Semi-supervised Graph Instance Transformer for Mental Health Inference

Abstract

Mobile sensing appears as a promising solution for health inference problem (e.g., influenza-like symptom recognition) by leveraging diverse smart sensors to capture fine-grained information about human behaviors and ambient contexts. Centralized training of machine learning models can place mobile users' sensitive information under privacy risks due to data breach and misexploitation. Federated Learning (FL) enables mobile devices to collaboratively learn global models without the exposure of local private data. However, there are challenges of on-device FL deployment using mobile sensing: 1) long-term and continuously collected mobile sensing data may exhibit domain shifts as sensing objects (e.g. humans) have varying behaviors as a result of internal and/or external stimulus; 2) model retraining using all available data may increase computation and memory burden; and 3) the sparsity of annotated crowd-sourced data causes supervised FL to lack robustness. In this work, we propose FedMobile, an incremental semi-supervised federated learning algorithm, to train models semi-supervisedly and incrementally in a decentralized online fashion. We evaluate FedMobile using a real-world mobile sensing dataset for influenza-like symptom recognition. Our empirical results show that FedMobile-trained models achieve the best results in comparison to the selected baseline methods.

Graph Neural Networks in IoT: A Survey

“…The core motivation of using GNNs for modeling human behavior dynamics is that graph structured information can capture complex interactions among human behaviors by generating explicit topological representations to enhance the expressive power of the sensory data to further improve the prediction performance. Various works have demonstrated improved performance by applying GNNs in modeling human behaviors using the sensory data compared to other non-GNN approaches, particularly in domains such as social interaction detection, mobility prediction, cognition and physical activity recognition [ [64]. We generalize the process of using GNNs to model HSD as a framework and show it in Fig.…”

“…Graph modeling can transform unstructured human state information into graph structured representation that connects individual human states effectively. In general, there are two different classes of graph modeling in HSD: 1) using graphs to generate explicit representation of human [61,63,64] produced graph representations from multi-channel human behaviors, including social interaction, mobility, and physical activity, and used Bluetooth encounters as nodes in social interaction graphs, visited places as nodes in mobility trajectory graphs, and body movement as nodes in physical activity transition graphs. In the studies of monitoring brain electrical activities, electrical signals produced from segmented regions on human scalp indicate the neuron activities in the corresponding brain region [186].…”

“…In graph representation of human behavior (body) dynamics or latent human state transitions, edge connections describe the mechanism about human state (behavior) changes in their daily lives. For example, in human mobility trajectories, edge connections between two nodes in the trajectories imply people traveling from one place to another, as demonstrated in the works [60,61,63,64]. Dong et al linked sequential human states (mobility behavior, physical activity, social interaction) to construct graphs.…”

“…For instance, Wang et al [278] comprehensively reviewed how mobile sensing technologies affect personal and population health. Some existing research used sensor clustering as nodes and modeled behaviors based on the input sensor graphs [61,64]. Other works directly used sensors as graph nodes, and modeled the interactions between each sensor pair as edges, have proven to be helpful in fluid intake monitoring [255,256], sleep stage detection [110], and structural health monitoring [26].…”

“…The first type of GNNs for HSD can be characterized as static GNNs, in which the authors consider the spatial features of the topology in human states and sensor networks, as demonstrated in these works[60,61,106,145,176,246,256,269,327]. The second type of GNNs for HSD can be characterized as dynamic GNNs, in which the authors consider both spatial and temporal variations of the topology in human states and sensor networks, as demonstrated in these works[63,64,94,155,255]. Static GNNs: topological variations in HSD graph representations cause human state information to have different message transit pathways when graph convolutions activate neural information propagation.…”