Multimodal large language models have shown powerful abilities to understand and reason across text and images, but their massive size and computational cost limit real-world deployment. This research systematically examines how multimodal models can be made more efficient without severely sacrificing performance. By analyzing lightweight architectures, visual token compression strategies, efficient training methods, and compact language backbones, the study maps out the technical pathways that reduce memory demand and inference cost. The work highlights how efficiency-focused design enables multimodal intelligence to move beyond cloud-based systems toward broader, more accessible applications, including mobile devices and edge computing environments.

Hair loss and graying, the earliest visible hallmarks of skin aging, result from the functional decline of hair follicle stem cells (HFSCs) and their niche. Dr. Zhao and colleagues conducted a comprehensive analysis of human scalp samples using single-cell RNA sequencing (11 samples, 57,181 cells in total) and spatial transcriptomics (1 sample) to detail the mechanisms involved. The study confirmed the transitional stages of three mitotic keratinocyte subtypes. Comparison of middle-aged and young scalps revealed three key age-associated changes: activated AP-1 transcription factor complex in keratinocytes; up-regulated DCT gene in melanocytes; and a dramatic decrease in BMP and non-canonical WNT (ncWNT) signaling within the critical dermal papilla-keratinocyte crosstalk. This breakdown of essential inter-cellular communication and activation of stress signals provides valuable, cell-resolved insights into hair follicle aging, supporting the development of future regenerative therapies targeting these pathways.

Large-scale Low Earth Orbit (LEO) constellations have become a focal point for providing round-the-clock high-fidelity information services. However, their efficient and economical batch deployment faces severe challenges from growing demands and multiple constraints, with existing methods struggling to address the computational complexity in large-scale scenarios. To meet this pressing need, this study published in the Chinese Journal of Aeronautics proposes an innovative deployment optimization framework. At its core, it constructs a novel partial time-expanded network and employs an efficient hybrid algorithm to significantly reduce constraint explosion, enhancing solution efficiency and scalability. The framework supports dual-channel, multi-configuration rocket strategies and flexible deployment under multiple mission triggers through weighted optimization. Ultimately, it effectively reduces deployment costs, improves optimization efficiency, and provides reliable decision support for large-scale constellation deployment.

The challenge of resource allocation for UAV swarms in dynamic and uncertain electromagnetic environments has been investigated for years. In a recent breakthrough published in the Chinese Journal of Aeronautics, a novel intelligent decision-making framework that addresses incomplete interference information has emerged. This innovative framework integrates fuzzy logic for uncertainty modeling, dynamic constrained multi-objective optimization, and transfer learning, enabling UAV swarms to achieve autonomous and efficient spectrum allocation under rapidly changing conditions while maintaining both communication performance and security.

Gust load alleviation is a crucial topic for the practical application of high-altitude long-endurance unmanned aerial vehicles. Passive flexible wingtips mitigate gust loads by naturally adapting their shape to airflow disturbances. Without active control or energy input, they passively relieve aerodynamic peaks, smooth transient loads, and enhance flight stability and structural safety, offering a lightweight, reliable, and energy-efficient gust alleviation solution.