Scam detection remains a critical challenge in cybersecurity as adversaries craft messages that evade automated filters. We propose a Hierarchical Scam Detection System (HSDS) that combines a lightweight multi-model voting front end with a fine-tuned LLaMA 3.1 8B Instruct back end to improve accuracy and robustness against adversarial attacks. An ensemble of four classifiers provides preliminary predictions through majority vote, and ambiguous cases are escalated to the fine-tuned model, which is optimized with adversarial training to reduce misclassification. Experiments show that this hierarchical design both improves adversarial scam detection and shortens inference time by routing most cases away from the LLM, outperforming traditional machine-learning baselines and proprietary LLM baselines. The findings highlight the effectiveness of a hybrid voting mechanism and adversarial fine-tuning in fortifying LLMs against evolving scam tactics, enhancing the resilience of automated scam detection systems.

Personalized image preference assessment aims to evaluate an individual user's image preferences by relying only on a small set of reference images as prior information. Existing methods mainly focus on general preference assessment, training models with large-scale data to tackle well-defined tasks such as text-image alignment. However, these approaches struggle to handle personalized preference because user-specific data are scarce and not easily scalable, and individual tastes are often diverse and complex. To overcome these challenges, we introduce a common preference profile that serves as a bridge across users, allowing large-scale user data to be leveraged for training profile prediction and capturing complex personalized preferences. Building on this idea, we propose a reasoning-based personalized image preference assessment framework that follows a \textit{predict-then-assess} paradigm: it first predicts a user's preference profile from reference images, and then provides interpretable, multi-dimensional scores and assessments of candidate images based on the predicted profile. To support this, we first construct a large-scale Chain-of-Thought (CoT)-style personalized assessment dataset annotated with diverse user preference profiles and high-quality CoT-style reasoning, enabling explicit supervision of structured reasoning. Next, we adopt a two-stage training strategy: a cold-start supervised fine-tuning phase to empower the model with structured reasoning capabilities, followed by reinforcement learning to incentivize the model to explore more reasonable assessment paths and enhance generalization. Furthermore, we propose a similarity-aware prediction reward to encourage better prediction of the user's preference profile, which facilitates more reasonable assessments exploration. Extensive experiments demonstrate the superiority of the proposed method.

Selecting an appropriate reasoning method for a given query remains a key challenge in language model generation. Existing approaches typically generate multiple candidate responses and use an aggregation strategy to select the output answer, often assuming that more candidate answers yield higher accuracy. We revisit this assumption through a rigorous theoretical analysis, deriving accuracy bounds for standard aggregation methods under fixed generation distributions and candidate sizes. Building on these insights, we introduce EPIC, an Ensemble Planning with Contrastive learning framework to learn a shared representation space that captures both model reasoning abilities and query-method compatibility. EPIC incorporates our probability bounds as a regularizer in a utility-driven optimization that balances accuracy and computational cost. Experiments on diverse mathematical reasoning tasks show that EPIC consistently selects optimal reasoning methods, improving accuracy while reducing computational overhead. Our code can be found at https://github.com/nguyenngocbaocmt02/EPIC.

Human smuggling networks are increasingly adaptive and difficult to analyze. Legal case documents offer critical insights but are often unstructured, lexically dense, and filled with ambiguous or shifting references, which pose significant challenges for automated knowledge graph (KG) construction. While recent LLM-based approaches improve over static templates, they still generate noisy, fragmented graphs with duplicate nodes due to the absence of guided extraction and coreference resolution. The recently proposed CORE-KG framework addresses these limitations by integrating a type-aware coreference module and domain-guided structured prompts, significantly reducing node duplication and legal noise. In this work, we present a systematic ablation study of CORE-KG to quantify the individual contributions of its two key components. Our results show that removing coreference resolution results in a 28.25% increase in node duplication and a 4.32% increase in noisy nodes, while removing structured prompts leads to a 4.29% increase in node duplication and a 73.33% increase in noisy nodes. These findings offer empirical insights for designing robust LLM-based pipelines for extracting structured representations from complex legal texts.

Large language models (LLMs) have transformed natural language processing (NLP), enabling applications from content generation to decision support. Retrieval-Augmented Generation (RAG) improves LLMs by incorporating external knowledge but also introduces security risks, particularly from data poisoning, where the attacker injects poisoned texts into the knowledge database to manipulate system outputs. While various defenses have been proposed, they often struggle against advanced attacks. To address this, we introduce RAGuard, a detection framework designed to identify poisoned texts. RAGuard first expands the retrieval scope to increase the proportion of clean texts, reducing the likelihood of retrieving poisoned content. It then applies chunk-wise perplexity filtering to detect abnormal variations and text similarity filtering to flag highly similar texts. This non-parametric approach enhances RAG security, and experiments on large-scale datasets demonstrate its effectiveness in detecting and mitigating poisoning attacks, including strong adaptive attacks.

Despite significant work on low-bit quantization-aware training (QAT), there is still an accuracy gap between such techniques and native training. To address this, we introduce CAGE (Curvature-Aware Gradient Estimation), a new QAT method that augments the straight-through estimator (STE) gradient with a curvature-aware correction designed to counteract the loss increase induced by quantization. CAGE is derived from a multi-objective view of QAT that balances loss minimization with the quantization constraints, yielding a principled correction term that depends on local curvature information. On the theoretical side, we introduce the notion of Pareto-optimal solutions for quantized optimization, and establish that CAGE yields strong convergence guarantees in the smooth non-convex setting. In terms of implementation, our approach is optimizer-agnostic, but we provide a highly-efficient implementation that leverages Adam statistics. CAGE significantly improves upon the prior state-of-the-art methods in terms of accuracy, for similar computational cost: for QAT fine-tuning, it halves the compression accuracy loss relative to the prior best method, while for QAT pre-training of Llama models, its accuracy for 3-bit weights-and-activations (W3A3) matches the accuracy achieved at 4-bits (W4A4) with the prior best method. The official implementation can be found over https://github.com/IST-DASLab/CAGE .

Diffusion language models (DLMs) have emerged as a promising alternative to the long-dominant autoregressive (AR) paradigm, offering a parallelable decoding process that could yield greater efficiency. Yet, in practice, current open-source DLMs often underperform their AR counterparts in speed, limiting their real-world utility. This work presents a systematic study of DLM efficiency, identifying key issues in prior evaluation methods. Through empirical benchmarking and a theoretical analysis, we demonstrate that AR models generally achieve higher throughput, while DLMs consistently lag. We also investigate acceleration strategies, finding that techniques like dual cache and parallel decoding mainly offer gains at small batch sizes, with their benefits diminishing upon scaling. Our findings underscore the necessity of robust evaluation methods and improved acceleration strategies to advance research on DLMs.

Open-vocabulary object detection (OVD) models offer remarkable flexibility by detecting objects from arbitrary text queries. However, their zero-shot performance in specialized domains like Remote Sensing (RS) is often compromised by the inherent ambiguity of natural language, limiting critical downstream applications. For instance, an OVD model may struggle to distinguish between fine-grained classes such as "fishing boat" and "yacht" since their embeddings are similar and often inseparable. This can hamper specific user goals, such as monitoring illegal fishing, by producing irrelevant detections. To address this, we propose a cascaded approach that couples the broad generalization of a large pre-trained OVD model with a lightweight few-shot classifier. Our method first employs the zero-shot model to generate high-recall object proposals. These proposals are then refined for high precision by a compact classifier trained in real-time on only a handful of user-annotated examples - drastically reducing the high costs of RS imagery annotation.The core of our framework is FLAME, a one-step active learning strategy that selects the most informative samples for training. FLAME identifies, on the fly, uncertain marginal candidates near the decision boundary using density estimation, followed by clustering to ensure sample diversity. This efficient sampling technique achieves high accuracy without costly full-model fine-tuning and enables instant adaptation, within less then a minute, which is significantly faster than state-of-the-art alternatives.Our method consistently surpasses state-of-the-art performance on RS benchmarks, establishing a practical and resource-efficient framework for adapting foundation models to specific user needs.

Recent LLM benchmarks have tested models on a range of phenomena, but are still focused primarily on natural language understanding for extraction of explicit information, such as QA or summarization, with responses often targeting information from individual sentences. We are still lacking more challenging, and importantly also multilingual, benchmarks focusing on implicit information and pragmatic inferences across larger documents in the context of discourse tracking: integrating and aggregating information across sentences, paragraphs and multiple speaker utterances. To this end, we present DiscoTrack, an LLM benchmark targeting a range of tasks across 12 languages and four levels of discourse understanding: salience recognition, entity tracking, discourse relations and bridging inference. Our evaluation shows that these tasks remain challenging, even for state-of-the-art models.

Speculative decoding has become a standard way to accelerate LLM inference: a small drafter proposes multiple tokens and a large target model verifies them once per speculation length. Recently, scaling of the LLM vocabulary has pushed the number of tokens to grow substantially. While verification over the full vocabulary leaves the target model largely unaffected, the O(|V|d) parameters in the drafter's output head become a latency bottleneck, slowing the entire pipeline. Contemporary methods (e.g., FR-Spec, VocabTrim) restrict the drafter's vocabulary to a fixed top frequent subset of the target model's vocabulary. Although this reduces draft-time compute, it is brittle, since: (i) frequency lists are corpus-dependent and require retuning to generalize, and (ii) static shortlists suppress rare or domain-specific tokens, lowering the expected number of tokens per verification step. We propose DynaSpec, a context-dependent dynamic shortlisting mechanism that is robust, speeds up drafting, and generalizes across diverse tasks. Concretely, we introduce lightweight, coarse-grained meta-classifiers that route contexts to a small number of token clusters; the union of the top-k selected clusters forms the drafter's shortlist, while verification retains the full vocabulary and exactness. The meta-classifier finishes its computation earlier than the drafter's hidden state generation by exploiting parallel execution of draft encoding and meta shortlisting on separate streams. Across standard speculative decoding benchmarks, DynaSpec delivers consistent improvements in mean accepted length, for Llama-3-8B, reaching upto 98.2% of full-vocabulary performance, while fixed-shortlist baselines attain only 84.4%. By leveraging context-dependent selection, DynaSpec achieves up to a 2.18 times increase in generated tokens compared to 1.91 times for fixed-vocabulary approaches.