Mobile Computing | Ying Wu College of Computing

Screen Shot 2024-02-06 at 1.47.48 PM.png

Cristian Borcea
Professor
borcea@njit.edu

Research Areas: Mobile computing and sensing, machine learning for mobility, predictive computational advertising

Federated Learning for Mobile Sensing Data

Federated Learning (FL) is a distributed machine learning paradigm that enables privacy-aware training and inference on mobile devices with help from the cloud. FL can enable a wide range of mobile apps that use machine learning models on mobile sensing data. We created FLSys,
a mobile-cloud FL system that can be a key component of an ecosystem of FL models and apps. FLSys balances model performance with resource consumption, tolerates communication failures, achieves scalability, provides advanced privacy mechanisms and supports third-party apps and models. We also created a system for fine-grained location prediction of mobile users, based on GPS traces collected on smart phones. Applications that benefit from this system include video quality adaptation as a function of the 5G network quality at the predicted user locations and augmented reality apps that speed up content rendering based on predicted user locations.

Screen Shot 2024-02-06 at 1.47.56 PM.png

Cong Shi
Assistant Professor
cong.shi@njit.edu

Research Areas: Mobile security, robust and trustworthy machine learning

Solving the WiFi Sensing Dilemma in Reality Leveraging Conformal Prediction

WiFi sensing has demonstrated its great convenience and contactless sensing capabilities in supporting a broad array of applications. However, designing a ubiquitous WiFi sensing system for heterogeneous scenarios in practice is still a big dilemma as the system performs poorly under domain variations. In this project, we aim to investigate reliable WiFi sensing based
on a statistical assessment approach, named conformal prediction. The proposed approach quantifies the conformity between new testing WiFi samples and the training samples for prediction, which enhances the reliability of deep learning models without generating new features or retraining.