greatly enhances convenience for the end-user by providing minimal disruptions to his or her standard sleep routine. A series of experiments were conducted in order to investigate the capacity of the system to detect activity during sleep. System performance was benchmarked against both a wrist-worn accelerometer as well as a smartphone application placed adjacent to the subject on the bed. Analysis demonstrates a statistically significant correlation between features computed from the system's output and the filtered activity data produced by the application, with maximum p values on the order of 10-3. Comparison with activity estimates formulated from the wrist-worn accelerometer output suggests stronger agreement, as indicated by increased correlation coefficient values.

A wearable system for asymmetric contactless human sensing.

A wearable sensor capable of detecting the presence of humans within a front-facing 90-degree sector of varying radius is demonstrated herein. The system offers extensive applicability across a variety of scenarios where detecting the parameters of human interaction, including separation distance and duration, is of value. Sensing is accomplished using an ultrasonic distance and passive infrared sensor. This design improves upon previous approaches presented in the literature by eliminating privacy concerns associated with audio and video capture, and also relaxing the requirement that both interacting individuals be in possession of dedicated hardware. A KNN classifier is developed using data obtained from a designed indoor experiment intended to demonstrate system robustness across geometries consistent with those observed in the target application. Employing a set of only three features, an overall accuracy rate of 94.2% is realized for detecting human interactions occurring within a 90-degree sector of three-foot radius.

Financial capacity of older African Americans with amnestic mild cognitive impairment.

This study investigated financial abilities of 154 patients with mild cognitive impairment (MCI) (116 white, 38 African American) using the Financial Capacity Instrument (FCI). In a series of linear regression models, we examined the effect of race on FCI performance and identified preliminary predictor variables that mediated observed racial differences on the FCI. Prior/premorbid abilities were identified. Predictor variables examined in the models included race and other demographic factors (age, education, sex), performance on global cognitive measures (MMSE, DRS-2 Total Score), history of cardiovascular disease (hypertension, diabetes, hypercholesterolemia), and a measure of educational achievement (WRAT-3 Arithmetic). African American patients with MCI performed below white patients with MCI on 6 of the 7 FCI domains examined and on the FCI total score. WRAT-3 Arithmetic emerged as a partial mediator of group differences on the FCI, accounting for 54% of variance. In contrast, performance on global cognitive measures and history of cardiovascular disease only accounted for 14% and 2%, respectively, of the variance. Racial disparities in financial capacity seem to exist among patients with amnestic MCI. Basic academic math skills related to educational opportunity and quality of education account for a substantial proportion of the group difference in financial performance.

Brain N-acetylaspartate is reduced in Parkinson disease with dementia.

Persons with Parkinson disease (PD) are at risk of developing dementia. Of the dementias affecting patients with PD, PD with dementia (PDD) is not well understood, although brain imaging studies to date have observed characteristic patterns of brain atrophy. Metabolic differences have been observed in magnetic resonance spectroscopy (MRS) studies comparing patients with PDD to nondemented PD patients, although it is unclear whether PDD patients have abnormally low MRS ratios compared with healthy age-matched adults. In this study, 12 patients with PDD, 12 patients with PD and no dementia, and 12 age-matched healthy older adults underwent MRS of the posterior cingulate gyrus. Patients with PDD showed lower N-acetylaspartate/creatine (NAA/Cr) compared with controls (P=0.004) and compared with nondemented PD patients (P=0.003). No abnormalities were observed in choline/Cr or myo-Inositol/Cr. NAA/Cr was correlated with mental status in patients with PD and in patients with PDD (r=0.56; P=0.029). The findings suggest that reduced NAA/Cr of the posterior cingulate could be used as a marker for dementia in patients with PD. Future studies investigating the utility of brain MRS as a predictor of dementia in PD and comparing brain metabolism in PDD with other dementias seem warranted.