Yifei Gao, Junhong Ye, Yifan Yang, Jiaqi Wang, Yi Zhang, Zhang Ruichen, Jitao Sang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Due to their multimodal capabilities, Vision-Language Models (VLMs) have found numerous impactful applications in real-world scenarios. However, recent studies have revealed that VLMs are vulnerable to image-based adversarial attacks. Traditional targeted adversarial attacks require specific targets and labels, limiting their real-world impact. We present AnyAttack, a self-supervised framework that transcends the limitations of conventional attacks through a novel foundation model approach. By pretraining on the massive LAION-400M dataset without label supervision, AnyAttack achieves unprecedented flexibility - enabling any image to be transformed into an attack vector targeting any desired output across different VLMs. This approach fundamentally changes the threat landscape, making adversarial capabilities accessible at an unprecedented scale. Our extensive validation across five open-source VLMs (CLIP, BLIP, BLIP2, InstructBLIP, and MiniGPT-4) demonstrates AnyAttack’s effectiveness across diverse multimodal tasks. Most concerning, Any-Attack seamlessly transfers to commercial systems including Google Gemini, Claude Sonnet, Microsoft Copilot and OpenAI GPT, revealing a systemic vulnerability requiring immediate attention.

Cross-video question answering presents significant challenges beyond traditional single-video understanding, particularly in establishing meaningful connections across video streams and managing the complexity of multi-source information retrieval. We introduce VideoForest, a novel framework that addresses these challenges through person-anchored hierarchical reasoning, enabling effective cross-video understanding without requiring end-to-end training. VideoForest integrates three key innovations: 1) a human-anchored feature extraction mechanism that employs ReID and tracking algorithms to establish robust spatiotemporal relationships across multiple video sources; 2) a multi-granularity spanning tree structure that hierarchically organizes visual content around person-level trajectories; and 3) a multi-agent reasoning framework that efficiently traverses this hierarchical structure to answer complex queries. To evaluate our method, we develop CrossVideoQA, a comprehensive benchmark specifically designed for person-centric cross-video analysis. Experimental results demonstrate VideoForest's superior performance in cross-video reasoning tasks, achieving 71.93% accuracy in person recognition, 83.75% in behavior analysis, and 51.67% in summarization and reasoning.

The widespread deployment of Remote Direct Memory Access (RDMA) in datacenter networks increases the stringency for convergence speed when congestion occurs. Fast convergence significantly reduces buffer occupancy, which in turn lessens the probability of triggering Priority-based Flow Control (PFC). Besides, the propagation delay becomes shorter with rapidly growing link speed, which correspondingly makes the queueing delay a major part of end-to-end latency in datacenter networks. Fast convergence and low buffer occupancy become more essential for lowering queue delay and flow complete time. In this paper, we present ERA, an ecn-ratio-based congestion control scheme, which contributes to fast convergence for datacenter networks. ERA consists of two fundamental components: (i) an ECN-marking-ratio-based queue buffer occupancy estimating (QBOE) solution and (ii) a queue-building-rate driven rate adjustment (QDRA) mechanism to achieve fast convergence in several control periods. We conduct extensive experiments to evaluate the performance of ERA, and the results show that ERA greatly accelerates the convergence process compared to other solutions. ERA achieves low tail latency and low buffer occupancy simultaneously.

RDMA has been widely adopted by distributed applications for its low latency, high throughput, and near-zero CPU overhead. However, high availability is still lacked for RDMA-based applications, especially in the cases of network failure or NIC upgrade. In this paper, we propose DTS, a scheme for RDMA-based applications to provide high availability by adopting the active and standby transports and switching them transparently and fast when necessary. We implement the prototype of DTS based on the Libfabric RxD provider. Our evaluation demonstrates that with DTS existing RDMA-based applications can be deployed with zero downtime and the performance degradation is ignorable.

The widespread deployment of Remote Direct Memory Access (RDMA) in datacenter networks increases the stringency for convergence speed when congestion occurs. Fast convergence significantly reduces buffer occupancy, which in turn lessens the probability of triggering Priority-based Flow Control (PFC). Besides, the propagation delay becomes shorter with rapidly growing link speed, which correspondingly makes the queueing delay a major part of end-to-end latency. Fast convergence and low buffer occupancy become more essential for lowering queue delay and flow complete time. We present DQCC (Double-Q Congestion Control), a fast-converging congestion control scheme, which consists of two fundamental components: (i) an ECN-marking-ratio-based queue buffer occupancy estimating (QBOE) solution and (ii) a queue-building-rate driven rate adjustment (QDRA) mechanism to achieve fast convergence. We conduct extensive experiments to evaluate the performance of DQCC, and the results show that DQCC greatly accelerates the convergence process. DQCC achieves low tail latency and low buffer occupancy simultaneously.

Heterogeneous Network (HetNet), where Small cell Base Stations (SBSs) are densely deployed to offload traffic from macro Base Stations (BSs), is identified as a key solution to meet the unprecedented mobile traffic demand. The high density of SBSs are designed for peak traffic hours and consume an unnecessarily large amount of energy during off-peak time. In this paper, we propose a Deep Reinforcement-Learning (DRL) based SBS activation strategy that activates the optimal subset of SBSs to significantly lower the energy consumption without compromising the quality of service. In particular, we formulate the SBS on/off switching problem into a Markov Decision Process that can be solved by Actor Critic (AC) reinforcement learning methods. To avoid prohibitively high computational and storage costs of conventional tabular-based approaches, we propose to use deep neural networks to approximate the policy and value functions in the AC approach. Moreover, to expedite the training process, we adopt a Deep Deterministic Policy Gradient (DDPG) approach together with a novel action refinement scheme. Through extensive numerical simulations, we show that the proposed scheme greatly outperforms the existing methods in terms of both energy efficiency and computational efficiency. We also show that the proposed scheme can scale to large system with polynomial complexities in both storage and computation.

With the emergence of new applications, e.g., video gaming, file sharing and proximity aware social networks, high data-rate traffics will be likely to occur between devices in the vicinity of each other. Enabling Device to Device (D2D) communication between such devices can efficiently offload these traffics from the base station. Meanwhile, by reusing the cellular spectrum, D2D has the potential to significantly improve the spectral efficiency. However, if not controlled carefully, the interference caused by D2D communication may severely degrade the performance of regular cellular services. To resolve this issue, this paper proposes a simple guard-zone based mode selection scheme as an efficient and scalable mechanism to manage the interference in D2D underlaid cellular systems. With the help of stochastic geometry, we rigorously analyze the interference and throughput of such system. The results indicate the existence of an optimal guard zone radius that maximizes the total system throughput. Thanks to the tractability of our analysis, the optimal guard zone radius can be efficiently obtained. Through extensive numerical analysis, we show that the success probability and aggregate throughput obtained at the optimal guard zone radius are significantly greater than those without guard zone.