Current multimodal large language models (MLLMs) struggle with hour-level video understanding, facing significant challenges not only in modeling the substantial information volume of long videos but also in overcoming the memory wall and resource constraints during both training and inference. Although recent training-free approaches have alleviated resource demands by compressing visual features, their reliance on incomplete visual information limits the performance potential. To address these limitations, we propose Adaptive Pivot Visual information Retrieval (APVR), a training-free framework that hierarchically retrieves and retains sufficient and important visual information. It breakthroughs the memory wall limitation via two complementary components: Pivot Frame Retrieval employs query expansion and iterative spatio-semantic confidence scoring to identify relevant video frames, and Pivot Token Retrieval performs query-aware attention-driven token selection within up to 1024 pivot frames. This dual granularity approach enables the processing of hour-long videos while maintaining semantic fidelity. Experimental validations on three different baseline MLLMs demonstrate significant performance improvements up to 9.5%, 4.6% and 9.7% on LongVideoBench, VideoMME and MLVU, respectively. APVR achieves state-of-the-art results for both training-free and training-based approaches.

Continual graph learning is rapidly emerging as an important role in a variety of real-world applications such as online product recommendation systems and social media. While achieving great success, existing works on continual graph learning ignore the information from multiple modalities (e.g., visual and textual features) as well as the rich dynamic structural information hidden in the ever-changing graph data and evolving tasks. However, considering multimodal continual graph learning with evolving topological structures poses great challenges: i) it is unclear how to incorporate the multimodal information into continual graph learning and ii) it is nontrivial to design models that can capture the structure-evolving dynamics in continual graph learning. To tackle these challenges, in this paper we propose a novel Multimodal Structure-evolving Continual Graph Learning (MSCGL) model, which continually learns both the model architecture and the corresponding parameters for Adaptive Multimodal Graph Neural Network (AdaMGNN). To be concrete, our proposed MSCGL model simultaneously takes social information and multimodal information into account to build the multimodal graphs. In order for continually adapting to new tasks without forgetting the old ones, our MSCGL model explores a new strategy with joint optimization of Neural Architecture Search (NAS) and Group Sparse Regularization (GSR) across different tasks. These two parts interact with each other reciprocally, where NAS is expected to explore more promising architectures and GSR is in charge of preserving important information from the previous tasks. We conduct extensive experiments over two real-world multimodal continual graph scenarios to demonstrate the superiority of the proposed MSCGL model. Empirical experiments indicate that both the architectures and weight sharing across different tasks play important roles in affecting the model performances.

This article presents a linear active disturbance rejection control (LADRC) scheme for two-mass systems (TMSs) based on a singular perturbation (SP) approach. In the proposed scheme, the dynamics of TMSs are separated into a quasi-steady-state system at a slow time scale and a boundary-layer system at a fast time scale via the SP method. The quasi-steady-state model is used as the nominal model to design the LADRC, where internal model control (IMC) rules are employed for prescribed tracking performance. This not only reduces the model order but also makes the gain tuning straightforward, which is crucial for practical use. Based on the boundary-layer model, the active damping method is designed for vibration suppression, which aims at achieving a higher stable bandwidth of the IMC to better attenuate the residual disturbance of linear extended state observer such that the tracking accuracy can be improved. Stability of the full system was analyzed via the extended Tikhonov's theorem. Comparative experiments were carried out to verify the effectiveness of the proposed scheme.

Value iteration networks (VINs) have been demonstrated to have a good generalization ability for reinforcement learning tasks across similar domains. However, based on our experiments, a policy learned by VINs still fail to generalize well on the domain whose action space and feature space are not identical to those in the domain where it is trained. In this paper, we propose a transfer learning approach on top of VINs, termed Transfer VINs (TVINs), such that a learned policy from a source domain can be generalized to a target domain with only limited training data, even if the source domain and the target domain have domain-specific actions and features. We empirically verify that our proposed TVINs outperform VINs when the source and the target domains have similar but not identical action and feature spaces. Furthermore, we show that the performance improvement is consistent across different environments, maze sizes, dataset sizes as well as different values of hyperparameters such as number of iteration and kernel size.

Today social media has become the primary source for news. Via social media platforms, fake news travel at unprecedented speeds, reach global audiences and put users and communities at great risk. Therefore, it is extremely important to detect fake news as early as possible. Recently, deep learning based approaches have shown improved performance in fake news detection. However, the training of such models requires a large amount of labeled data, but manual annotation is time-consuming and expensive. Moreover, due to the dynamic nature of news, annotated samples may become outdated quickly and cannot represent the news articles on newly emerged events. Therefore, how to obtain fresh and high-quality labeled samples is the major challenge in employing deep learning models for fake news detection. In order to tackle this challenge, we propose a reinforced weakly-supervised fake news detection framework, i.e., WeFEND, which can leverage users' reports as weak supervision to enlarge the amount of training data for fake news detection. The proposed framework consists of three main components: the annotator, the reinforced selector and the fake news detector. The annotator can automatically assign weak labels for unlabeled news based on users' reports. The reinforced selector using reinforcement learning techniques chooses high-quality samples from the weakly labeled data and filters out those low-quality ones that may degrade the detector's prediction performance. The fake news detector aims to identify fake news based on the news content. We tested the proposed framework on a large collection of news articles published via WeChat official accounts and associated user reports. Extensive experiments on this dataset show that the proposed WeFEND model achieves the best performance compared with the state-of-the-art methods.

A previously developed pneumatic muscles' (PMs) actuated gait exoskeleton (with only knee joint) has been demonstrated in achieving appropriate actuation torque, range of motion (ROM), and control bandwidth for task-specific gait training. While the adopted multi-input-multi-output (MIMO) sliding mode (SM) strategy has preliminarily implemented simultaneous control of the exoskeleton's angular trajectory and compliance, its efficacy with human users during gait cycles has not been investigated. This article presents an improved bi-joint gait rehabilitation exoskeleton (BiGREX) with integrated human hip and knee joints. The results with 12 healthy subjects demonstrated that the system's compliance can be effectively adjusted while guiding the subjects walking in predefined trajectories.

In this paper, a scenario of full-duplex (FD)-mode relay aided device-to-device (D2D) communications network is investigated, with base station (BS) equipped in an unmanned aerial vehicle (UAV). The FD-mode relay is capable of improving the capacity of the proposed heterogeneous network. Furthermore, the interference imposed on the conventional cellular users (CUs) by the D2D users, which may erode the Signal-to-Interference-plus-Noise-Ratio (SINR) of each CU, is analyzed, with a limited transmit power of the UAV BS considered. Furthermore, the expression of UAV BS's downlink transmission is derived, following which the influence of mobile UAV BS's coverage on the performance of the proposed heterogeneous network is analyzed.

The impact of both maximal ratio combining (MRC) and relay selection on the physical layer security in wireless communication systems is studied by analyzing some important factors, including the probability characteristic of the legitimate receiver (Bob) and malicious eavesdropper (Eve)'s end-to-end signal-to-noise ratio (SNR), the secrecy outage probability, and the average secrecy channel capacity over Rayleigh fading Channel. We assume that Bob receives its data from both the relay and the source by cooperative communication, provided that MRC is employed at the receiver. Compared to the conventional MRC methods, a higher spatial diversity order can be exploited by performing relay selection in the proposed method, as validated via both theoretical analysis and numerical simulation. The theoretical closed-form expressions for some figures of merit, e.g., the secrecy outage probability and the average secrecy capacity, are all consistent with the numerical results.

The performance analysis of optimal relay decode-and-forward (DF) selection for both the full-duplex (FD) and half-duplex (HD) relaying modes is studied, with some important factors such as the distributions of the received signal-to-noise ratio (SNR), the outage probability and the average channel capacity, etc., being taken into account. Different from the conventional relay selection schemes, the trade-off between the FD and HD modes is studied, with the former suffering from the impact of residual self-interference while the latter consuming more channel resources than the former by allocating two orthogonal channels for transmission and reception. The optimal power allocation (OPA) subject to individual power constrains (IPC) and sum power constrains (SPC) are also analyzed in the proposed FD scheme. In particular, the exact closed-form expressions for outage probability of the proposed FD relay selection scheme over independent and identically distributed (i.i.d.) Rayleigh fading channels are derived in this paper, with the validity of the proposed analysis being proven by simulation. It is also shown that the proposed FD scheme outperforms the HD mode in terms of average channel capacity by about 33.1%, provided that the self-interference can be successfully suppressed below the noise power level.

The impact of an imperfect channel estimation on the amplify-and-forward (AF) mode cooperative communications systems is studied, with some important factors, including the probability characteristic of the secondary user's end-to-end signal-to-noise ratio (SNR), the outage probability, the symbol error probability (SEP) and the lower bound on the capacity, etc, being analyzed. As compared to the conventional relay selection schemes, less channel state information (CSI) feedback is required in the proposed method due to an outdated channel estimation being tolerable. The proposed scheme is validated by carrying out both theoretical analysis and numerical simulation, and the theoretical closed-form expressions for some figures of merit, including the outage probability, the SEP and the lower bound on the capacity, are consistent with the numerical results. The simulation results also prove that the performance of the proposed scheme is impacted considerably by some other critical parameters, including the number of relays, the channel correlation coefficient and the interference threshold. In the presence of multiple candidate relays, an optimum solution in terms of either outage probability or SEP performance can always be found within the SNR range of [0dB, 10dB].

The impact of outdated channel state information (CSI) on the capacity of amplify-and-forward (AF) partial relay selection systems is studied in this paper. The closed-form expressions for the distribution of received signal-to-noise ratio (SNR) in a multi-relay cooperative communications system is first derived, with independent and identically distributed (i.i.d.) Rayleigh fading channels being assumed in each wireless link. After that, the theoretical closed-form expressions for both outage probability and channel capacity of partial relay selection are derived, with four classical adaptive transmission techniques, including the constant power with optimal rate adaption (ORA), the optimal power and rate adaption (OPRA), the channel inversion with fixed rate (CIFR) and truncated channel inversion with fixed rate (TIFR), being considered. Numerical analysis proves that the channel capacity of partial relay selection is impacted considerably by some critical parameters, including the number of relays, the channel correlation coefficient and the end-to-end SNR, etc. It's also exhibited in the numerical results that among the four adaptive transmission techniques, the diversity order of OPRA is larger than that of TIFR, and the OPRA outperforms TIFR with about 0.15 bits/s/Hz in terms of average channel capacity.