This paper addresses the challenges of feeder link switching in non-terrestrial networks (NTNs) by leveraging low Earth orbit (LEO) mega-constellations with multiple shells and multiple gateway (GW) deployments. Due to the high mobility of LEO satellites and frequent feeder link disruptions, it is critical to maintain reliable feeder link connectivity. To tackle these challenges, we propose a novel bilevel optimization framework that combines virtual network embedding (VNE) with advanced feeder link switching strategies. To manage the trade-off between link quality and switching frequency, lower-level optimization maximizes the data transmitted per feeder link connection using a path-finding algorithm with bisection search. The upper-level VNE problem enhances the connectivity between the satellites in the LEO mega-constellation and the ground infrastructure by constructing a dedicated virtual network for NTN GWs. Simulation results based on the Starlink Gen II constellation demonstrate that our proposed method consistently ensures 100% satellite activation, while baseline benchmarks achieve 93% satellite activation rates in the real-world GW deployments. Our optimization framework is well-suited for multiple shell deployments of LEO mega-constellations with multiple GW supports, paving the way for globally connected NTNs.

This paper proposes a globally scalable multicontroller software-defined networking (SDN) framework tailored for low Earth orbit (LEO) mega-constellation. To handle the rapidly changing topology and massive scale of such networks, the novel topology virtualization technique and dynamic mapping strategy are introduced. The framework creates a stable logical network abstraction of the moving satellites that shields the control-plane from constant topological changes. Unlike conventional approaches that require frequent intercontroller cooperation, we formulate a location-based linear assignment problem and derive a closed-form mapping solution to directly assign satellites to SDN controllers without additional cooperation. Extensive evaluations on the Starlink Gen II constellation demonstrate that our framework achieves efficient load balancing among Earth-fixed cells across LEO mega-constellation while significantly reducing computational complexity. Our main contribution is to address the critical challenges of mobility management in dynamic, global-scale LEO mega-constellation, thereby paving the way for enhanced global coverage and efficient use of resources in next-generation communication systems.

Multimodal learning has gained increasing importance across various fields, offering the ability to integrate data from diverse sources such as images, text, and personalized records, which are frequently observed in medical domains. However, in scenarios where some modalities are missing, many existing frameworks struggle to accommodate arbitrary modality combinations, often relying heavily on a single modality or complete data. This oversight of potential modality combinations limits their applicability in real-world situations. To address this challenge, we propose Flex-MoE (Flexible Mixture-of-Experts), a new framework designed to flexibly incorporate arbitrary modality combinations while maintaining robustness to missing data. The core idea of Flex-MoE is to first address missing modalities using a new missing modality bank that integrates observed modality combinations with the corresponding missing ones. This is followed by a uniquely designed Sparse MoE framework. Specifically, Flex-MoE first trains experts using samples with all modalities to inject generalized knowledge through the generalized router ($\mathcal{G}$-Router). The $\mathcal{S}$-Router then specializes in handling fewer modality combinations by assigning the top-1 gate to the expert corresponding to the observed modality combination. We evaluate Flex-MoE on the ADNI dataset, which encompasses four modalities in the Alzheimer's Disease domain, as well as on the MIMIC-IV dataset. The results demonstrate the effectiveness of Flex-MoE, highlighting its ability to model arbitrary modality combinations in diverse missing modality scenarios. Code is available at: \url{https://github.com/UNITES-Lab/flex-moe}.