Current evaluation methods for autonomous driving prediction models rely heavily on simplistic metrics such as Average Displacement Error (ADE) and Final Displacement Error (FDE). While these metrics offer basic performance assessments, they fail to capture the nuanced behavior of prediction modules under complex, interactive, and safety-critical driving scenarios. For instance, existing benchmarks do not distinguish the influence of nearby versus distant agents, nor systematically test model robustness across varying multi-agent interactions. This paper addresses this critical gap by proposing a novel testing framework that evaluates prediction performance under diverse scene structures, saying, map context, agent density and spatial distribution. Through extensive empirical analysis, we quantify the differential impact of agent proximity on target trajectory prediction and identify scenario-specific failure cases that are not exposed by traditional metrics. Our findings highlight key vulnerabilities in current state-of-the-art prediction models and demonstrate the importance of scenario-aware evaluation. The proposed framework lays the groundwork for rigorous, safety-driven prediction validation, contributing significantly to the identification of failure-prone corner cases and the development of robust, certifiable prediction systems for autonomous vehicles.

Harmful fine-tuning issues present significant safety challenges for fine-tuning-as-a-service in large language models. Existing alignment-stage defenses, e.g., Vaccine, Repnoise, Booster, and T-Vaccine, mitigate harmful fine-tuning issues by enhancing the model's robustness during the alignment phase. While these methods have been proposed to mitigate the issue, they often overlook a critical upstream factor: the role of the original safety-alignment data. We observe that their defense performance and computational efficiency remain constrained by the quality and composition of the alignment dataset. To address this limitation, we propose Pharmacist, a safety alignment data curation solution that enhances defense against harmful fine-tuning by selecting a high-quality and safety-critical core subset from the original alignment data. The core idea of Pharmacist is to train an alignment data selector to rank alignment data. Specifically, up-ranking high-quality and safety-critical alignment data, down-ranking low-quality and non-safety-critical data. Empirical results indicate that models trained on datasets selected by Pharmacist outperform those trained on datasets selected by existing selection methods in both defense and inference performance. In addition, Pharmacist can be effectively integrated with mainstream alignment-stage defense methods. For example, when applied to RepNoise and T-Vaccine, using the dataset selected by Pharmacist instead of the full dataset leads to improvements in defense performance by 2.60\% and 3.30\%, respectively, and enhances inference performance by 3.50\% and 1.10\%. Notably, it reduces training time by 56.83\% and 57.63\%, respectively. Our code is available at https://github.com/Lslland/Pharmacist.

Document summarization enables efficient extraction of user-relevant content but is inherently shaped by individual subjectivity, making it challenging to identify subjective salient information in multifaceted documents. This complexity underscores the necessity for personalized summarization. However, training models for personalized summarization has so far been challenging, particularly because diverse training data containing both user preference history (i.e., click-skip trajectory) and expected (gold-reference) summaries are scarce. The MS/CAS PENS dataset is a valuable resource but includes only preference history without target summaries, preventing end-to-end supervised learning, and its limited topic-transition diversity further restricts generalization. To address this, we propose $\mathrm{PerAugy}$, a novel cross-trajectory shuffling and summary-content perturbation based data augmentation technique that significantly boosts the accuracy of four state-of-the-art baseline (SOTA) user-encoders commonly used in personalized summarization frameworks (best result: $\text{0.132}$$\uparrow$ w.r.t AUC). We select two such SOTA summarizer frameworks as baselines and observe that when augmented with their corresponding improved user-encoders, they consistently show an increase in personalization (avg. boost: $\text{61.2\%}\uparrow$ w.r.t. PSE-SU4 metric). As a post-hoc analysis of the role of induced diversity in the augmented dataset by \peraugy, we introduce three dataset diversity metrics -- $\mathrm{TP}$, $\mathrm{RTC}$, and \degreed\ to quantify the induced diversity. We find that $\mathrm{TP}$ and $\mathrm{DegreeD}$ strongly correlate with user-encoder performance on the PerAugy-generated dataset across all accuracy metrics, indicating that increased dataset diversity is a key factor driving performance gains.

The quickly growing popularity of AI companions poses risks to mental health, personal wellbeing, and social relationships. Past work has identified many individual factors that can drive human-companion interaction, but we know little about how these factors interact and evolve over time. In Study 1, we surveyed AI companion users (N = 303) to map the psychological pathway from users' mental models of the agent to parasocial experiences, social interaction, and the psychological impact of AI companions. Participants' responses foregrounded multiple interconnected variables (agency, parasocial interaction, and engagement) that shape AI companionship. In Study 2, we conducted a longitudinal study with a subset of participants (N = 110) using a new generic chatbot. Participants' perceptions of the generic chatbot significantly converged to perceptions of their own companions by Week 3. These results suggest a longitudinal model of AI companionship development and demonstrate an empirical method to study human-AI companionship.

Deep learning models have attracted lots of research attention in time series classification (TSC) task in the past two decades. Recently, deep neural networks (DNN) have surpassed classical distance-based methods and achieved state-of-the-art performance. Despite their promising performance, deep neural networks (DNNs) have been shown to rely on spurious correlations present in the training data, which can hinder generalization. For instance, a model might incorrectly associate the presence of grass with the label ``cat" if the training set have majority of cats lying in grassy backgrounds. However, the shortcut behavior of DNNs in time series remain under-explored. Most existing shortcut work are relying on external attributes such as gender, patients group, instead of focus on the internal bias behavior in time series models. In this paper, we take the first step to investigate and establish point-based shortcut learning behavior in deep learning time series classification. We further propose a simple detection method based on other class to detect shortcut occurs without relying on test data or clean training classes. We test our proposed method in UCR time series datasets.

Conventional continual pretraining (CPT) for large language model (LLM) domain adaptation often suffers from catastrophic forgetting and limited domain capacity. Existing strategies adopt layer expansion, introducing additional trainable parameters to accommodate new knowledge. However, the uniform expansion and updates still entangle general and domain learning, undermining its effectiveness. Our pilot studies reveal that LLMs exhibit functional specialization, where layers and units differentially encode general-critical capabilities, suggesting that parameter expansion and optimization should be function-aware. We then propose ADEPT, Adaptive Expansion and Dynamic Decoupled Tuning for continual pretraining, a two-stage framework for domain-adaptive CPT. ADEPT first performs General-Competence Guided Selective Layer Expansion, duplicating layers least critical for the general domain to increase representational capacity while minimizing interference with general knowledge. It then applies Adaptive Unit-Wise Decoupled Tuning, disentangling parameter units within expanded layers according to their general-domain importance and assigning asymmetric learning rates to balance knowledge injection and retention. Experiments on mathematical and medical benchmarks show that ADEPT outperforms full-parameter CPT by up to 5.76% on the general domain and 5.58% on the target domain with only 15% of parameters tuned and less than 50% training time. Ablation studies, theoretical analysis, and extended investigations further demonstrate the necessity of targeted expansion and decoupled optimization, providing new principles for efficient and robust domain-adaptive CPT. Our code is open-sourced at https://github.com/PuppyKnightUniversity/ADEPT

Abstract-- Deep learning's remarkable performance in natural language processing (NLP) faces critical interpretability challenges, particularly in high-stakes domains like healthcare and finance where model transparency is essential. While concept bottleneck models (CBMs) have enhanced interpretabil-ity in computer vision by linking predictions to human-understandable concepts, their adaptation to NLP remains understudied with persistent limitations. Existing approaches either enforce rigid binary concept activations that degrade textual representation quality or obscure semantic interpretability through latent concept embeddings, while failing to capture dynamic concept interactions crucial for understanding linguistic nuances like negation or contextual modification. This paper proposes the C oncept L anguage M odel N etwork (CLMN), a novel neural-symbolic framework that reconciles performance and interpretability through continuous concept embeddings enhanced by fuzzy logic-based reasoning. CLMN addresses the information loss in traditional CBMs by projecting concepts into an interpretable embedding space while preserving human-readable semantics, and introduces adaptive concept interaction modeling through learnable neural-symbolic rules that explicitly represent how concepts influence each other and final predictions. By supplementing original text features with concept-aware representations and enabling automatic derivation of interpretable logic rules, our framework achieves superior performance on multiple NLP benchmarks while providing transparent explanations.

When modeling a given type of data, we consider it to involve two key aspects: 1) identifying relevant elements (e.g., image pixels or textual words) to a central element, as in a convolutional receptive field, or to a query element, as in self-attention, and 2) encoding these tokens effectively. Self-attention can adaptively identify these elements but relies on absolute positional embedding for structural representation learning. In contrast, convolution encodes elements in a relative manner, yet their fixed kernel size limits their ability to adaptively select the relevant elements. In this paper, we introduce Translution, an operation that unifies the adaptive identification capability of self-attention and the relative encoding advantage of convolution. However, this integration leads to a substantial increase in the number of parameters, exceeding most currently available computational resources. Therefore, we propose a lightweight variant of Translution, named α-Translution. Experiments on computer vision and natural language processing tasks show that Translution (including α-Translution) achieves superior accuracy compared to self-attention. The code is available at https://github.com/hehefan/Translution.

The three-dimensional bin packing problem (3D-BPP) is widely applied in logistics and warehousing. Existing learning-based approaches often neglect practical stability-related constraints and exhibit limitations in generalizing across diverse bin dimensions. To address these limitations, we propose a novel deep reinforcement learning framework, One4Many-StablePacker (O4M-SP). The primary advantage of O4M-SP is its ability to handle various bin dimensions in a single training process while incorporating support and weight constraints common in practice. Our training method introduces two innovative mechanisms. First, it employs a weighted reward function that integrates loading rate and a new height difference metric for packing layouts, promoting improved bin utilization through flatter packing configurations. Second, it combines clipped policy gradient optimization with a tailored policy drifting method to mitigate policy entropy collapse, encouraging exploration at critical decision nodes during packing to avoid suboptimal solutions. Extensive experiments demonstrate that O4M-SP generalizes successfully across diverse bin dimensions and significantly outperforms baseline methods. Furthermore, O4M-SP exhibits strong practical applicability by effectively addressing packing scenarios with stability constraints.

Medical vision-language models (VLMs) excel at image-text understanding but typically rely on a single-pass reasoning that neglects localized visual cues. In clinical practice, however, human experts iteratively scan, focus, and refine the regions of interest before reaching a final diagnosis. To narrow this machine-human perception gap, we introduce ViT AR, a novel VLM framework that emulates the iterative reasoning process of human experts through a cognitive chain of "think-act-rethink-answer". ViT AR treats medical images as interactive objects, enabling models to engage multi-step visual reasoning. To support this approach, we curate a high-quality instruction dataset comprising 1K interactive examples that encode expert-like diagnostic behaviors. In addition, a 16K visual question answering training data has been curated towards fine-grained visual diagnosis. We introduce a two-stage training strategy that begins with supervised fine-tuning to guide cognitive trajectories, followed by the reinforcement learning to optimize decision-making. Extensive evaluations demonstrate that ViT AR outperforms strong state-of-the-art models. Visual attention analysis reveals that from the "think" to "rethink" rounds, ViT AR increasingly anchors visual grounding to clinically critical regions and maintains high attention allocation to visual tokens during reasoning, providing mechanistic insight into its improved performance. These findings demonstrate that embedding expert-style iterative thinking chains into VLMs enhances both performance and trustworthiness of medical AI. Medical vision-language models (VLMs) have evolved from task-specific architectures to versatile frameworks, advancing large-scale medical image annotation (Xie et al., 2024), outcome prediction (Zhong et al., 2025), and clinical reasoning (Chen et al., 2024a). Powered by large language models (LLMs), systems such as LLaV A-Med (Li et al., 2023) and Lingshu (Xu et al., 2025) can engage human-like clinical dialogues and act as visual assistants. Nevertheless, current VLMs typically perform a single-pass inference strategy (Zhang et al., 2024), generating predictions from the entire images without explicitly identifying key visual cues that is vital for decision-making. In the realm of medical imaging diagnosis, human experts follow an iterative cognitive process essentially comprising a multiscale observation (Aggarwal et al., 2021). Clinicians begin with a global image examination to locate suspicious regions of interest (ROIs).