Search for: All records

This study applies machine learning to identify menstrual cycle phases using physiological signals recorded from a wrist-worn device. These signals include skin temperature, electrodermal activity (EDA), interbeat interval (IBI), and heart rate (HR), and were collected without requiring participant input. Data from 65 cycles across 18 subjects were analyzed, and multiple classifiers including random forest (RF) models were trained to classify the phases. Using a leave-last-cycle-out approach, and features from non-overlapping fixed-size windows, the RF model achieved 87% accuracy and an area under the receiver operating characteristic curve (AUC-ROC) of 0.96 when classifying three phases (period, ovulation, and luteal). For daily phase tracking using a sliding window, the RF model achieved 68% accuracy and an AUC-ROC of 0.77 when classifying four phases (period, follicular, ovulation, luteal). While these results highlight the potential of wrist-based physiological signals to enable automated phase tracking, reduce the burden of self-reporting, and improve access to cycle tracking solutions, further validation is needed to enhance the results. 

Facial attribute recognition is conventionally computed from a single image. In practice, each subject may have multiple face images. Taking the eye size as an example, it should not change, but it may have different estimation in multiple images, which would make a negative impact on face recognition. Thus, how to compute these attributes corresponding to each subject rather than each single image is a profound work. To address this question, we deploy deep training for facial attributes prediction, and we explore the inconsistency issue among the attributes computed from each single image. Then, we develop two approaches to address the inconsistency issue. Experimental results show that the proposed methods can handle facial attribute estimation on either multiple still images or video frames, and can correct the incorrectly annotated labels. The experiments are conducted on two large public databases with annotations of facial attributes.