Training-free KV cache compression is essential for deploying vision-language GUI agents under memory and latency constraints, yet existing methods are designed for generic language workloads and ignore the distinctive structure of GUI interaction traces. We characterize three GUI-specific workload properties — high inter-frame visual redundancy, extremely small UI-element spatial footprints, and near-uniform cross-layer attention sparsity — that cause existing schemes to retain as few as 39% of oracle-important KV pairs at the standard 20% budget. To address this, we propose **ST-Lite**, a training-free compression scheme whose three components each target one property: Trajectory-aware Semantic Gating (TSG) filters redundant historical frames, Component-centric Spatial Saliency (CSS) preserves fine-grained element boundaries, and a flat per-layer budget avoids hierarchical misallocation. Across seven GUI benchmarks and two backbones in the deployment-relevant β ∈ [10%, 40%] window, **ST-Lite** consistently outperforms all existing compression baselines — matching or exceeding Full Cache task accuracy on the primary backbone at the 20% budget — while delivering up to 2.35× decoding speedup at fivefold compression. Code: https://github.com/94wen94/ST-Lite

Knowledge-based visual question answering (KB-VQA) requires vision-language models to understand images and use external knowledge, especially for rare entities and long-tail facts. Most existing retrieval-augmented generation (RAG) methods adopt a fixed pipeline that sequentially retrieves information, filters it, and then produces an answer. Such a design makes it difficult to adapt to diverse question types. Moreover, it separates retrieval from reasoning, making it hard for the model to decide when to search, how to refine queries, or when to stop. As a result, the retrieved evidence is often poorly aligned with the question. To address these limitations, we reformulate KB-VQA as a search-agent problem and model the solving process as a multi-step decision-making procedure. At each step, the agent selects one of four actions-Answer, Image Retrieval, Text Retrieval, and Caption-based on its current information state. We further design an automated pipeline to collect multi-step trajectories that record the agent’s reasoning process, tool usage, and intermediate decisions. These trajectories are then used as supervision for fine-tuning. Experiments on InfoSeek and E-VQA demonstrate that our method achieves state-of-the-art performance, consistently outperforming prior baselines and confirming the effectiveness of our framework.

Large vision-language models (VLMs) are increasingly deployed in safety-critical settings, yet existing visual jailbreak research has focused almost exclusively on autoregressive architectures, leaving an important emerging family unstudied: multimodal discrete diffusion vision-language models (dVLMs). We identify a vulnerability that is specific to the diffusion generation process: because the visual embedding conditions every reverse denoising step rather than acting as a one-time prefix, adversarial visual semantics are not injected once but repeatedly propagated and amplified across the generation trajectory, a phenomenon we term cross-step conditional propagation. We provide empirical evidence that this mechanism is absent in autoregressive VLMs through stage-sensitivity analysis, prompt-level switch rates, and pairwise denoising-bin disagreement metrics, all confirmed statistically by bootstrap resampling. Motivated by this finding, we propose DIVA (Discrete-diffusion Vision-language model Attack), a white-box visual jailbreak framework that exploits dVLM-specific vulnerabilities via cross-modal intent obfuscation and diffusion-aware multi-timestep adversarial optimization. Across three dVLMs (LLaDA-V, MMaDA, and LLaDA2.0-Uni), DIVA reaches 58.8%, 67.7%, and 69.1% HADES ASR under the Beaver reward-model metric, substantially outperforming visual jailbreak baselines designed for autoregressive models. Our code are available at https://github.com/loststars2002/DIVA.

In the unsupervised self-evolution of Multimodal Large Language Models, the quality of feedback signals during post-training is pivotal for stable and effective learning. However, existing self-evolution methods predominantly rely on majority voting to select the most frequent output as the pseudo-golden answer, which may stem from the model’s intrinsic biases rather than guaranteeing the objective correctness of the reasoning paths. To counteract the degradation, we propose Continuous Softened Retracing reSampling (CSRS) in MLLM self-evolution. Specifically, we introduce a Retracing Re-inference Mechanism (RRM) that the model re-inferences from anchor points to expand the exploration of long-tail reasoning paths. Simultaneously, we propose Softened Frequency Reward (SFR), which replaces binary rewards with continuous signals, calibrating reward based on the answers’ frequency across sampled reasoning sets. Furthermore, incorporated with Visual Semantic Perturbation (VSP), CSRS ensures the model prioritizes mathematical logic over visual superficiality. Experimental results demonstrate that CSRS significantly enhances the reasoning performance of Qwen2.5-VL-7B on benchmarks such as MathVision. We achieve state-of-the-art (SOTA) results in unsupervised self-evolution on geometric tasks.

Large Language Models (LLMs) serving as long-term personal companions require memory systems that support complex reasoning over extended dialogues. However, existing Retrieval-Augmented Generation (RAG) frameworks treat memory as passive storage, suffering from semantic collapse—where subjective beliefs and objective facts become indistinguishable—and rigid navigation that fails to connect dispersed information. We introduce CogMem, a cognitive memory architecture that reconstructs dialogue context through explicit structure. CogMem organizes memory via the PEC²F (Person-Event-Concept-Claim-Fact) schema, using dedicated Claim nodes to separate subjective attributions from objective records and prevent information contamination. Rather than static retrieval, CogMem employs a deterministic controller with four cognitive operators (anchoring, traversal, intersection, evidence grounding) that dynamically navigate the graph based on query intent. Comprehensive evaluation on LoCoMo and LongMemEval shows that CogMem achieves state-of-the-art performance across all task categories, surpassing neural memory systems and graph retrieval methods. These results demonstrate that structured cognitive modeling significantly improves long-term memory accuracy and robustness compared to passive vector retrieval.