Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs remains poorly understood. We present a sensitivity analysis that probes how LLMs extract harmful sentences embedded in long inputs. We construct long inputs by combining neutral and harmful sentences, and systematically vary four factors: input length (600--30,000 tokens), the proportion of harmful sentences (0.01--0.50), harm realization (explicit vs. implicit), and the position of harmful sentences within the input (beginning, middle, end), enabling a controlled stress-test evaluation. Experiments across toxic, offensive, and hate content, and across LLaMA-3.1, Qwen-2.5, and Mistral, reveal consistent patterns: sensitivity is non-monotonic with respect to harmful prevalence, peaking at moderate levels; sensitivity degrades as input length increases; harmful sentences placed earlier in the input are more strongly prioritized; and explicit harm is more reliably identified than implicit harm. These findings provide a systematic view of how LLMs prioritize harmful sentences in long input under controlled stress conditions, highlighting both emerging strengths and remaining challenges for safety-related use.

When we read, we make predictions about upcoming words; these predictions influence our reading behavior. The success of large language models (LLMs), which, like humans, make predictions about upcoming words, has motivated their use as models of human linguistic prediction. Surprisingly, in the last few years, as LLMs' ability to predict the next word has improved, their ability to explain reading behavior has declined. We argue this is because current LLMs can predict upcoming words much better than human readers can. This 'superhumanness' is driven by LLMs' extensive training data, stronger long-term memory of training examples, and stronger short-term memory. We advocate for LLMs with human-like memory and for new experiments to measure the alignment between humans and LLMs, and outline directions towards achieving these goals.

Diffusion models achieve state-of-the-art image generation but often produce semantic inconsistencies, or hallucinations. Existing inference-time guidance methods rely on external signals or architectural modifications, adding computational overhead. We propose $\mathbf{T}$angential $\mathbf{A}$mplifying $\mathbf{G}$uidance $\mathbf{(TAG)}$, a training-free, architecture-agnostic, plug-and-play guidance method that operates purely on trajectory signals. TAG uses an intermediate sample as a projection basis and amplifies the tangential components of the estimated score to correct the sampling trajectory. A first-order Taylor analysis shows that this steers the state toward higher-probability regions of the data manifold, reducing inconsistencies and improving fidelity while adding negligible overhead to existing samplers. Code is available at our Project Page (https://hyeon-cho.github.io/TAG/).

Large language models (LLMs) demonstrate strong reasoning abilities via Chain-of-Thought (CoT), but their token-level generation encourages local decisions and lacks global planning, often leading to redundant or inaccurate reasoning. Existing methods, such as tree-based search and reinforcement learning (RL), attempt to address this issue but incur high computational costs and still struggle to produce reliable reasoning trajectories. To address these challenges, we propose Plan-Then-Action Enhanced Reasoning with Group Relative Policy Optimization (PTA-GRPO), a two-stage framework designed to jointly improve high-level planning and fine-grained CoT reasoning. Specifically, in the first stage, a given LLM is responsible for summarizing CoT reasoning into compact high-level guidance, which is then leveraged for supervised fine-tuning. Then, we introduce a guidance-aware reinforcement learning method that jointly optimizes the final output and the quality of guidance, enhancing reasoning effectiveness. We evaluate PTA-GRPO on ten reasoning benchmarks across mathematics and natural sciences, using five diverse base models spanning multiple data modalities. The results show that PTA-GRPO consistently delivers significant improvements across models and tasks, demonstrating strong effectiveness and generalization.

Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\textit{``Intermediate"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate the intermediate verification step, and compromise accuracy by relying on approximate heuristics. We propose $\underline{\textit{Hi}}\textit{erarchical }\underline{\textit{Spec}}\textit{ulative Decoding (HiSpec)}$, a framework for high-throughput speculative decoding that exploits $\textit{early-exit (EE) models}$ for low-overhead intermediate verification. EE models allow tokens to exit early by skipping layer traversal and are explicitly trained so that hidden states at selected layers can be interpreted, making them uniquely suited for intermediate verification without drastically increasing compute and memory overheads. To improve resource-efficiency even further, we design a methodology that enables HiSpec to re-use key-value caches and hidden states between the draft, intermediate verifier, and target models. To maintain accuracy, HiSpec periodically validates the draft tokens accepted by the intermediate verifier against the target model. Our evaluations using various representative benchmarks and models show that HiSpec improves throughput by 1.28$\times$ on average and by up to 2.01$\times$ compared to the baseline single-layer speculation without compromising accuracy.

Transfer learning is crucial for medical imaging, yet the selection of source datasets often relies on researchers' intuition rather than systematic principles, which can impact the generalizability of algorithms and, thus, patient outcomes. This study investigates these decisions through a task-based survey with machine learning practitioners. Unlike prior work that benchmarks models and experimental setups, we take a human-computer interaction (HCI) perspective on how practitioners select source datasets. Our findings indicate that choices are task-dependent and influenced by community practices, dataset properties, and computational (data embedding), or perceived visual or semantic similarity. However, similarity ratings and expected performance are not always aligned, challenging a traditional "more similar is better" view. Moreover, ethical and fairness considerations remain largely absent from source dataset sections. Participants often used ambiguous terminology, which suggests a need for clearer definitions and tools to make them explicit and usable. By clarifying these heuristics and introducing a conceptual framework of transfer learning factors, this work provides practical insights for more systematic source selection in transfer learning.

As LLMs rapidly saturate existing benchmarks, automated benchmark creation using LLMs (LLM-as-a-benchmark) -- where a model generates test inputs (LLM-as-a-testset) and evaluates outputs (LLM-as-an-evaluator) -- has gained traction as a cheap alternative to human curation. We show that this paradigm has a fundamental problem: LLM-generated benchmarks systematically favor the model that created them. Using machine translation as our primary testbed, we find that self-bias arises from two additive sources, LLM-as-a-testset and LLM-as-an-evaluator, and their combination amplifies the effect. Crucially, even when test data is generated with explicit diversity controls, each model's implicit stylistic tendencies produce homogeneous, model-specific outputs that inflate its own scores. Increasing source text diversity, using our proposed diversity metric, partially mitigates this bias. Self-bias is strong enough to cause each model to rank itself first, overriding the peer-consensus ordering. We confirm that the phenomenon extends to open-ended generation on the Chatbot Arena task.

Graphical User Interface (GUI) grounding is commonly framed as a coordinate prediction task -- given a natural language instruction, generate on-screen coordinates for actions such as clicks and keystrokes. However, recent Vision Language Models (VLMs) often fail to predict accurate numeric coordinates when processing GUI images with high resolutions and complex layouts. To address this issue, we reframe GUI grounding as an interactive search task, where the VLM generates actions to move a cursor in the GUI to locate UI elements. At each step, the model determines the target object, evaluates the spatial relations between the cursor and the target, and moves the cursor closer to the target conditioned on the movement history. In this interactive process, the rendered cursor provides visual feedback to help the model align its predictions with the corresponding on-screen locations. We train our GUI grounding model, GUI-Cursor, using multi-step online reinforcement learning with a dense trajectory-based reward function. Experimental results demonstrate that GUI-Cursor surpasses strong baselines in GUI grounding and agentic tasks, achieving superior performance with the same base models while requiring less training data. Further analysis shows that GUI-Cursor learns to adaptively conduct more steps on more difficult examples, and it obtains better spatial reasoning capability on out-of-distribution domains.

High-level representations have become a central focus in enhancing AI transparency and control, shifting attention from individual neurons or circuits to structured semantic directions that align with human-interpretable concepts. While the Linear Representation Hypothesis (LRH) suggests that such directions emerge in representations, it remains unclear how these representations originate and why they become increasingly stable across layers. To solve this issue, we introduce the Input-Space Linearity Hypothesis, positing that concept-aligned directions originate in the input space and are steadily maintained with increasing depth. We then propose the Spectral Principal Path (SPP) framework, which formalizes how deep networks progressively distill linear representations along the spectral principal directions. We provide rigorous stability guarantees for the SPP based on the Wedin $\sinΘ$ perturbation theorem, identifying testable conditions, including spectral gap and context incoherence, that jointly ensure layer-wise directional preservation. By bridging theoretical analysis with empirical evidence, this work identifies a spectral view of how linear representations arise in LLMs, and suggests potential implications for concept-level controllable, robust, and coherent approaches to fairness and transparency in modern AI systems.

Most existing methods for concept unlearning in text-to-image diffusion models minimize a mean squared error (MSE) loss between the denoiser outputs conditioned on a target and an anchor concept, which is implicitly the KL divergence between two Gaussians. We generalize this objective to any $f$-divergence, recovering MSE as the KL instance, and identify a family of $α$-divergences whose Gaussian closed-form yields cheap, MSE-like training objectives. For the remaining $f$-divergences, we provide a min-max objective based on the variational formulation of the $f$-divergence. We theoretically analyze and numerically validate how different $f$-divergences impact the gradient magnitude and the convergence properties of the algorithm, affecting the quality of unlearning. For instance, we observe that the Hellinger closed-form instance consistently dominates MSE across multiple scenarios. More generally, the proposed unified framework offers a flexible paradigm for selecting the optimal divergence based on the application and user goal, allowing for finer control over the trade-off between unlearning efficacy and generative fidelity.

Search-augmented large language models (LLMs) have advanced information-seeking tasks by integrating retrieval into generation, reducing users' cognitive burden compared to traditional search systems. Yet they remain insufficient for fully addressing diverse user needs, which requires recognizing how the same query can reflect different intents across users and delivering information in preferred forms. While recent systems such as ChatGPT and Gemini attempt personalization by leveraging user histories, systematic evaluation of such personalization is under-explored. To address this gap, we propose BESPOKE, the realistic benchmark for evaluating personalization in search-augmented LLMs. BESPOKE is designed to be both realistic, by collecting authentic chat and search histories directly from humans, and diagnostic, by pairing responses with fine-grained preference scores and feedback. The benchmark is constructed through long-term, deeply engaged human annotation, where human annotators contributed their own histories, authored queries with detailed information needs, and evaluated responses with scores and diagnostic feedback. Leveraging BESPOKE, we conduct systematic analyses that reveal key requirements for effective personalization in information-seeking tasks, providing a foundation for fine-grained evaluation of personalized search-augmented LLMs. Our code and data are available at https://augustinlib.github.io/BESPOKE/.

The pretraining-finetuning paradigm is a crucial strategy in metallic surface defect detection for mitigating the challenges posed by data scarcity. However, its implementation presents a critical dilemma. Pretraining on natural image datasets such as ImageNet, faces a significant domain gap. Meanwhile, naive self-supervised pretraining on in-domain industrial data is often ineffective due to the inability of existing learning objectives to distinguish subtle defect patterns from complex background noise and textures. To resolve this, we introduce Anomaly-Guided Self-Supervised Pretraining (AGSSP), a novel paradigm that explicitly guides representation learning through anomaly priors. AGSSP employs a two-stage framework: (1) it first pretrains the model's backbone by distilling knowledge from anomaly maps, encouraging the network to capture defect-salient features; (2) it then pretrains the detector using pseudo-defect boxes derived from these maps, aligning it with localization tasks. To enable this, we develop a knowledge-enhanced method to generate high-quality anomaly maps and collect a large-scale industrial dataset of 120,000 images. Additionally, we present two small-scale, pixel-level labeled metallic surface defect datasets for validation. Extensive experiments demonstrate that AGSSP consistently enhances performance across various settings, achieving up to a 10\% improvement in mAP@0.5 and 11.4\% in mAP@0.5:0.95 compared to ImageNet-based models. All code, pretrained models, and datasets are publicly available at https://clovermini.github.io/AGSSP-Dev/.

Large language models (LLMs) can translate natural language instructions into executable action plans for robotics, autonomous driving, and other domains. Yet, deploying LLM-driven planning in the physical world demands strict adherence to safety and regulatory constraints, which current models often violate due to hallucination or weak alignment. Traditional data-driven alignment methods, such as Direct Preference Optimization (DPO), require costly human labeling, while recent formal-feedback approaches still depend on resource-intensive fine-tuning. In this paper, we propose LAD-VF, a fine-tuning-free framework that leverages formal verification feedback for automated prompt engineering. By introducing a formal-verification-informed text loss integrated with LLM-AutoDiff, LAD-VF iteratively refines prompts rather than model parameters. This yields three key benefits: (i) scalable adaptation without fine-tuning; (ii) compatibility with modular LLM architectures; and (iii) interpretable refinement via auditable prompts. Experiments in robot navigation and manipulation tasks demonstrate that LAD-VF substantially enhances specification compliance, improving success rates from 60% to over 90%. Our method thus presents a scalable and interpretable pathway toward trustworthy, formally-verified LLM-driven control systems.

With the improving semantic understanding capability of Large Language Models (LLMs), they exhibit a greater awareness and alignment with human values, but this comes at the cost of transparency. Although promising results are achieved via experimental analysis, an in-depth understanding of the LLM's internal workings is unavoidable to comprehend the reasoning behind the re-ranking, which provides end users with an explanation that enables them to make an informed decision. Moreover, in newly developed systems with limited user engagement and insufficient ranking data, accurately re-ranking content remains a significant challenge. While various training methods affect the training of LLMs and generate inference, our analysis has found that some training methods exhibit better explainability than others, implying that an accurate semantic understanding has not been learned through all training methods; instead, abstract knowledge has been gained to optimize evaluation, which raises questions about the true reliability of LLMs. Therefore, in this work, we analyze how different training methods affect the semantic understanding of the re-ranking task in LLMs and investigate whether these models can generate more informed textual reasoning to overcome the challenges of transparency or LLMs and limited training data. To analyze the LLMs for re-ranking tasks, we utilize a relatively small ranking dataset from the environment and the Earth science domain to re-rank retrieved content. Furthermore, we also analyze the explainable information to see if the re-ranking can be reasoned using explainability.

This work tackles the challenge of continuous sign language segmentation, a key task with huge implications for sign language translation and data annotation. We propose a transformer-based architecture that models the temporal dynamics of signing and frames segmentation as a sequence labeling problem using the Begin-In-Out (BIO) tagging scheme. Our method leverages the HaMeR hand features, and is complemented with 3D Angles. Extensive experiments show that our model achieves state-of-the-art results on the DGS Corpus, while our features surpass prior benchmarks on BSLCorpus.

Group Activity Detection (GAD) involves recognizing social groups and their collective behaviors in videos. Vision Foundation Models (VFMs), like DINOv2, offer excellent features but are pretrained on object-centric data. We find that naively substituting them into existing GAD pipelines actually degrades performance, exposing structured group-aware decoding as the true bottleneck. We introduce ProGraD, a structured relational-reasoning framework for GAD built on top of frozen VFMs. At its core is a lightweight two-layer GroupContext Transformer that explicitly models actor-group associations and aggregates global context to infer collective behavior. Learnable group prompts serve as a minimal conditioning mechanism to guide the frozen backbone toward socially relevant representations, while the relational decoder performs the core reasoning over actors and groups. This design jointly infers group locations, memberships, and activities in a single pass using only 10M trainable parameters - less than half of prior methods. On the Cafe benchmark with multiple concurrent social groups, ProGraD improves the state-of-the-art by 6.5% Group mAP$@$1.0 and 8.2% Group mAP$@$0.5. On Social-CAD, it achieves state-of-the-art social and membership accuracy. ProGraD further produces interpretable attention maps that provide insights into actor-group reasoning.

Large Language Models (LLMs) have achieved remarkable performance on a wide range of Natural Language Processing (NLP) benchmarks, often surpassing human-level accuracy. However, their reliability in high-stakes domains such as medicine, particularly in low-resource languages, remains underexplored. In this work, we introduce PersianMedQA, a large-scale dataset of 20,785 expert-validated multiple-choice Persian medical questions from 14 years of Iranian national medical exams, spanning 23 medical specialties and designed to evaluate LLMs in both Persian and English. We benchmark 41 state-of-the-art models, including general-purpose, Persian, and medical LLMs, in zero-shot and chain-of-thought (CoT) settings. Our results show that closed-weight general models (e.g., GPT-4.1) consistently outperform all other categories, achieving 83.09% accuracy in Persian and 80.7% in English, while Persian LLMs such as Dorna underperform significantly (e.g., 34.9% in Persian), often struggling with both instruction-following and domain reasoning. We also analyze the impact of translation, showing that while English performance is generally higher, 3-10% of questions can only be answered correctly in Persian due to cultural and clinical contextual cues that are lost in translation. Finally, we demonstrate that model size alone is insufficient for robust performance without strong domain or language adaptation. PersianMedQA provides a foundation for evaluating bilingual and culturally grounded medical reasoning in LLMs. The dataset, along with a bilingual medical dictionary, is available: https://huggingface.co/datasets/MohammadJRanjbar/PersianMedQA .

The recently proposed Large Concept Model (LCM) generates text by predicting a sequence of sentence-level embeddings and training with either mean-squared error or diffusion objectives. We present SONAR-LLM, a decoder-only transformer that "thinks" in the same continuous SONAR embedding space, yet is supervised through token-level cross-entropy propagated via the frozen SONAR decoder. This hybrid objective retains the semantic abstraction of LCM while eliminating its diffusion sampler and restoring a likelihood-based training signal. Across model sizes from 39M to 1.3B parameters, SONAR-LLM attains competitive generation quality. We report scaling trends, ablations, benchmark results, and release the complete training code and all pretrained checkpoints to foster reproducibility and future research.

Photorealistic talking-head avatars are becoming increasingly common in virtual meetings, gaming, and social platforms. These avatars allow for more immersive communication, but they also introduce serious security risks. One emerging threat is impersonation: an attacker can steal a user's avatar, preserving his appearance and voice, making it nearly impossible to detect its fraudulent usage by sight or sound alone. In this paper, we explore the challenge of biometric verification in such avatar-mediated scenarios. Our main question is whether an individual's facial motion patterns can serve as reliable behavioral biometrics to verify their identity when the avatar's visual appearance is a facsimile of its owner. To answer this question, we introduce a new dataset of realistic avatar videos created using a state-of-the-art one-shot avatar generation model, GAGAvatar, with genuine and impostor avatar videos. We also propose a lightweight, explainable spatio-temporal Graph Convolutional Network architecture with temporal attention pooling, that uses only facial landmarks to model dynamic facial gestures. Experimental results demonstrate that facial motion cues enable meaningful identity verification with AUC values approaching 80%. The proposed benchmark and biometric system are available for the research community in order to bring attention to the urgent need for more advanced behavioral biometric defenses in avatar-based communication systems.

Existing studies typically investigate domain shift and category shift as independent problems, however, in real-world scenarios, the two types of shifts often occur simultaneously and interact, leading to significant degradation in detection performance. To address this, we propose and systematically study a novel problem-Open-Domain Open-Vocabulary (ODOV) object detection-which aims to evaluate a model's ability to adapt to the compound domain and category shifts in real-world environments.We construct a new benchmark, OD-LVIS, which contains 46,949 images spanning 15 diverse real-world scenarios and 1,203 categories, for assessing object detection performance. Furthermore, we propose a novel ODOV detection baseline that fully leverages VLM's powerful multi-modal alignment capabilities and introduces two key mechanisms to enhance both category and domain generalization. One is the Domain-Agnostic Category Prompt (DAPmt), which strengthens category semantics while attenuating domain representations, enabling pure category representation. The other is the Domain Projection and Grafting (DP&G) module, which incorporates domain-specific features from input images, allowing the model to dynamically generalize across diverse open domains. These two components enable the model to maintain effective detection performance under simultaneous category and domain variations in real-world scenarios. We provide extensive benchmark evaluations for the proposed ODOV detection task and report experimental results. These results validate the soundness of the ODOV task, the practicality of the OD-LVIS dataset, and the superiority of the method.

Motivation: Structure-based drug design (SBDD) has advanced with deep generative models, but bridging the gap between continuous atomic coordinates and discrete atom types remains a challenge. Current approaches, such as diffusion and flow matching models, often fail to unify these heterogeneous modalities, relying on separate strategies or ill-fitting Euclidean metrics for discrete variables. This lack of a consistent framework limits generative models' ability to capture the geometric and chemical structure of protein-ligand complexes. Results: We present MolPIF, a parameter interpolation flow mechanism designed to unify the generation of continuous and discrete molecular variables. Unlike traditional flow models that operate in sample space, MolPIF interpolates between distributions in the parameter space, theoretically recovering Wasserstein-2 optimal transport for continuous coordinates and establishing Fisher-Rao geodesics for discrete atom types. We further incorporate a geometry-enhanced learning strategy to improve the capture of atomic contexts. Extensive evaluations on the CrossDocked2020 dataset demonstrate that MolPIF outperforms baselines in binding affinity, chemical validity, geometric fidelity and chemical space coverage. Additionally, MolPIF exhibits versatility in lead optimization and offers flexible prior distribution selection (such as Laplace), establishing a robust paradigm for SBDD. Availability: Source code is freely available at https://github.com/BLEACH366/MolPIF. Supplementary information: Supplementary data are available at Bioinformatics.

Chain-of-Thought (CoT) prompting significantly enhances model reasoning, yet its internal mechanisms remain poorly understood. We analyze CoT's operational principles by reversely tracing information flow across decoding, projection, and activation phases. Our quantitative analysis suggests that CoT may serve as a decoding space pruner, leveraging answer templates to guide output generation, with higher template adherence strongly correlating with improved performance. Furthermore, we surprisingly find that CoT modulates neuron engagement in a task-dependent manner: reducing neuron activation in open-domain tasks, yet increasing it in closed-domain scenarios. These findings offer a novel mechanistic interpretability framework and critical insights for enabling targeted CoT interventions to design more efficient and robust prompts. We released our code and data at https://anonymous.4open.science/r/cot-D247.

The rapid advancement of video diffusion models has been hindered by fundamental limitations in temporal modeling, particularly the rigid synchronization of frame evolution imposed by conventional scalar timestep variables. While task-specific adaptations and autoregressive models have sought to address these challenges, they remain constrained by computational inefficiency, catastrophic forgetting, or narrow applicability. In this work, we present \textbf{Pusa} V1.0, a versatile model that leverages \textbf{vectorized timestep adaptation (VTA)} to enable fine-grained temporal control within a unified video diffusion framework. Note that VTA is a non-destructive adaptation, which means that it fully preserves the capabilities of the base model. Unlike conventional methods like Wan-I2V, which finetune a base text-to-video (T2V) model with abundant resources to do image-to-video (I2V), we achieve comparable results in a zero-shot manner after an ultra-efficient finetuning process based on VTA. Moreover, this method also unlocks many other zero-shot capabilities simultaneously, such as start-end frames and video extension -- all without task-specific training. Meanwhile, it keeps the T2V capability from the base model. Mechanistic analyses also reveal that our approach preserves the foundation model's generative priors while surgically injecting temporal dynamics, avoiding the combinatorial explosion inherent to the vectorized timestep. This work establishes a scalable, efficient, and versatile paradigm for next-generation video synthesis, democratizing high-fidelity video generation for research and industry alike.

Advances in vision-based sensors and computer vision algorithms have significantly improved the analysis and understanding of traffic scenarios. To facilitate the use of these improvements for road safety, this survey systematically categorizes the critical elements that demand attention in traffic scenarios and comprehensively analyzes available vision-driven tasks and datasets. Compared to existing surveys that focus on isolated domains, our taxonomy categorizes attention-worthy traffic entities into two main groups that are anomalies and normal but critical entities, integrating ten categories and twenty subclasses. It establishes connections between inherently related fields and provides a unified analytical framework. Our survey highlights the analysis of 35 vision-driven tasks and comprehensive examinations and visualizations of 73 available datasets based on the proposed taxonomy. The cross-domain investigation covers the pros and cons of each benchmark with the aim of providing information on standards unification and resource optimization. Our article concludes with a systematic discussion of the existing weaknesses, underlining the potential effects and promising solutions from various perspectives. The integrated taxonomy, comprehensive analysis, and recapitulatory tables serve as valuable contributions to this rapidly evolving field by providing researchers with a holistic overview, guiding strategic resource selection, and highlighting critical research gaps.

The acquisition of accurately coloured, balanced images in an optical microscope can be a challenge even for experienced microscope operators. This article presents an entirely automatic mechanism for balancing the white level that allows the correction of the microscopic colour images adequately. The results of the algorithm have been confirmed experimentally on a set of two hundred microscopic images. The images contained scans of three microscopic specimens commonly used in pathomorphology. Also, the results achieved were compared with other commonly used white balance algorithms in digital photography. The algorithm applied in this work is more effective than the classical algorithms used in colour photography for microscopic images stained with hematoxylin-phloxine-saffron and for immunohistochemical staining images.

Reusable skills play a key role in improving LLM-based agents, but existing skill-evolution methods often fail to ensure that evolved skills both cover the knowledge required by the task and remain aligned with the target task. As a result, evolved skills could be incomplete or irrelevant. To address this limitation, we propose AlignEvoSkill, a skill-evolution framework that jointly models knowledge coverage and task alignment. Given failed task trajectories, AlignEvoSkill first identifies task-relevant knowledge tags, retrieves complementary prior skills, and adapts them into candidate skills that address missing knowledge. It then selects high-quality candidates using a joint filtering criterion based on knowledge-coverage and task-alignment scores. Experiments on 3 benchmarks with4 LLM backbones show a 34.7% relative gain of AlignEvoSkill over the non-evolution baseline and achieves a new SOTA in skill evolution with lower cost.

Detecting deceptive conversations on dynamic platforms is increasingly difficult due to evolving language patterns and Concept Drift (CD)-i.e., semantic or topical shifts that alter the context or intent of interactions over time. These shifts can obscure malicious intent or mimic normal dialogue, making accurate classification challenging. While Large Language Models (LLMs) show strong performance in natural language tasks, they often struggle with contextual ambiguity and hallucinations in risk-sensitive scenarios. To address these challenges, we present a Domain Knowledge (DK)-Enhanced LLM framework that integrates pretrained LLMs with structured, task-specific insights to perform fraud and concept drift detection. The proposed architecture consists of three main components: (1) a DK-LLM module to detect fake or deceptive conversations; (2) a drift detection unit (OCDD) to determine whether a semantic shift has occurred; and (3) a second DK-LLM module to classify the drift as either benign or fraudulent. We first validate the value of domain knowledge using a fake review dataset and then apply our full framework to SEConvo, a multiturn dialogue dataset that includes various types of fraud and spam attacks. Results show that our system detects fake conversations with high accuracy and effectively classifies the nature of drift. Guided by structured prompts, the LLaMA-based implementation achieves 98% classification accuracy. Comparative studies against zero-shot baselines demonstrate that incorporating domain knowledge and drift awareness significantly improves performance, interpretability, and robustness in high-stakes NLP applications.

Out-of-distribution (OOD) detection attempts to distinguish outlier samples to prevent models trained on the in-distribution (ID) dataset from producing unavailable outputs. Most OOD detection methods require many IID samples for training, which seriously limits their real-world applications. To this end, we target a challenging setting: few-shot OOD detection, where {Only a few {\em labeled ID} samples are available.} Therefore, few-shot OOD detection is much more challenging than the traditional OOD detection setting. Previous few-shot OOD detection works ignore the distinct diversity between different classes. In this paper, we propose a novel network: Adaptive Multi-prompt Contrastive Network (AMCN), which adapts the ID-OOD separation boundary by learning inter- and intra-class distribution. To compensate for the absence of OOD and scarcity of ID {\em image samples}, we leverage CLIP, connecting text with images, engineering learnable ID and OOD {\em textual prompts}. Specifically, we first generate adaptive prompts (learnable ID prompts, label-fixed OOD prompts and label-adaptive OOD prompts). Then, we generate an adaptive class boundary for each class by introducing a class-wise threshold. Finally, we propose a prompt-guided ID-OOD separation module to control the margin between ID and OOD prompts. Experimental results show that AMCN outperforms other state-of-the-art works.

Machine unlearning removes certain training data points and their influence from AI models (e.g., when a data owner revokes their consent to allow models to learn from the data). In this position paper, we propose to lift data-tracing machine unlearning to knowledge-tracing for foundation models (FMs). We support this position based on practical needs and insights from cognitive studies. Practically, tracing data cannot meet the diverse unlearning requests for FMs, which may be from regulators, enterprise users, product teams, etc., who have no access to FMs' massive training data. Instead, it is convenient for these parties to issue an unlearning request about the knowledge or capability FMs (should not) possess. Cognitively, knowledge-tracing unlearning aligns with how the human brain forgets more closely than tracing individual training data points does. We further discuss the nontrivial challenges in the knowledge-tracing machine unlearning paradigm. Finally, we provide a concrete case study about a vision-language FM to illustrate how an unlearner might instantiate the knowledge-tracing machine unlearning paradigm. Code is available at: https://1yuwen.github.io/Knowledge-Tracing-MU-Page.

Computational Music Generation is evolving towards non-conventional styles, demanding methods that enable precise and controllable blending of diverse music elements. In this work, we present a method for fine grained control using inference-time interventions on an autoregressive generative transformer, MusicGen. Through our approach, we achieve genre control by steering the residual stream using weights of a linear probe on it. By framing activation steering as a human-controllable interaction, our work highlights how interpretable model behaviors can empower in co-creative music generation.Audio samples demonstrating our method are available on our demo page.