Activities of daily living (ADL) identification with wearables has significant implications in healthy lifestyle management and offers an important sensor-based supervised learning research benchmark. Most ADL studies use single-modality (i.e., single sensor type like smartwatch only or earbuds only) data while multi-sensor data fusion studies using early-stage fusion of modalities from multi-sensors and multi-devices are emerging. To improve classification performance and model interpretability, leveraging early-stage fusion, late-stage fusion and individual modalities, we introduced novel modality-aware dynamic fusion (MAD-Fusion) models for multi-sensor data-fusion-based ADL identification. Based on early-stage fusion, we incorporated conformal prediction for uncertainty quantification, uncertainty late-stage fusion for cross-modality interpretability, and multi-modal strength-aware classification module. Trained on 36 independent subjects and tested on 4 independent subjects from Samsung ADL dataset with multi-sensor (accelerometers and gyroscopes) and multi-device (earbuds and smartwatch), MAD-Fusion not only achieved the state-of-the-art classification performance (accuracy: 0.9504, F1-score: 0.9142), but also enabled better interpretability of contributions and uncertainties from different modalities. The additional contribution of each building block is validated systematically. Furthermore, we validated MAD-Fusion’s superior performance on two public datasets in multi-sensor single-device settings (UCI-HAR and USC-HAD datasets). On all three datasets, MAD-Fusion manifested statistically significantly superiority comparing against baselines of single-modality, early-stage fusion and late-stage fusion (p< 0.001). To conclude, the novel MAD-Fusion models improve the classification performance, uncertainty quantification and interpretability for the ADL identification and can be applied to broader supervised learning areas requiring high performance, rigorous uncertainty quantification and model interpretability.

Slow-paced deep breathing exercises have many health benefits, including stress management, lowering blood pressure, pain management, and controlling pulmonary conditions. While biofeedback can significantly improve the efficacy of breathing exercises, existing approaches support limited biomarkers, such as breathing rate, for specific breathing exercises (e.g., equal-phase breathing) in particular conditions without considering the breath-holding phase or variation in device orientation. Therefore, there needs to be a more convenient and robust approach that can generate and deliver comprehensive digital breathing biomarkers to facilitate biofeedback for various types of breathing exercises. In this article, we present a system with lightweight algorithms to passively track mindful breathing in real-time using lower-power earbud motion sensors to extract fine-grained comprehensive breathing biomarkers for generating biofeedback on users’ breathing exercises. We utilize the earbud’s motion sensor data to detect nonbreathing head motion and develop an extensive set of breathing markers, including breathing phases, breathing depth, breathing rate, breathing symmetry, and breath-holding. Such a comprehensive set of biomarkers can enable engaging user experience and effective mindful breathing exercises toward better stress management and overall mental well-being. Moreover, we develop a physiologically informed, novel earbud orientation handling algorithm that makes our biomarkers more resilient to ear canal shape and size. Finally, we showcase potential use-cases based on the breathing biomarkers derived from our algorithms to provide biofeedback on user’s overall breathing performance.

Mobile systems face privacy challenges including coarsegrained privacy control and the inability to distinguish private and public files. We propose System-E, a novel system which can enhance the user privacy on mobile systems (e.g., Android) by (1) enabling users to set finer grained permissions for apps accessing data, and (2) enabling automatic classification of data (e.g., photos) at the storage layer (e.g., by using deep learning) to prevent potentially sensitive data from being stored/accessed with open permissions.

With the increase in data volumes, it is prudent to classify data depending on its criticality. One might prefer a cheap storage for a year old system logs but a highly fault tolerant storage for personal photos. The existing solutions include storing these two sets of data in two different systems or choosing a system with a fault tolerance level required by the most critical data. This means a higher storage footprint for the lesser critical data. In today's petabyte scale data centers, this will result in a significant increase in the overall storage footprint making the system unattractive. In addition, data storage solutions designed for traditional data centers have to be re-engineered to work at cloud scale. For instance, reliable storage using traditional RAID like systems are inflexible and do not provide a mechanism to easily change the level of redundancy once the system is set up. As a result, changes to the reliability requirements of data over time cannot be serviced efficiently. To address these problems, we propose Cider, which aims at providing an extremely flexible, reliable, and distributed block store using erasure codes. Cider takes a new approach to reduce the storage overhead by offering a variable degree of fault tolerance at a granularity of a single block. We achieve this by using a thin block translation layer and a block level metadata system. Cider can be readily used by many high performance and enterprise applications which require a block level interface. Through a case study of a novel flash based clustered storage system, we make a strong case for the adoption of Cider in future systems. We discuss various design trade-offs and implementation challenges associated with designing our system. Lastly, we discuss the preliminary implementation of our system and results from our initial experiments.

In this paper, we design a pluggable data management solution for modern mobile platforms (e.g., Android). Our goal is to allow data management mechanisms and policies to be implemented independently of core app logic. Our design allows a user to install data management solutions as apps, install multiple such solutions on a single device, and choose a suitable solution each for one or more apps. It allows app developers to focus their effort on app logic and helps the developers of data management solutions to achieve wider deployability. It also gives increased control of data management to end users and allows them to use different solutions for different apps. We present a prototype implementation of our design called BlueMountain, and implement several data management solutions for file and database management to demonstrate the utility and ease of using our design. We perform detailed microbenchmarks as well as end-to-end measurements for files and databases to demonstrate the performance overhead incurred by our implementation.

BlueMountain is a system that enables building pluggable data management solutions which can be linked with any Android app at runtime, without requiring any modifications to the Android platform. BlueMountain simplifies the app development, provides flexibility to end users, and works with existing apps.

This poster presents Pixelsior, a new mobile platform service for photo data management in mobile apps.

This paper presents PigOut, a system that enables federated data processing over multiple Hadoop clusters. Using PigOut, a user (such as a data analyst) can write a single script in a high-level language to efficiently use multiple Hadoop clusters. There is no need to manually write multiple scripts and coordinate the execution for different clusters. PigOut accomplishes this by automatically partitioning a single, user-supplied script into multiple scripts that run on different clusters. Additionally, PigOut generates workflow descriptions to coordinate execution across clusters. In doing so, PigOut leverages existing tools built around Hadoop, avoiding extra effort required from users or administrators. For example, PigOut uses Pig Latin, a popular query language for Hadoop MapReduce, in a (virtually) unmodified form. Through our evaluation with PigMix, the standard benchmark for Pig, we demonstrate that PigOut's automatically-generated scripts and workflow definitions have comparable performance to manual, hand-tuned ones. We also report our experience with manually writing multiple scripts for a set of federated clusters, and compare the process with PigOut's automated approach.