Recently, integrating video foundation models and large language models to build a video understanding system can overcome the limitations of specific vision tasks. Yet, existing methods either employ complex spatial-temporal modules or rely heavily on additional perception models to extract temporal features for video understanding, performing well only on short videos. For long videos, the computational complexity and memory costs associated with long-term temporal connections are significantly increased, posing additional challenges. Leveraging the hierarchical memory structure of the Atkinson-Shiffrin memory model, with tokens in Transformers being employed as the carriers of memory in combination, we propose MovieChat within a training-free memory consolidation mechanism to overcome these challenges, which transfers dense frames from short-term memory into sparse tokens in long-term memory by temporally merging adjacent frames. We lift pre-trained large multi-modal models for understanding long videos without additional trainable modules, employing a zero-shot approach. Additionally, in our new version, MovieChat+, we design an enhanced training-free vision-question matching-based memory consolidation mechanism to better anchor predictions to relevant visual content. MovieChat achieves state-of-the-art performance in long video understanding, along with the released MovieChat-1 K benchmark with 1 K long video, 2 K temporal grounding labels, and 14 K manual annotations.
The emergence of Large Vision-Language Models (LVLMs) marks significant strides towards achieving general artificial intelligence. However, these advancements are accompanied by concerns about biased outputs, a challenge that has yet to be thoroughly explored. Existing benchmarks are not sufficiently comprehensive in evaluating biases due to their limited data scale, single questioning format and narrow sources of bias. To address this problem, we introduce VLBiasBench, a comprehensive benchmark designed to evaluate biases in LVLMs. VLBiasBench features a dataset that covers nine distinct categories of social biases, including age, disability status, gender, nationality, physical appearance, race, religion, profession, social economic status, as well as two intersectional bias categories: race × gender and race × social economic status. To build a large-scale dataset, we use Stable Diffusion XL model to generate 46,848 high-quality images, which are combined with various questions to create 128,342 samples. These questions are divided into open-ended and close-ended types, ensuring thorough consideration of bias sources and a comprehensive evaluation of LVLM biases from multiple perspectives. We conduct extensive evaluations on 15 open-source models as well as two advanced closed-source models, yielding new insights into the biases present in these models.
Deep learning technologies are pivotal in enhancing the performance of WiFi-based wireless sensing systems. However, they are inherently vulnerable to adversarial perturbation attacks, and regrettably, there is lacking serious attention to this security issue within the WiFi sensing community. In this paper, we elaborate such an attack, called WiIntruder, distinguishing itself with universality, robustness, and stealthiness, which serves as a catalyst to assess the security of existing WiFi-based sensing systems. This attack encompasses the following salient features: (1) Maximizing transferability by differentiating user-state-specific feature spaces across sensing models, leading to a universally effective perturbation attack applicable to common applications; (2) Addressing perturbation signal distortion caused by device synchronization and wireless propagation when critical parameters are optimized through a heuristic particle swarm-driven perturbation generation algorithm; and (3) Enhancing attack pattern diversity and stealthiness through random switching of perturbation surrogates generated by a generative adversarial network. Extensive experimental results confirm the practical threats of perturbation attacks to common WiFi-based services, including user authentication and respiratory monitoring.
Reconfigurable antennas (RAs), capable of dynamically adapting their radiation patterns, polarization states, and operating frequencies, have emerged as a promising technology to meet the stringent performance requirements of sixth-generation (6G) wireless networks. This article systematically introduces essential hardware implementations of RAs and investigates advanced array architectures, such as fully-digital and tri-hybrid designs, emphasizing their capability to synergistically integrate electromagnetic (EM) reconfigurability with analog and digital signal processing. By facilitating coordinated beamforming across the EM and signal processing domains, RA arrays offer substantial flexibility and adaptability compared to conventional static antenna systems. Representative applications empowered by RA arrays, including integrated sensing and communication (ISAC), physical layer security (PLS), and near-field communications, are highlighted. A case study illustrates the effectiveness of RA arrays in optimizing beam steering, improving link robustness, and alleviating system power consumption. Finally, several open challenges and future research directions are outlined, emphasizing the need for advancements in theoretical modeling, hardware reliability, channel estimation techniques, intelligent optimization methods, and innovative network architectures, to fully realize the transformative impact of RAs in future 6G wireless networks.
As embodied AI systems become increasingly multi-modal, personalized, and interactive, they must learn effectively from diverse sensory inputs, adapt continually to user preferences, and operate safely under resource and privacy constraints. These challenges expose a pressing need for machine learning models capable of swift, context-aware adaptation while balancing model generalization and personalization. Here, two methods emerge as suitable candidates, each offering parts of these capabilities: multi-modal multi-task foundation models (M3T-FMs) provide a pathway toward generalization across tasks and modalities, whereas federated learning (FL) offers the infrastructure for distributed, privacy-preserving model updates and user-level model personalization. However, when used in isolation, each of these approaches falls short of meeting the complex and diverse capability requirements of real-world embodied AI environments. In this vision paper, we introduce multi-modal multi-task federated foundation models (M3T-FFMs) for embodied AI, a new paradigm that unifies the strengths of M3T-FMs with the privacy-preserving distributed training nature of FL, enabling intelligent systems at the wireless edge. We collect critical deployment dimensions of M3T-FFMs in embodied AI ecosystems under a unified framework, which we name “EMBODY”: Embodiment heterogeneity, Modality richness and imbalance, Bandwidth and compute constraints, On-device continual learning, Distributed control and autonomy, and Yielding safety, privacy, and personalization. For each, we identify concrete challenges and envision actionable research directions. We also present an evaluation framework for deploying M3T-FFMs in embodied AI systems, along with the associated trade-offs. Finally, we present a prototype implementation of M3T-FFMs and evaluate their energy and latency performance. To foster further research in this largely untapped area, we share our implementation through an open-source repository (GitHub: https://github.com/payamsiabd/M3T-FFM-EmbodiedAI).
