Resource-Constrained Federated Learning with Heterogeneous Labels and Models for Human Activity Recognition

Abstract

One of the most significant applications in pervasive computing for modeling user behavior is Human Activity Recognition (HAR). Such applications necessitate us to characterize insights from multiple resource-constrained user devices using machine learning techniques for effective personalized activity monitoring. On-device Federated Learning proves to be an extremely viable option for distributed and collaborative machine learning in such scenarios, and is an active area of research. However, there are a variety of challenges in addressing statistical (non-IID data) and model heterogeneities across users. In addition, in this paper, we explore a new challenge of interest -to handle heterogeneities in labels (activities) across users during federated learning. To this end, we propose a framework with two different versions for federated label-based aggregation, which leverage overlapping information gain across activities -one using Model Distillation Update, and the other using Weighted α-update. Empirical evaluation on the Heterogeneity Human Activity Recognition (HHAR) dataset (with four activities for effective elucidation of results) indicates an average deterministic accuracy increase of at least ∼11.01% with the model distillation update strategy and ∼9.16% with the weighted α-update strategy. We demonstrate the ondevice capabilities of our proposed framework by using Raspberry Pi 2, a single-board computing platform.

Zero-Shot Federated Learning with New Classes for Audio Classification

“…We initially consider eight labels with an initial Public Dataset in both datasets before streaming new classes (Table 1). We simulate two scenarios for testing just our zero-shot framework -1) new classes only (homogeneous) with limited users and FL iterations (3 users and 10 FL iterations) for effective analysis of results, 2) new classes with statistical heterogeneities in both labels and models as performed in [13], (10 users and 30 FL iterations). This exhibits near-real-time model heterogeneities as shown in Table 2, and effective convergence.…”

“…Typically, statistical heterogeneities are widely observed in practical FL settings, hence Choice 1 handles heterogeneities in local and global update steps [13], while Choice 2 handles new classes in our proposed framework.…”

“…One way of handling model heterogeneities and independence in a federated learning setting is by using knowledge distillation [11] with a common student model architecture on each local device [12]. Label and model heterogeneities are handled in an inertial Human Activity Recognition scenario in [13]. Federated learning for keyword spotting [14], and new class learning and identification in various speech recognition settings are addressed in [15,16].…”

Deep Learning in Human Activity Recognition with Wearable Sensors: A Review on Advances

“…Bettini et al [281] presented a personalized semi-supervised federated learning method that built a global activity model and leveraged transfer learning for user personalization. Besides, Gudur and Perepu [282] implemented on-device federated learning using model distillation update and so-called weighted αupdates strategies to resolve model heterogeneities on a resource-limited embedded system (Raspberry Pi), which proved its effectiveness and efficiency. Opportunity: Blockchain is a new hot topic around the world.…”

Federated Learning for Smart Healthcare: A Survey

“…The proposed approach is examined on the Animals-10 dataset [75], showing that the classification accuracy is increased by about 16.7%. This approach is also tested on the human activity recognition dataset with an average increase of 9.153% 11.01% in [76] and [77], respectively. Recently, the work in [78] develops a personalized FL framework to predict the pain by using face images.…”