Multimodal human action recognition (HAR) utilizes complementary data for activity classification. Built on traditional HAR tasks, recent advances in Large Language Models (LLMs) enable detailed descriptions and causal reasoning of human actions, advancing new tasks of human action understanding (HAU) and human action reasoning (HARn). However, most LLMs, especially multimodal Large Vision-Language Models (LVLMs), struggle with modalities other than RGB images, like depth, IMU, ormmWave, due to a lack of large-scale datasets in these task domains. Existing HAR datasets provide only coarse-grained annotations, in-sufficient for depicting the detailed action dynamics required in HAU and HARn tasks. Simply combining annotations and generating captions with LLMs often lacks necessary logical and spatiotemporal consistency. In this paper, we introduce CUHK-X, a large-scale multi-modal dataset and benchmarks for HAR, HAU, and HARn. It includes 64,267 samples of 40 actions performed by 30 participants across two indoor environments, covering diverse daily scenarios. To address the challenge of spatiotemporal inconsistencies in captions, we propose a prompt-based scene creation method that leverages LLMs to generate logically connected activity sequences. CUHK-X also includes three benchmarks with six tasks to evaluate state-of-the-art models. Experimental results show average accuracies of 76.52% for HAR, 40.76% for HAU, and 70.25% for HARn. This large-scale multimodal dataset aims to empower the research community to apply, develop, and adapt data-intensive learning techniques for a wide range of human activity-related tasks.

Federated Learning (FL) empowers wireless intelligent applications, by leveraging distributed data of edge clients for training without compromising privacy. Client selection is inevitable in FL, since clients have diversified learning capabilities arising from heterogeneous computing and communication resources. Existing methods like fair-selection and dropping-straggler are either inefficient or unfair (resulting in a less effective trained model). Therefore, we propose FedAoI, an Age-of-Information (AoI) based client selection policy. FedAoI ensures fairness by allowing all clients, including stragglers, to submit their model updates while maintaining high training efficiency by keeping round completion times short. This trade-off is achieved by minimizing Peak-AoI (PAoI), the interval between a client's consecutive participations. An optimization problem is formulated by minimizing the Expected-Weighted-Sum-of-PAoI. This NP-hard problem is addressed with a two-step sub-optimal algorithm, PriorS. It first calculates client priority in a round using Lyapunov optimization and then selects the highest-priority clients through G-FPFC (Greedy minimization of the round weighted-sum-of-PAoI with First-Priority-First-Considered). Simulation results demonstrate that, compared to fair-selection, FedAoI improves average efficiency by 83.8% and achieves an average model accuracy of 97.3% (or at the cost of averaging 2.7% degradation in model accuracy). Compared to dropping-straggler, FedAoI reduces the average model accuracy degradation from 9.5% to 2.7%.