Recent progress in vision-language models (VLMs) has led to impressive results in document understanding tasks, but their high computational demands remain a challenge. To mitigate the compute burdens, we propose a lightweight token pruning framework that filters out non-informative background regions from document images prior to VLM processing. A binary patch-level classifier removes non-text areas, and a max-pooling refinement step recovers fragmented text regions to enhance spatial coherence. Experiments on real-world document datasets demonstrate that our approach substantially lowers computational costs, while maintaining comparable accuracy.

This study presents DiagCoT, a multi-stage framework that applies supervised fine-tuning to general-purpose vision-language models (VLMs) to emulate radiologists' stepwise diagnostic reasoning using only free-text reports. DiagCoT combines contrastive image-report tuning for domain alignment, chain-of-thought supervision to capture inferential logic, and reinforcement tuning with clinical reward signals to enhance factual accuracy and fluency. On the MIMIC-CXR benchmark, DiagCoT improved zero-shot disease classification AUC from 0.52 to 0.76 (absolute gain of 0.24), pathology grounding mIoU from 0.08 to 0.31 (absolute gain of 0.23), and report generation BLEU from 0.11 to 0.33 (absolute gain of 0.22). It outperformed state-of-the-art models including LLaVA-Med and CXR-LLAVA on long-tailed diseases and external datasets. By converting unstructured clinical narratives into structured supervision, DiagCoT offers a scalable approach for developing interpretable and diagnostically competent AI systems for radiology.

Meeting minimum data rate constraints is a significant challenge in wireless communication systems, particularly as network complexity grows. Traditional deep learning approaches often address these constraints by incorporating penalty terms into the loss function and tuning hyperparameters empirically. However, this heuristic treatment offers no theoretical convergence guarantees and frequently fails to satisfy QoS requirements in practical scenarios. Building upon the structure of the WMMSE algorithm, we first extend it to a multi-channel setting with QoS constraints, resulting in the enhanced WMMSE (eWMMSE) algorithm, which is provably convergent to a locally optimal solution when the problem is feasible. To further reduce computational complexity and improve scalability, we develop a GNN-based algorithm, JCPGNN-M, capable of supporting simultaneous multi-channel allocation per user. To overcome the limitations of traditional deep learning methods, we propose a principled framework that integrates GNN with a Lagrangian-based primal-dual optimization method. By training the GNN within the Lagrangian framework, we ensure satisfaction of QoS constraints and convergence to a stationary point. Extensive simulations demonstrate that JCPGNN-M matches the performance of eWMMSE while offering significant gains in inference speed, generalization to larger networks, and robustness under imperfect channel state information. This work presents a scalable and theoretically grounded solution for constrained resource allocation in future wireless networks.

Due to hardware and software improvements, an increasing number of AI models are deployed on-device. This shift enhances privacy and reduces latency, but also introduces security risks distinct from traditional software. In this article, we examine these risks through the real-world case study of SafetyCore, an Android system service incorporating sensitive image content detection. We demonstrate how the on-device AI model can be extracted and manipulated to bypass detection, effectively rendering the protection ineffective. Our analysis exposes vulnerabilities of on-device AI models and provides a practical demonstration of how adversaries can exploit them.

Drought stress is a major threat to global crop productivity, making its early and precise detection essential for sustainable agricultural management. Traditional approaches, though useful, are often time-consuming and labor-intensive, which has motivated the adoption of deep learning methods. In recent years, Convolutional Neural Network (CNN) and Vision Transformer architectures have been widely explored for drought stress identification; however, these models generally rely on a large number of trainable parameters, restricting their use in resource-limited and real-time agricultural settings. To address this challenge, we propose a novel lightweight hybrid CNN framework inspired by ResNet, DenseNet, and MobileNet architectures. The framework achieves a remarkable 15-fold reduction in trainable parameters compared to conventional CNN and Vision Transformer models, while maintaining competitive accuracy. In addition, we introduce a machine unlearning mechanism based on a gradient norm-based influence function, which enables targeted removal of specific training data influence, thereby improving model adaptability. The method was evaluated on an aerial image dataset of potato fields with expert-annotated healthy and drought-stressed regions. Experimental results show that our framework achieves high accuracy while substantially lowering computational costs. These findings highlight its potential as a practical, scalable, and adaptive solution for drought stress monitoring in precision agriculture, particularly under resource-constrained conditions.

Colorectal diseases, including inflammatory conditions and neoplasms, require quick, accurate care to be effectively treated. Traditional diagnostic pipelines require extensive preparation and rely on separate, individual evaluations on histological images and colonoscopy footage, introducing possible variability and inefficiencies. This pilot study proposes a unified deep learning network that uses convolutional neural networks (CN N s) to classify both histopathological slides and colonoscopy video frames in one pipeline. The pipeline integrates class-balancing learning, robust augmentation, and calibration methods to ensure accurate results. Static colon histology images were taken from the PathMNIST dataset, and the lower gastrointestinal (colonoscopy) videos were drawn from the HyperKvasir dataset. The CNN architecture used was ResNet-50. This study demonstrates an interpretable and reproducible diagnostic pipeline that unifies multiple diagnostic modalities to advance and ease the detection of colorectal diseases.

The widespread distribution of Large Language Models (LLMs) through public platforms like Hugging Face introduces significant security challenges. While these platforms perform basic security scans, they often fail to detect subtle manipulations within the embedding layer. This work identifies a novel class of deployment phase attacks that exploit this vulnerability by injecting imperceptible perturbations directly into the embedding layer outputs without modifying model weights or input text. These perturbations, though statistically benign, systematically bypass safety alignment mechanisms and induce harmful behaviors during inference. We propose Search based Embedding Poisoning(SEP), a practical, model agnostic framework that introduces carefully optimized perturbations into embeddings associated with high risk tokens. SEP leverages a predictable linear transition in model responses, from refusal to harmful output to semantic deviation to identify a narrow perturbation window that evades alignment safeguards. Evaluated across six aligned LLMs, SEP achieves an average attack success rate of 96.43% while preserving benign task performance and evading conventional detection mechanisms. Our findings reveal a critical oversight in deployment security and emphasize the urgent need for embedding level integrity checks in future LLM defense strategies.

Questionnaire-based surveys are foundational to social science research and public policymaking, yet traditional survey methods remain costly, time-consuming, and often limited in scale. This paper explores a new paradigm: simulating virtual survey respondents using Large Language Models (LLMs). We introduce two novel simulation settings, namely Partial Attribute Simulation (PAS) and Full Attribute Simulation (FAS), to systematically evaluate the ability of LLMs to generate accurate and demographically coherent responses. In PAS, the model predicts missing attributes based on partial respondent profiles, whereas FAS involves generating complete synthetic datasets under both zero-context and context-enhanced conditions. We curate a comprehensive benchmark suite, LLM-S^3 (Large Language Model-based Sociodemographic Survey Simulation), that spans 11 real-world public datasets across four sociological domains. Our evaluation of multiple mainstream LLMs (GPT-3.5/4 Turbo, LLaMA 3.0/3.1-8B) reveals consistent trends in prediction performance, highlights failure modes, and demonstrates how context and prompt design impact simulation fidelity. This work establishes a rigorous foundation for LLM-driven survey simulations, offering scalable and cost-effective tools for sociological research and policy evaluation. Our code and dataset are available at: https://github.com/dart-lab-research/LLM-S-Cube-Benchmark

As on-device LLMs(e.g., Apple on-device Intelligence) are widely adopted to reduce network dependency, improve privacy, and enhance responsiveness, verifying the legitimacy of models running on local devices becomes critical. Existing attestation techniques are not suitable for billion-parameter Large Language Models (LLMs), struggling to remain both time- and memory-efficient while addressing emerging threats in the LLM era. In this paper, we present AttestLLM, the first-of-its-kind attestation framework to protect the hardware-level intellectual property (IP) of device vendors by ensuring that only authorized LLMs can execute on target platforms. AttestLLM leverages an algorithm/software/hardware co-design approach to embed robust watermarking signatures onto the activation distributions of LLM building blocks. It also optimizes the attestation protocol within the Trusted Execution Environment (TEE), providing efficient verification without compromising inference throughput. Extensive proof-of-concept evaluations on LLMs from Llama, Qwen, and Phi families for on-device use cases demonstrate AttestLLM's attestation reliability, fidelity, and efficiency. Furthermore, AttestLLM enforces model legitimacy and exhibits resilience against model replacement and forgery attacks.

The rapid development of the low-altitude economy emphasizes the critical need for effective perception and intent recognition of non-cooperative unmanned aerial vehicles (UAVs). The advanced generative reasoning capabilities of multimodal large language models (MLLMs) present a promising approach in such tasks. In this paper, we focus on the combination of UAV intent recognition and the MLLMs. Specifically, we first present an MLLM-enabled UAV intent recognition architecture, where the multimodal perception system is utilized to obtain real-time payload and motion information of UAVs, generating structured input information, and MLLM outputs intent recognition results by incorporating environmental information, prior knowledge, and tactical preferences. Subsequently, we review the related work and demonstrate their progress within the proposed architecture. Then, a use case for low-altitude confrontation is conducted to demonstrate the feasibility of our architecture and offer valuable insights for practical system design. Finally, the future challenges are discussed, followed by corresponding strategic recommendations for further applications.