Semi-Synchronous Federated Learning for Energy-Efficient Training and Accelerated Convergence in Cross-Silo Settings

There are situations where data relevant to machine learning problems are distributed across multiple locations that cannot share the data due to regulatory, competitiveness, or privacy reasons. Machine learning approaches that require data to be copied to a single location are hampered by the challenges of data sharing. Federated Learning (FL) is a promising approach to learn a joint model over all the available data across silos. In many cases, the sites participating in a federation have different data distributions and computational capabilities. In these heterogeneous environments existing approaches exhibit poor performance: synchronous FL protocols are communication efficient, but have slow learning convergence and high energy cost; conversely, asynchronous FL protocols have faster convergence with lower energy cost, but higher communication. In this work, we introduce a novel energy-efficient Semi-Synchronous Federated Learning protocol that mixes local models periodically with minimal idle time and fast convergence. We show through extensive experiments over established benchmark datasets in the computer-vision domain as well as in real-world biomedical settings that our approach significantly outperforms previous work in data and computationally heterogeneous environments .

Applying Multivariate Segmentation Methods to Human Activity Recognition From Wearable Sensors' Data

Background

Time-resolved quantification of physical activity can contribute to both personalized medicine and epidemiological research studies, for example, managing and identifying triggers of asthma exacerbations. A growing number of reportedly accurate machine learning algorithms for human activity recognition (HAR) have been developed using data from wearable devices (eg, smartwatch and smartphone). However, many HAR algorithms depend on fixed-size sampling windows that may poorly adapt to real-world conditions in which activity bouts are of unequal duration. A small sliding window can produce noisy predictions under stable conditions, whereas a large sliding window may miss brief bursts of intense activity.

Objective

We aimed to create an HAR framework adapted to variable duration activity bouts by (1) detecting the change points of activity bouts in a multivariate time series and (2) predicting activity for each homogeneous window defined by these change points.

Methods

We applied standard fixed-width sliding windows (4-6 different sizes) or greedy Gaussian segmentation (GGS) to identify break points in filtered triaxial accelerometer and gyroscope data. After standard feature engineering, we applied an Xgboost model to predict physical activity within each window and then converted windowed predictions to instantaneous predictions to facilitate comparison across segmentation methods. We applied these methods in 2 datasets: the human activity recognition using smartphones (HARuS) dataset where a total of 30 adults performed activities of approximately equal duration (approximately 20 seconds each) while wearing a waist-worn smartphone, and the Biomedical REAl-Time Health Evaluation for Pediatric Asthma (BREATHE) dataset where a total of 14 children performed 6 activities for approximately 10 min each while wearing a smartwatch. To mimic a real-world scenario, we generated artificial unequal activity bout durations in the BREATHE data by randomly subdividing each activity bout into 10 segments and randomly concatenating the 60 activity bouts. Each dataset was divided into ~90% training and ~10% holdout testing.

Results

In the HARuS data, GGS produced the least noisy predictions of 6 physical activities and had the second highest accuracy rate of 91.06% (the highest accuracy rate was 91.79% for the sliding window of size 0.8 second). In the BREATHE data, GGS again produced the least noisy predictions and had the highest accuracy rate of 79.4% of predictions for 6 physical activities.

Conclusions

In a scenario with variable duration activity bouts, GGS multivariate segmentation produced smart-sized windows with more stable predictions and a higher accuracy rate than traditional fixed-size sliding window approaches. Overall, accuracy was good in both datasets but, as expected, it was slightly lower in the more real-world study using wrist-worn smartwatches in children (BREATHE) than in the more tightly controlled study using waist-worn smartphones in adults (HARuS). We implemented GGS in an offline setting, but it could be adapted for real-time prediction with streaming data.

A Scalable Data Integration and Analysis Architecture for Sensor Data of Pediatric Asthma

According to the Centers for Disease Control, in the United States there are 6.8 million children living with asthma. Despite the importance of the disease, the available prognostic tools are not sufficient for biomedical researchers to thoroughly investigate the potential risks of the disease at scale. To overcome these challenges we present a big data integration and analysis infrastructure developed by our Data and Software Coordination and Integration Center (DSCIC) of the NIBIB-funded Pediatric Research using Integrated Sensor Monitoring Systems (PRISMS) program. Our goal is to help biomedical researchers to efficiently predict and prevent asthma attacks. The PRISMS-DSCIC is responsible for collecting, integrating, storing, and analyzing real-time environmental, physiological and behavioral data obtained from heterogeneous sensor and traditional data sources. Our architecture is based on the Apache Kafka, Spark and Hadoop frameworks and PostgreSQL DBMS. A main contribution of this work is extending the Spark framework with a mediation layer, based on logical schema mappings and query rewriting, to facilitate data analysis over a consistent harmonized schema. The system provides both batch and stream analytic capabilities over the massive data generated by wearable and fixed sensors.