The integration of egocentric (ego) and exocentric (exo) perspectives offers a promising avenue for advancing video understanding, yet the domain gap between these views presents significant challenges. In this work, we introduce EgoExoChat, a novel framework designed to bridge this gap by effectively fusing complementary information from both perspectives. Our approach incorporates an ego–exo token selector, which dynamically identifies and aligns informative tokens from ego and exo videos, ensuring fine-grained interaction details are preserved while maintaining global scene context. Furthermore, we enhance the Ego-Exo4D dataset by introducing dense video captioning (DVC) and refined annotations, enabling improved multiview reasoning. Experimental results demonstrate that EgoExoChat outperforms prior models across multiple video understanding tasks, showcasing superior generalization capabilities in complex video scenarios. These findings highlight the potential of ego–exo integration for advancing artificial intelligence (AI)-driven video reasoning and interaction modeling.
This article is concerned with data-driven control for a class of time-varying neural networks from noisy data. Here, the main contribution is that a dynamic data informatization scheme is established to represent time-varying neural networks. Different from previous works, a new data collection approach is proposed to collect T-instant data sequences by choosing different initial values. Then, an auxiliary function matrix is introduced to construct a dynamic data information equation, which is used to characterize the time-varying dynamics. Based on the established dynamic data information equation, several sufficient data-driven conditions are developed for time-varying neural networks under event-triggered control by employing Lyapunov stability analysis and an interval matrix approach. Without prior model parameters, the controller gains, trigger parameters, allowable parameter variation intervals, and data-driven Zeno-free behavior can be achieved for time-varying neural networks. The relationships among data sample size, allowable parameter variation intervals, and trigger parameters are well discussed. Finally, simulation comparison results and an application to image processing illustrate the effectiveness of the proposed data-driven control scheme.
Learning compact and discriminative representations of object relations is central to structured visual reasoning. In scene graph generation, these relations must be inferred from ambiguous visual cues under a heavy class imbalance; yet most methods encode a relation as a single real-valued vector that entangles interaction strength with a discriminative orientation, so frequent predicates overwrite the distinctions among rare ones. We propose CReaL, which fuses the subject–object pair with its context through complex multiplication, letting the context scale the magnitude and shift the phase of the relation. The result is split into an amplitude branch trained for interaction strength and a phase branch for the angular structure that discriminates similar relations, with branch-specific objectives that keep the two disentangled. A dynamic gating mechanism then fuses them with per-class weights computed from both branches, and the framework stays compatible with standard real-valued pipelines. On Visual Genome 150 (VG-150), CReaL improves mR@100 across predicate classification (PredCls), scene graph classification (SGCls), and scene graph detection (SGDet) over state-of-the-art methods, including a 5.9% gain in PredCls over the strongest knowledge-graph-based baseline despite using no external knowledge graph, and it attains the best overall score on Open Images v6 (OI-V6). Visualization and clustering analyses show that the amplitude and phase form complementary structures whose gated combination sharpens decision boundaries, indicating that explicit relational decomposition is an effective foundation for fine-grained relational understanding.
This article investigates output synchronization for discrete-time fractional-order Takagi-Sugeno fuzzy coupled neural networks (DFTSCNNs) with output couplings and time-varying delays under a fuzzy quantized controller. First, nonautonomous Halanay inequalities, both with and without the disturbance term, are proved in the discrete-time fractional-order (DF) framework. To achieve output synchronization of the proposed DFTSCNNs, we design a Takagi-Sugeno (T-S) fuzzy quantized controller. By combining linear matrix inequalities (LMIs), the Lyapunov direct method, and the proposed Halanay inequality, sufficient quantized output synchronization criteria are derived. Finally, the theoretical results are validated through three numerical examples, including an image-encryption application.
Recent studies have explored enhancing label efficiency in vision–language models (VLMs) for downstream tasks. Existing methods of pseudolabel learning (PL) would suffer from heavily incorrect hard pseudolabels. Nevertheless, active learning (AL) focuses on uncertain samples that can cause forgetting of VLMs’ existing knowledge. Can these two paradigms work together to enhance the performance of VLMs? To alleviate the issues mentioned above, we bridge the gap between the two via a resampling mechanism that balances the data distribution without increasing labeling costs. By offering a comprehensive and well-distributed labeled set, the downstream task models based on VLMs receive balanced gradient updates during training, which helps to reduce miscalibration and overfitting in models. Comprehensive experimental results demonstrate the effectiveness and general applicability of our method compared to state-of-the-art label-efficient approaches with VLMs across several benchmark datasets.
