Key research themes
1. How can machine learning and deep learning models improve path loss prediction accuracy across varied environments and network parameters?
This theme investigates the application of machine learning (ML) and deep learning (DL) techniques to enhance the accuracy and versatility of path loss prediction models. Traditional empirical and deterministic models often fail to capture complex environmental variability or require extensive computations and measurements. ML/DL models seek to provide accurate, computationally efficient, and adaptable path loss predictions across multiple frequencies, antenna heights, and diverse propagation environments by leveraging data-driven approaches and environmental feature extraction.
2. What roles do the underlying loss landscapes and optimization properties of neural networks play in developing effective path loss prediction models?
This theme explores theoretical and empirical insights into the optimization landscape of neural networks used for path loss and related prediction tasks. Understanding the loss surface geometry, presence of saddle points, local minima, and training dynamics is crucial for developing robust machine learning models for path loss prediction. Improved knowledge about neural network convergence behavior can guide selection of architectures, regularization parameters, and training strategies to obtain models that generalize well to complex wireless environments.
3. How can domain-specific empirical and hybrid modeling approaches be developed for path loss prediction in complex propagation environments such as vegetated agricultural and urban macro settings?
This theme focuses on designing specialized models for path loss prediction tailored to complex real-world propagation environments, including vegetated farmlands and urban macro environments with diverse obstacles. It highlights the need to capture distinctive physical characteristics and environmental interactions affecting signal attenuation by combining empirical measurements with advanced model fitting, hybrid machine learning structures, and domain-informed parameterizations to improve prediction reliability and support practical wireless network design in challenging environments.