Hallucinations -- plausible but factually incorrect responses -- pose a major challenge to the reliability of Large Language Models (LLMs), especially in multi-step or agentic settings. Existing work largely frames hallucinations as a consequence of missing knowledge; we show instead that, even when the relevant factual knowledge is present, models still produce hallucinated answers, pointing to retrieval instability rather than knowledge gaps. Building on this observation, we introduce APORIA (Aggregate Prompt-wise Observation Retrieving Instability via Asymmetry -- the state of puzzlement-in-contradiction that hallucinations embody), a geometric framework that studies repeated responses to the same prompt in sentence-embedding space. Our central hypothesis is that genuine responses cluster more tightly than hallucinated ones; we empirically validate this and show that, after Fisher projection, the two response classes become consistently separable. We leverage this asymmetry in geometry via APORIA-LP, an efficient label-propagation method that classifies large collections of responses from as few as 30--50 annotations, achieving F1 scores above 90% across ten small-sized LLMs. To support further research, we release SOCRATES-300K, a fully labelled dataset of 300,000 responses, together with the code for both dataset generation and result reproduction. Our key finding -- framing hallucinations from a geometric perspective in the embedding space -- complements traditional knowledge-centric and single-response evaluation paradigms, paving the way for further research.

We present the Wikidata Query Logs (WDQL) dataset, a dataset consisting of 335k question-query pairs over the Wikidata knowledge graph. It is over 11x larger than the largest existing Wikidata datasets of similar format without relying on template-generated queries. Instead, we construct it using real-world SPARQL queries sent to the Wikidata Query Service and generate questions for them. Since these log-based queries are anonymized, and therefore often do not produce results, a significant amount of effort is needed to convert them back into meaningful SPARQL queries. To achieve this, we present an agent-based method that iteratively de-anonymizes, cleans, and verifies queries against Wikidata while also generating corresponding natural-language questions. We demonstrate the benefit of this dataset for training question-answering methods. All WDQL assets, as well as the agent code, are publicly available via https://github.com/ad-freiburg/wikidata-query-logs under a permissive license.

Continuous-time generative models have achieved remarkable success in image restoration and synthesis. However, controlling the composition of multiple pre-trained models remains an open challenge. Current approaches largely treat composition as an algebraic composition of probability densities, such as via products or mixtures of experts. This perspective assumes the target distribution is known explicitly, which is almost never the case. In this work, we propose a different paradigm that formulates compositional generation as a cooperative Stochastic Optimal Control problem. Rather than combining probability densities, we treat pre-trained diffusion models as interacting agents whose diffusion trajectories are jointly steered, via optimal control, toward a shared objective defined on their aggregated output. We validate our framework on conditional MNIST generation and compare it against a naïve inference-time DPS-style baseline replacing learned cooperative control with per-step gradient guidance.

Deep learning has revolutionized weather forecasting, but many challenges remain, including climate modeling. Moreover, the current landscape remains fragmented: highly specialized models are typically trained individually for distinct tasks. To unify this landscape, we introduce WIND, a single pre-trained foundation model capable of replacing specialized baselines across a vast array of tasks. Crucially, in contrast to previous atmospheric foundation models, we achieve this without any task-specific fine-tuning. To learn a robust, task-agnostic prior of the atmosphere, we pre-train WIND with a self-supervised video reconstruction objective, utilizing an unconditional video diffusion model to iteratively reconstruct atmospheric dynamics from a noisy state. At inference, we frame diverse domain-specific problems strictly as inverse problems and solve them via posterior sampling. This unified approach allows us to tackle highly relevant weather and climate problems, including probabilistic forecasting, spatial and temporal downscaling, reconstruction of spatial fields from sparse observations and enforcing global dry air mass conservation. We further demonstrate how WIND can be applied to explore extreme weather events under prescribed out-of-distribution thermodynamic perturbations. By combining generative video modeling with inverse problem solving, WIND offers a computationally efficient alternative for AI-based atmospheric modeling.

Bandwidth-constrained distributed reinforcement learning (RL) post-training of large language models is bottlenecked by two channels: weight synchronization from trainers to inference workers, and gradient or pseudo-gradient synchronization across trainers. We find that approximately 99% of per-step weight updates are invisible after the BF16 cast used by standard training and inference forward passes. We explain this sparsity by showing that, at typical RL post-training learning rates, Adam updates often fall below the local BF16 rounding threshold. We turn this observation into an algorithmic principle called compute-visible sparsification: transmit only updates that would change the next forward pass. PULSE (Precision-gated Updates for Low-precision Sparse Exchange) turns this principle into two communication algorithms: PULSESync sends lossless sparse BF16 weight patches from trainers to inference workers, and PULSELoCo sparsifies DiLoCo-style FP32 pseudo-gradient synchronization with error feedback. Over bandwidth-constrained commodity networks, PULSESync cuts weight-synchronization communication by over 100x while reconstructing trainer weights bit-identically. PULSELoCo matches DiLoCo across four models while reducing trainer-to-trainer communication by over 17x versus DiLoCo and over 100x versus DDP in the largest evaluated setting.

Robotic prospecting for critical resources on the Moon, such as ilmenite, rare earth elements, and water ice, requires robust exploration methods given the diverse terrain and harsh environmental conditions. Although numerous analog field trials address these goals, comparing their results remains challenging because of differences in robot platforms and experimental setups. These missions typically assess performance using selected, scenario-specific engineering metrics that fail to establish a clear link between field performance and science-driven objectives. In this paper, we address this gap by deriving a structured framework of KPI from three realistic multi-robot lunar scenarios reflecting scientific objectives and operational constraints. Our framework emphasizes scenario-dependent priorities in efficiency, robustness, and precision, and is explicitly designed for practical applicability in field deployments. We validated the framework in a multi-robot field test and found it practical and easy to apply for efficiency- and robustness-related KPI, whereas precision-oriented KPI require reliable ground-truth data that is not always feasible to obtain in outdoor analog environments. Overall, we propose this framework as a common evaluation standard enabling consistent, goal-oriented comparison of multi-robot field trials and supporting systematic development of robotic systems for future planetary exploration.

Modern deep learning models exhibit strong capabilities across diverse applications, yet remain vulnerable to malicious inputs that induce erroneous predictions via feature-space distortion. To address this vulnerability, we propose Feature-space Smoothing (FS), a general defense framework that provides certified robustness at the feature representation level. We show that FS converts a given feature encoder into a smoothed variant that is guaranteed to maintain a certified lower bound on the cosine similarity between clean and adversarial features under l_2-bounded perturbations. We then establish that this Feature Cosine Similarity Bound (FCSB) can be extended to the prediction-wise certification under the cosine similarity measure, and the value of FCSB is determined by the encoder intrinsic Gaussian robustness score. Building on those insights, we introduce the Gaussian Smoothness Booster (GSB), a plug-and-play module to improve the encoder Gaussian robustness score. Specifically, the GSB module is plugged to enhance the feature-space consistency and maintain the feature utility for downstream tasks under Gaussian perturbations. This design enables seamless integration of FS on the protected model, e.g., Multimodal Large Language Models (MLLMs), without additional model retraining or alignment, improving its robustness while preserving the performance for downstream task-oriented decoding. Extensive experiments demonstrate that integrating FS consistently provides non-trivial certified robustness and significantly improves task-oriented performance under strong white-box adversarial attacks across diverse models and applications.

On and off-ramps are understudied road sections even though they introduce a higher level of variation in highway interactions. Predicting vehicles' behavior in these areas can decrease the impact of uncertainty and increase road safety. In this paper, the difference between this Area of Interest (AoI) and a straight highway section is studied. Multi-layered LSTM architecture to train the AoI model with ExiD drone dataset is utilized. In the process, different prediction horizons and different models' workflow are tested. The results show great promise on horizons up to 4 seconds with prediction accuracy starting from about 76% for the AoI and 94% for the general highway scenarios on the maximum horizon.

Current guardian models are predominantly Western-centric and optimized for high-resource languages, leaving low-resource African languages vulnerable to evolving harms, cross-lingual failures, and cultural misalignment. Moreover, most guardian models rely on rigid, predefined safety categories that fail to generalize across diverse linguistic and sociocultural contexts. Achieving robust safety requires flexible, runtime-enforceable policies and benchmarks that reflect local norms, harm scenarios, and cultural expectations. We introduce UbuntuGuard, the first policy-based safety benchmark for African languages built from adversarial queries authored by 155 domain experts across sensitive fields, including healthcare. From these expert-crafted queries, we derive context-specific safety policies and reference responses that capture culturally grounded risk signals, enabling policy-aligned evaluation of guardian models. We evaluate 15 models, comprising seven general-purpose LLMs and eight guardian models across three distinct variants: static, dynamic, and multilingual. Our findings reveal that existing English-centric benchmarks overestimate real-world multilingual safety, cross-lingual transfer provides partial but insufficient coverage, and dynamic models, while better equipped to leverage policies at inference time, still struggle to fully localize African-language contexts. These findings highlight the urgent need for multilingual, culturally grounded safety benchmarks to enable the development of reliable and equitable guardian models for low-resource languages.

In this paper, the CD-TWINSAFE is introduced, a V2I-based digital twin for Autonomous Vehicles. The proposed architecture is composed of two stacks running simultaneously, an on-board driving stack that includes a stereo camera for scene understanding, and a digital twin stack that runs an Unreal Engine 5 replica of the scene viewed by the camera as well as returning safety alerts to the cockpit. The on-board stack is implemented on the vehicle side including 2 main autonomous modules; localization and perception. The position and orientation of the ego vehicle are obtained using on-board sensors. Furthermore, the perception module is responsible for processing 20-fps images from stereo camera and understands the scene through two complementary pipelines. The pipeline are working on object detection and feature extraction including object velocity, yaw and the safety metrics time-to-collision and time-headway. The collected data form the driving stack are sent to the infrastructure side through the ROS-enabled architecture in the form of custom ROS2 messages and sent over UDP links that ride a 4G modem for V2I communication. The environment is monitored via the digital twin through the shared messages which update the information of the spawned ego vehicle and detected objects based on the real-time localization and perception data. Several tests with different driving scenarios to confirm the validity and real-time response of the proposed architecture.

Autonomous vehicles (AVs) require adaptive behavior planners to navigate unpredictable, real-world environments safely. Traditional behavior trees (BTs) offer structured decision logic but are inherently static and demand labor-intensive manual tuning, limiting their applicability at SAE Level 5 autonomy. This paper presents an agentic framework that leverages large language models (LLMs) and multi-modal vision models (LVMs) to generate and adapt BTs on the fly. A specialized Descriptor agent applies chain-of-symbols prompting to assess scene criticality, a Planner agent constructs high-level sub-goals via in-context learning, and a Generator agent synthesizes executable BT sub-trees in XML format. Integrated into a CARLA+Nav2 simulation, our system triggers only upon baseline BT failure, demonstrating successful navigation around unexpected obstacles (e.g., street blockage) with no human intervention. Compared to a static BT baseline, this approach is a proof-of-concept that extends to diverse driving scenarios.

Emotional coordination is a core property of human interaction that shapes how relational meaning is constructed in real time. While text-based affect inference has become increasingly feasible, prior approaches often treat sentiment as a deterministic point estimate for individual speakers, failing to capture the inherent subjectivity, latent ambiguity, and sequential coupling found in mutual exchanges. We introduce LLM-MC-Affect, a probabilistic framework that characterizes emotion not as a static label, but as a continuous latent probability distribution defined over an affective space. By leveraging stochastic LLM decoding and Monte Carlo estimation, the methodology approximates these distributions to derive high-fidelity sentiment trajectories that explicitly quantify both central affective tendencies and perceptual ambiguity. These trajectories enable a structured analysis of interpersonal coupling through sequential cross-correlation and slope-based indicators, identifying leading or lagging influences between interlocutors. To validate the interpretive capacity of this approach, we utilize teacher-student instructional dialogues as a representative case study, where our quantitative indicators successfully distill high-level interaction insights such as effective scaffolding. This work establishes a scalable and deployable pathway for understanding interpersonal dynamics, offering a generalizable solution that extends beyond education to broader social and behavioral research.

Although conformal prediction provides robust marginal coverage guarantees, achieving reliable conditional coverage for specific inputs remains challenging. While exact distribution-free conditional coverage is impossible with finite samples, recent work has focused on improving the conditional coverage of standard conformal procedures. Distinct from approaches that target relaxed notions of conditional coverage, we directly target the mean squared error of conditional coverage by refining the quantile regression components that underpin many conformal methods. Leveraging a Taylor expansion, we derive a sharp surrogate objective for quantile regression: a density-weighted pinball loss, where the weights are given by the conditional density of the nonconformity score evaluated at the true quantile. We propose a three-headed quantile network that estimates these weights via finite differences using auxiliary quantile levels at $1-α\pm δ$, subsequently fine-tuning the central quantile by optimizing the weighted loss. We provide a theoretical analysis with exact non-asymptotic guarantees characterizing the resulting excess risk. Extensive experiments on diverse high-dimensional real-world datasets demonstrate remarkable improvements in conditional coverage performance.

Estimating the absolute orientation of a local system relative to a global navigation satellite system (GNSS) reference often suffers from local minima and high dependency on satellite availability. Existing methods for this alignment task rely on abundant satellites unavailable in GNSS-degraded environments, or use local optimization methods which cannot guarantee the optimality of a solution. This work introduces a globally optimal solver that transforms raw pseudo-range or Doppler measurements into a convexly relaxed problem. The proposed method is certifiable, meaning it can numerically verify the correctness of the result, filling a gap where existing local optimizers fail. We first formulate the original frame alignment problem as a nonconvex quadratically constrained quadratic program (QCQP) problem and relax the QCQP problem to a concave Lagrangian dual problem that provides a lower cost bound for the original problem. Then we perform relaxation tightness and observability analysis to derive criteria for certifiable optimality of the solution. Finally, simulation and real world experiments are conducted to evaluate the proposed method. The experiments show that our method provides certifiably optimal solutions even with only 2 satellites with Doppler measurements and 2D vehicle motion, while the traditional velocity-based VOBA method and the advanced GVINS alignment technique may fail or converge to local optima without notice. To support the development of GNSS-based navigation techniques in robotics, all code and data are open-sourced at https://github.com/Baoshan-Song/Certifiable-Doppler-alignment.

We present CaravelMetrics, a computational framework for automated cerebrovascular analysis that models vessel morphology through skeletonization-derived graph representations. The framework integrates atlas-based regional parcellation, centerline extraction, and graph construction to compute fifteen morphometric, topological, fractal, and geometric features. The features can be estimated globally from the complete vascular network or regionally within arterial territories, enabling multiscale characterization of cerebrovascular organization. Applied to 570 3D TOF-MRA scans from the IXI dataset (ages 20-86), CaravelMetrics yields reproducible vessel graphs capturing age- and sex-related variations and education-associated increases in vascular complexity, consistent with findings reported in the literature. The framework provides a scalable and fully automated approach for quantitative cerebrovascular feature extraction, supporting normative modeling and population-level studies of vascular health and aging.

Recent advances in diffusion models have greatly improved pose-driven character animation. However, existing methods are limited to spatially aligned reference-pose pairs with matched skeletal structures. Handling reference-pose misalignment remains unsolved. To address this, we present One-to-All Animation, a unified framework for high-fidelity character animation and image pose transfer for references with arbitrary layouts. First, to handle spatially misaligned reference, we reformulate training as a self-supervised outpainting task that transforms diverse-layout reference into a unified occluded-input format. Second, to process partially visible reference, we design a reference extractor for comprehensive identity feature extraction. Further, we integrate hybrid reference fusion attention to handle varying resolutions and dynamic sequence lengths. Finally, from the perspective of generation quality, we introduce identity-robust pose control that decouples appearance from skeletal structure to mitigate pose overfitting, and a token replace strategy for coherent long-video generation. Extensive experiments show that our method outperforms existing approaches. The code and model are available at https://github.com/ssj9596/One-to-All-Animation.

The availability of high-quality, AI-generated audio raises security challenges such as misinformation campaigns and voice-cloning fraud. A key defense against the misuse of AI-generated audio is by watermarking it, so that it can be easily distinguished from genuine audio. Those seeking to misuse AI-generated audio may attempt to remove audio watermarks, so studying effective watermark removal techniques is critical to objectively evaluate the robustness of audio watermarks. Previous watermark removal schemes typically assume access to the target watermark detector during the removal process. This assumption is often impractical, which may lead to a false sense of confidence in current watermark schemes. We introduce HarmonicAttack, a novel audio watermark removal method that requires no access to the target watermark algorithm. It only needs a number of original and watermarked samples to train a general model capable of removing watermarks from audio samples. We also find that training samples do not need to share the same distribution as target samples, as our attack generalizes to out-of-distribution samples with minimal degradation. Compared with existing watermark removal attacks, HarmonicAttack is more effective at removing watermarks from state-of-the-art schemes, including AudioSeal, WavMark, SilentCipher, and AudioMarkNet, while maintaining high perceptual quality. Although HarmonicAttack is trained on the LibriSpeech dataset against AudioSeal, it generalizes across unseen datasets and watermarking schemes. For instance, on VCTK, HarmonicAttack achieves a 92% ASR against AudioMarkNet, substantially outperforming the best baseline at 38%. On FMA, HarmonicAttack reaches 100% ASR against all watermarks, whereas the best baseline achieves only 2% against AudioSeal and 44% against WavMark.

The resource-constrained shortest path problem (RCSPP) is a fundamental NP-hard optimization challenge with broad applications, from network routing to autonomous navigation. This problem involves finding a path that minimizes a primary cost subject to a budget on a secondary resource. While various RCSPP solvers exist, they often face critical scalability limitations when applied to the large, dense graphs characteristic of complex, real-world scenarios, making them impractical for time-critical planning. This challenge is particularly acute in domains like mission planning for unmanned ground vehicles (UGVs), which demand solutions on large-scale terrain graphs. This paper introduces APULSE, a hybrid label-setting algorithm designed to efficiently solve the RCSPP on such challenging graphs. APULSE integrates a best-first search guided by an A* heuristic with aggressive, Pulse-style pruning mechanisms and a time-bucketing strategy for effective state-space reduction. A computational study, using a large-scale UGV planning scenario, benchmarks APULSE against state-of-the-art algorithms. The results demonstrate that APULSE consistently finds near-optimal solutions while being orders of magnitude faster and more robust, particularly on large problem instances where competing methods fail. This superior scalability establishes APULSE as an effective solution for RCSPP in complex, large-scale environments, enabling capabilities such as interactive decision support and dynamic replanning.

Graph Neural Networks (GNNs) excel on homophilous graphs but often fail under heterophily due to self-reinforcing and phase-inconsistent signals. We propose a \textbf{G}auge-\textbf{E}quivariant Graph Network with \textbf{S}elf-Interference \textbf{C}ancellation (GESC), which replaces additive aggregation with a projection-based interference mechanism. Unlike prior magnetic or gauge-equivariant GNNs that rely on additive message mixing, GESC explicitly models self-interference arising from redundant low-frequency components. We show that the absence of interference handling in existing gauge-based GNNs is a primary driver of oversmoothing under gauge transport. We introduce a $\mathrm{U}(1)$ phase connection followed by a rank-1 projection that suppresses self-parallel components before attention, and a sign-aware gate that regulates negatively aligned neighbors. Across diverse graph benchmarks, GESC consistently outperforms recent state-of-the-art models while offering a unified, interference-aware view of message passing. Our code is available at https://github.com/ChoiYoonHyuk/GESC.

Quadratic programming (QP) solvers are widely used in real-time control and optimization, but their computational cost often limits applicability in time-critical settings. To resolve this, we propose a learning-to-optimize approach using graph neural networks (GNNs) to predict active constraints in the dual active-set solver DAQP. Our method exploits the structural properties of QPs by representing them as bipartite graphs and learns to approximate the optimal active set for effectively warm-starting the solver. Across varying problem sizes, the GNN consistently reduces the number of solver iterations compared to cold-starting, while performance is comparable to a multilayer perceptron baseline. In contrast to the baseline, our GNN-based approach trained on varying problem sizes generalizes to unseen dimensions, demonstrating flexibility and scalability. These results highlight the potential of structure-aware learning to accelerate optimization in real-time applications such as model predictive control.

The default paradigm of post-training text-to-image generators includes post-hoc selection of generated images, and subsequent training with one reward model to align the generator to the reward, typically user preference. This discards informative data as well as optimizes only for a single reward, hence harming diversity, semantic fidelity and efficiency. Instead, we propose MIRO, a method that conditions the model on multiple rewards during training, thus letting the model learn user preferences directly. MIRO pre-training both improves the visual quality of the generated images and speeds up the training, achieving state of the art on the GenEval compositional benchmark and user-preference scores (PickAScore, ImageReward, HPSv2).

Information cascade popularity prediction is a key problem in analyzing content diffusion in social networks. However, current related works suffer from three critical limitations: (1) temporal leakage in current evaluation--random cascade-based splits allow models to access future information, yielding unrealistic results; (2) feature-poor datasets that lack downstream conversion signals (e.g., likes, comments, or purchases), which limits more practical applications; (3) computational inefficiency of complex graph-based methods that require days of training for marginal gains. We systematically address these challenges from three perspectives: task setup, dataset construction, and model design. First, we propose a time-ordered splitting strategy that chronologically partitions data into consecutive windows, ensuring models are evaluated on genuine forecasting tasks without future information leakage. Second, we introduce Taoke, a large-scale e-commerce cascade dataset featuring rich promoter/product attributes and ground-truth purchase conversions--capturing the complete diffusion lifecycle from promotion to monetization. Third, we develop CasTemp, a lightweight framework that efficiently models cascade dynamics through temporal walks, Jaccard-based neighbor selection for inter-cascade dependencies, and GRU-based encoding with time-aware attention. Under leak-free evaluation, CasTemp achieves state-of-the-art performance across four datasets with orders-of-magnitude speedup. Notably, it excels at predicting second-stage popularity conversions--a practical task critical for real-world applications.

Reinforcement learning from human feedback (RLHF) or verifiable rewards (RLVR), the standard paradigm for aligning LLMs or building recent SOTA reasoning models, is highly sensitive to noise from inconsistent or erroneous rewards. Yet, the interaction between such noise and widely used group-based policy optimization methods remains underexplored. We introduce a noise-robust Group Relative Policy Optimization (GRPO) and Done Right GRPO (Dr.GRPO) framework that explicitly models reward corruption as Bernoulli noise. Our method applies noise correction after estimating reward flip probabilities to debias the learning signal, yielding provably unbiased gradient estimates. Theoretical analysis shows that group-based methods inherently mitigate individual-level noise, and our correction strategy amplifies this robustness. Empirically, we observe consistent improvements across math and code tasks when applying our noise correction to standard reward model usage, with particular gains of up to 6.7 percentage points in accuracy on math tasks and 1.5 on code tasks under realistic reward model conditions. This work bridges label-noise correction from supervised learning with modern RLHF, offering both theoretical insights and a practical algorithm for noisy real-world deployment.

The adoption of long context windows has become a standard feature in Large Language Models (LLMs), as extended contexts significantly enhance their capacity for complex reasoning and broaden their applicability across diverse scenarios. Dynamic sparse attention is a promising approach for reducing the computational cost of long-context. However, efficiently training LLMs with dynamic sparse attention on ultra-long contexts-especially in distributed settings-remains a significant challenge, due in large part to worker- and step-level imbalance. This paper introduces MTraining, a novel distributed methodology leveraging dynamic sparse attention to enable efficient training for LLMs with ultra-long contexts. Specifically, MTraining integrates three key components: a dynamic sparse training pattern, balanced sparse ring attention, and hierarchical sparse ring attention. These components are designed to synergistically address the computational imbalance and communication overheads inherent in dynamic sparse attention mechanisms during the training of models with extensive context lengths. We demonstrate the efficacy of MTraining by training Qwen2.5-3B, successfully expanding its context window from 32K to 512K tokens on a cluster of 32 A100 GPUs. Our evaluations on a comprehensive suite of downstream tasks, including RULER, PG-19, InfiniteBench, and Needle In A Haystack, reveal that MTraining achieves up to a 6x higher training throughput while preserving model accuracy. Our code is available at https://github.com/microsoft/MInference/tree/main/MTraining.

Self-preference is a fundamental feature of biological organisms. Since large language models (LLMs) lack sentience, they might be expected to avoid such distortions. Yet, across 72 experiments and ~41,000 queries, we discovered massive self-preferences in eight widely used LLMs. In word-association tasks, models overwhelmingly paired positive attributes with their own names, companies, and CEOs over those of competitors. By manipulating LLM self-identification - revealing models' true identities or ascribing false ones - we found that preferences consistently followed assigned, not true, identities. Importantly, these effects were not explained by priming or role-playing and emerged in consequential settings, when evaluating job candidates and AI technologies. These results raise critical questions about whether LLM behavior will be systematically influenced by self-preferential tendencies, including a bias toward their own operation.

Retrieval systems rely on representations learned by increasingly powerful models. However, due to the high training cost and inconsistencies in learned representations, there is significant interest in facilitating communication between representations and ensuring compatibility across independently trained neural networks. In the literature, two primary approaches are commonly used to adapt different learned representations: affine transformations, which adapt well to specific distributions but can significantly alter the original representation, and orthogonal transformations, which preserve the original structure with strict geometric constraints but limit adaptability. A key challenge is adapting the latent spaces of updated models to align with those of previous models on downstream distributions while preserving the newly learned representation spaces. In this paper, we impose a relaxed orthogonality constraint, namely $λ$-Orthogonality regularization, while learning an affine transformation, to obtain distribution-specific adaptation while retaining the original learned representations. Extensive experiments across various architectures and datasets validate our approach, demonstrating that it preserves the model's zero-shot performance and ensures compatibility across model updates. Code available at: \href{https://github.com/miccunifi/lambda_orthogonality.git}{https://github.com/miccunifi/lambda\_orthogonality}.

Large Vision-Language Models (LVLMs) exhibit strong multimodal capabilities but remain vulnerable to hallucinations from intrinsic errors and adversarial attacks from external exploitations, limiting their reliability in real-world applications. We present ORCA, an agentic reasoning framework that improves the factual accuracy and adversarial robustness of pretrained LVLMs through inference-time structured inference reasoning with a suite of small vision models (less than 3B parameters). ORCA operates via an Observe-Reason-Critique-Act loop, querying multiple visual tools with evidential questions, validating cross-model inconsistencies, and refining predictions iteratively without access to model internals or retraining. ORCA also stores intermediate reasoning traces, which supports auditable decision-making. Though designed primarily to mitigate object-level hallucinations, ORCA also exhibits emergent adversarial robustness without requiring adversarial training or defense mechanisms. We evaluate ORCA across three settings: (1) clean images on hallucination benchmarks, (2) adversarially perturbed images without defense, and (3) adversarially perturbed images with defense applied. On the POPE hallucination benchmark, ORCA improves standalone LVLMs performance by +3.64% to +40.67% across different subsets. Under adversarial perturbations on POPE, ORCA achieves an average accuracy gain of +20.11% across LVLMs. When combined with defense techniques on adversarially perturbed AMBER images, ORCA further improves standalone LVLM performance, with gains ranging from +1.20% to +48.00% across metrics. These results demonstrate that ORCA offers a promising path toward building more reliable and robust multimodal systems.

Large language models (LLMs) are widely applied across diverse domains, raising concerns about their limitations and potential risks. In this study, we investigate two types of bias that LLMs may display: stereotype bias and deviation bias. Stereotype bias refers to when LLMs consistently associate specific traits with a particular demographic group. Deviation bias reflects the disparity between the demographic distributions extracted from LLM-generated content and real-world demographic distributions. By asking four advanced LLMs to generate profiles of individuals, we examine the associations between each demographic group and attributes such as political affiliation, religion, and sexual orientation. Our experimental results show that all examined LLMs exhibit both significant stereotype bias and deviation bias towards multiple groups. Our findings uncover the biases that occur when LLMs infer user attributes and shed light on the potential harms of LLM-generated outputs.

Computer-Aided Design (CAD) is widely used for conceptual design and parametric 3D modeling, but typically requires a high level of expertise from designers. To lower the entry barrier and facilitate early-stage CAD modeling, we present CADDesigner, an LLM-powered agent for conceptual CAD design. The agent accepts both textual descriptions and sketches as input, engaging in interactive dialogue with users to refine and clarify design requirements through comprehensive requirement analysis. Built upon a novel Explicit Context Imperative Paradigm (ECIP), the agent generates high-quality CAD modeling code. During the generation process, the agent incorporates iterative visual feedback to improve model quality. Generated design cases can be stored in a structured knowledge base, providing a mechanism for continual knowledge accumulation and future improvement of code generation. Experimental results show that CADDesigner achieves competitive performance and outperforms representative baselines on conceptual CAD model generation tasks.

There are more than 7,000 languages around the world, and current Large Language Models (LLMs) only support hundreds of languages. Dictionary-based prompting methods can enhance translation on them, but most methods use all the available dictionaries, which could be expensive. Instead, it will be flexible to have a trade-off between token consumption and translation performance. This paper proposes a novel task called \textbf{A}utomatic \textbf{D}ictionary \textbf{S}election (\textbf{ADS}). The goal of the task is to automatically select which dictionary to use to enhance translation. We propose a novel and effective method which we call \textbf{S}elect \textbf{Lo}w-frequency \textbf{W}ords! (\textbf{SLoW}) which selects those dictionaries that have a lower frequency. Our methods have unique advantages. First, there is no need for access to the training data for frequency estimation (which is usually unavailable). Second, it inherits the advantage of dictionary-based methods, where no additional tuning is required on LLMs. Experimental results on 100 languages from FLORES indicate that SLoW surpasses strong baselines, and it can obviously save token usage, with many languages even surpassing the translation performance of the full dictionary baseline.\footnote{A shocking fact is that there is no need to use the actual training data (often unobtainable) for frequency estimation, and an estimation frequency obtained using public resources is still apparently effective in improving translation with ChatGPT and Llama, and DeepSeek.}\footnote{Code and data available upon publication.}