Federated learning systems for mobile sensing data

Author

Xiaopeng Jiang, New Jersey Institute of TechnologyFollow

Author ORCID Identifier

0000-0002-0814-4514

Document Type

Dissertation

Date of Award

8-31-2024

Degree Name

Doctor of Philosophy in Computing Sciences - (Ph.D.)

Department

Computer Science

First Advisor

Cristian Borcea

Second Advisor

Yi Chen

Third Advisor

Xiaoning Ding

Fourth Advisor

Hai Nhat Phan

Fifth Advisor

Guy Jacobson

Abstract

Federated Learning (FL) has emerged as a new distributed Deep Learning (DL) paradigm that enables privacy-aware training and inference on mobile devices with help from the cloud. This dissertation presents a comprehensive exploration of FL with mobile sensing data, covering systems, applications, and optimizations.

First, a mobile-cloud FL system, FLSys, is designed to balance model performance with resource consumption, tolerate communication failures, and achieve scalability. In FLSys, different DL models with different FL aggregation methods can be trained and accessed concurrently by different apps. In addition, FLSys provides advanced privacy-preserving mechanisms and a common API for third-party app developers to access FL models. FLSys adopts a modular design and is implemented in Android and AWS cloud. Extended from FLSys, ZoneFL exploits a mobile-edge-cloud architecture to adapt models to user behaviors in different geographical zones to further improve scalability and model utility. Both FLSys and ZoneFL are evaluated with real-world deployments to showcase the superior model performance, scalability, and fault-tolerance.

Second, Federated Meta-Location Learning (FMLL) is proposed on smart phones for fine-grained location prediction, based on GPS traces collected on the phones. FMLL has three components: a meta-location generation module, a prediction model, and a FL framework. The meta-location generation module represents the user location data as relative points in an abstract 2-Dimensional (2D) space, which enables learning across different physical spaces. The model fuses Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNN) layers, where BiLSTM learns the speed and direction of the mobile users, and CNN learns information such as user movement preferences. FMLL uses federated learning to protect user privacy and reduce bandwidth consumption.

Third, Complement Sparsification (CS) is presented as an FL pruning mechanism that achieves low bidirectional communication overhead between the server and the clients, low computation overhead at the clients, and good model accuracy. CS uses a complementary and collaborative pruning at the server and the clients. At each round, CS creates a global sparse model that contains the weights that capture the general data distribution of all clients, while the clients create local sparse models with the weights pruned from the global model to capture the local trends. For improved model performance, these two types of complementary sparse models are aggregated into a dense model in each round, which is subsequently pruned in an iterative process.

Fourth, Federated Continual Learning (FCL) is explored as a more intricate FL scenario wherein data accumulates over time and undergoes distributional changes. A framework, Concept Matching (CM), is introduced for efficient FCL. The CM framework groups client models into model clusters, and then uses novel CM algorithms to build different global models for different concepts in FL over time. In each round, the server sends the global concept models to the clients. To avoid catastrophic forgetting, each client selects the concept model best-matching the implicit concept of the current data for fine-tuning. To avoid interference among client models with different concepts, the server clusters the models representing the same concept, aggregates the model weights in each cluster, and updates each global concept model with a cluster model of the same concept. Since the server does not know the concepts captured by the aggregated cluster models, a theoretical grounded server CM algorithm is proposed to effectively update a global concept model with a matching cluster model.

Recommended Citation

Jiang, Xiaopeng, "Federated learning systems for mobile sensing data" (2024). Dissertations. 1774.
https://digitalcommons.njit.edu/dissertations/1774