Browsing by Author "Li, Boning"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Extraction and Interpretation of Deep Autoencoder-based Temporal Features from Wearables for Forecasting Personalized Mood, Health, and Stress(2020-08-20) Li, Boning; Sano, Akane; Veeraraghavan, AshokHigh-resolution wearable sensor data contain physiological and behavioral information that can be utilized to predict and eventually improve human health and wellbeing. We propose a semi-supervised deep neural network framework to automatically learn features from passively collected multi-modal sensor data. This process can be personalized by finetuning the general features with participant-specific data. Then, using the learned features, we performed personalized prediction of subjective wellbeing scores with high precision. We also provide visual explanation and statistical interpretation of the automatically learned features and the prediction models. In this study, we explored multiple implementations of our framework including locally connected linear network, convolutional neural network, recurrent neural network, and visual attention network. The framework was evaluated using wearable sensor data and wellbeing labels collected from college students (total 6391 days from N=239). Sensor data include skin temperature, skin conductance, and acceleration; wellbeing scores include self-reported mood, health and stress ranged from 0 to 100. Compared to the prediction performance based on hand-crafted features, the proposed framework achieved higher precision with a smaller number of features. Our results show promising potentials of predicting self-reported mood, health, and stress accurately using an interpretable deep learning framework, ultimately for developing real-time health and wellbeing monitoring and intervention systems that can benefit various populations.Item Graph-based Learning for Efficient Resource Allocation and Management in Wireless Networks(2024-10-07) Li, Boning; Segarra, SantiagoWireless network optimization faces challenges in several aspects. System-level objective functions are typically non-convex due to multiuser interference, rendering NP-hard problems. Moreover, real-world factors like delay and energy requirements could translate into non-convex constraints. As a result, traditional practices tend to be suboptimal and often rely on computationally expensive iterations or simulations. This inefficiency limits the design and management of scalable wireless networks. To improve efficiency, this thesis leverages deep learning techniques, with a focus on the permutation-equivariant graph neural networks (GNNs), across various wireless applications. The first part of this thesis presents a graph-based trainable framework for power allocation, namely the unfolded successive concave approximation (USCA), to maximize weighted-sum energy efficiency (WSEE) in wireless interference networks. The second part revisits power allocation and, by devising a primal-dual (PD) learning framework, handles non-convex constraints specific to the context of wireless federated learning (FL). Finally, the third part creates a learnable digital twin (DT) for efficient network evaluation, a thousandfold more efficient tool than simulators for network design and management. These three bodies of work demonstrate comprehensive theoretical and experimental results highlighting the superior performance of the proposed architectures over existing approaches. Overall, this work seeks to integrate deep learning into wireless applications by innovating specialized architectures that retain the core structure of original regimes. Key advantages include better interpretability than non-specialized end-to-end learning, enhanced generalizability across varying network configurations, and faster inference speed compared to optimization-based methods and simulators. The methodologies presented herein have substantial potential for diverse wireless applications, including network design, resource allocation, and network management.