Recent studies have shown that long chain-of-thought (CoT) reasoning can significantly enhance the performance of large language models (LLMs) on complex tasks. However, this benefit is yet to be demonstrated in the domain of video understanding, since most existing benchmarks lack the reasoning depth required to demonstrate the advantages of extended CoT chains. While recent efforts have proposed benchmarks aimed at video reasoning, the tasks are often knowledge-driven and do not rely heavily on visual content. To bridge this gap, we introduce VideoReasonBench, a benchmark designed to evaluate vision-centric, complex video reasoning. To ensure visual richness and high reasoning complexity, each video in VideoReasonBench depicts a sequence of fine-grained operations on a latent state that is only visible in part of the video. The questions evaluate three escalating levels of video reasoning skills: recalling observed visual information, inferring the content of latent states, and predicting information beyond the video. Under such task setting, models have to precisely recall multiple operations in the video, and perform step-by-step reasoning to get correct final answers for these questions. Using VideoReasonBench, we comprehensively evaluate 18 state-of-the-art multimodal LLMs (MLLMs), finding that most perform poorly on complex video reasoning -- e.g., GPT-4o achieves only 6.9% accuracy -- while the thinking-enhanced Gemini-2.5-Pro significantly outperforms others with 56.0% accuracy. Our investigations on "test-time scaling" further reveal that extended thinking budget, while offering none or minimal benefits on existing video benchmarks, is essential for improving the performance on VideoReasonBench.

Large language models (LLMs) are increasingly integrated in software development, but ensuring correctness in LLM-generated code remains challenging and often requires costly manual review. Verifiable code generation -- jointly generating code, specifications, and proofs of code-specification alignment -- offers a promising path to address this limitation and further unleash LLMs' benefits in coding. Yet, there exists a significant gap in evaluation: current benchmarks often focus on only individual components rather than providing a holistic evaluation framework of all tasks. In this paper, we introduce VERINA (Verifiable Code Generation Arena), a high-quality benchmark enabling a comprehensive and modular evaluation of code, specification, and proof generation as well as their compositions. VERINA consists of 189 manually curated coding tasks in Lean, with detailed problem descriptions, reference implementations, formal specifications, and extensive test suites. Our extensive evaluation of state-of-the-art LLMs reveals significant challenges in verifiable code generation, especially in proof generation, underscoring the need for improving LLM-based theorem provers in verification domains. The best model, OpenAI o3, achieves a 72.6\% code correctness rate, 52.3\% for specification soundness and completeness, and a mere 4.9\% proof success rate (based on one trial per task). We hope VERINA will catalyze progress in verifiable code generation by providing a rigorous and comprehensive benchmark. We release our dataset on https://huggingface.co/datasets/sunblaze-ucb/verina and our evaluation code on https://github.com/sunblaze-ucb/verina.

Dense Associative Memories (DenseAMs) are generalizations of Hopfield networks, which have superior information storage capacity and can store training data points (memories) at local minima of the energy landscape. When the amount of training data exceeds the critical memory storage capacity of these models, new local minima, which are different from the training data, emerge. In Associative Memory these emergent local minima are called $\textit{spurious}\; \textit{states}$, which hinder memory retrieval. In this work, we examine diffusion models (DMs) through the DenseAM lens, viewing their generative process as an attempt of a memory retrieval. In the small data regimes, DMs create distinct attractors for each training sample, akin to DenseAMs below the critical memory storage. As the training data size increases, they transition from memorization to generalization. We identify a critical intermediate phase, predicted by DenseAM theory -- the spurious states. In generative modeling, these states are no longer negative artifacts but rather are the first signs of generative capabilities. We characterize the basins of attraction, energy landscape curvature, and computational properties of these previously overlooked states. Their existence is demonstrated across a wide range of architectures and datasets.

Autonomous mobility systems increasingly operate in dense and dynamic environments where perception occlusions, limited sensing coverage, and multi-agent interactions pose major challenges. While onboard sensors provide essential local perception, they often struggle to maintain reliable situational awareness in crowded urban or indoor settings. This article presents the Cloud-based Autonomous Mobility (CAM) framework, a generalized architecture that integrates infrastructure-based intelligent sensing with cloud-level coordination to enhance autonomous operations. The system deploys distributed Intelligent Sensor Nodes (ISNs) equipped with cameras, LiDAR, and edge computing to perform multi-modal perception and transmit structured information to a cloud platform via high-speed wireless communication. The cloud aggregates observations from multiple nodes to generate a global scene representation for other autonomous modules, such as decision making, motion planning, etc. Real-world deployments in an urban roundabout and a hospital-like indoor environment demonstrate improved perception robustness, safety, and coordination for future intelligent mobility systems.

Text-to-multiview (T2MV) diffusion models have shown great promise in generating multiple views of a scene from a single text prompt. While few-step backbones enable real-time T2MV generation, they often compromise key aspects of generation quality, such as per-view fidelity and cross-view consistency. Reinforcement learning (RL) finetuning offers a potential solution, yet existing approaches designed for single-image diffusion do not readily extend to the few-step T2MV setting, as they neglect cross-view coordination and suffer from weak learning signals in few-step regimes. To address this, we propose MVC-ZigAL, a tailored RL finetuning framework for few-step T2MV diffusion models. Specifically, its core insights are: (1) a new MDP formulation that jointly models all generated views and assesses their collective quality via a joint-view reward; (2) a novel advantage learning strategy that exploits the performance gains of a self-refinement sampling scheme over standard sampling, yielding stronger learning signals for effective RL finetuning; and (3) a unified RL framework that extends advantage learning with a Lagrangian dual formulation for multiview-constrained optimization, balancing single-view and joint-view objectives through adaptive primal-dual updates under a self-paced threshold curriculum that harmonizes exploration and constraint enforcement. Collectively, these designs enable robust and balanced RL finetuning for few-step T2MV diffusion models, yielding substantial gains in both per-view fidelity and cross-view consistency. Code is available at https://github.com/ZiyiZhang27/MVC-ZigAL.

Recent advances have shown success in eliciting strong reasoning abilities in multimodal large language models (MLLMs) through rule-based reinforcement learning (RL) with outcome rewards. However, this paradigm typically lacks supervision over the thinking process leading to the final outcome. As a result, the model may learn sub-optimal reasoning strategies, which can hinder its generalization ability. In light of this, we propose SophiaVL-R1, as an attempt to add reward signals for the thinking process in this paradigm. To achieve this, we first train a thinking reward model that evaluates the quality of the entire thinking process. Given that the thinking reward may be unreliable for certain samples due to reward hacking, we propose the Trust-GRPO method, which assigns a trustworthiness weight to the thinking reward during training. This weight is computed based on the thinking reward comparison of responses leading to correct answers versus incorrect answers, helping to mitigate the impact of potentially unreliable thinking rewards. Moreover, we design an annealing training strategy that gradually reduces the thinking reward over time, allowing the model to rely more on the accurate rule-based outcome reward in later training stages. Experiments show that our SophiaVL-R1 surpasses a series of reasoning MLLMs on various benchmarks (e.g., MathVisita, MMMU), demonstrating strong reasoning and generalization capabilities. Notably, our SophiaVL-R1-7B even outperforms LLaVA-OneVision-72B on most benchmarks, despite the latter having 10 times more parameters. All code, models, and datasets are made publicly available at https://github.com/kxfan2002/SophiaVL-R1.

False negatives pose a critical challenge in vision-language pretraining (VLP) due to the many-to-many correspondence between images and texts in large-scale datasets. These false negatives introduce conflicting supervision signals that degrade the learned embedding space and diminish the effectiveness of hard negative sampling. In this paper, we propose FALCON (False-negative Aware Learning of COntrastive Negatives), a learning-based mini-batch construction strategy that adaptively balances the trade-off between hard and false negatives during VLP. Rather than relying on fixed heuristics, FALCON employs a negative mining scheduler that dynamically selects negative samples of appropriate hardness for each anchor instance during mini-batch construction, guided by a proxy for cross-modal alignment improvement. Experimental results demonstrate that FALCON significantly improves performance across three vision-language learning frameworks (ALBEF, BLIP-2, SigLIP-2) and a broad range of downstream tasks and evaluation settings, underscoring its effectiveness and robustness in mitigating the impact of false negatives.

Streetscapes are an essential component of urban space. Their assessment is presently either limited to morphometric properties of their mass skeleton or requires labor-intensive qualitative evaluations of visually perceived qualities. This paper introduces SAGAI: Streetscape Analysis with Generative Artificial Intelligence, a modular workflow for scoring street-level urban scenes using open-access data and vision-language models. SAGAI integrates OpenStreetMap geometries, Google Street View imagery, and a lightweight version of the LLaVA model to generate structured spatial indicators from images via customizable natural language prompts. The pipeline includes an automated mapping module that aggregates visual scores at both the point and street levels, enabling direct cartographic interpretation. It operates without task-specific training or proprietary software dependencies, supporting scalable and interpretable analysis of urban environments. Two exploratory case studies in Nice and Vienna illustrate SAGAI's capacity to produce geospatial outputs from vision-language inference. The initial results show strong performance for binary urban-rural scene classification, moderate precision in commercial feature detection, and lower estimates, but still informative, of sidewalk width. Fully deployable by any user, SAGAI can be easily adapted to a wide range of urban research themes, such as walkability, safety, or urban design, through prompt modification alone.

Hyperspectral image (HSI) classification (HSIC) requires effective modeling of complex spatial-spectral dependencies under limited labeled data and high dimensionality. While transformer-based models have shown strong capability in capturing long-range contextual information, they often introduce redundant attention patterns, which limits their effectiveness for fine-grained HSI analysis. To address these challenges, this paper proposes MemFormer, a lightweight transformer architecture for HSIC that incorporates a dynamic memory-enhanced attention mechanism. The proposed design augments multi-head self-attention with a compact global memory module that progressively aggregates contextual information across layers, enabling efficient modeling of long-range dependencies while reducing attention redundancy. In addition, a Spatial-Spectral Positional Embedding (SSPE) is used to jointly encode spatial continuity and spectral ordering, providing structurally consistent representations without relying on convolution-based positional encodings. Extensive experiments conducted on three benchmark hyperspectral datasets, including Indian Pines, WHU-Hi-HanChuan, and WHU-Hi-HongHu, demonstrate that MemFormer achieves superior classification performance compared to representative convolutional, hybrid, and transformer-based methods. On the Indian Pines dataset, MemFormer attains an overall accuracy of up to 99.55\%, average accuracy of 99.38\%, and a $κ$ coefficient of 99.49\%, highlighting its effectiveness and efficiency for HSIC.

Reward models (RMs) play a crucial role in Reinforcement Learning from Human Feedback by serving as proxies for human preferences in aligning large language models. However, they suffer from various biases which could lead to reward hacking. In this paper, we identify a model preference bias in RMs, where they systematically assign disproportionately high scores to responses from certain policy models, leading to unfair judgments. To mitigate this bias, we propose a calibration method named CHatbot Arena calibrated Reward Modeling (CHARM) that leverages Elo scores from the Chatbot Arena to construct debiased preference datasets and adjust reward model scoring. We conduct extensive experiments on reward model benchmarks and human preference alignment. Results demonstrate that our calibrated RMs achieve improved evaluation accuracy on RM-Bench and the Chat-Hard domain of RewardBench, exhibit a stronger correlation with human preferences by producing scores more closely aligned with Elo rankings and improve downstream post-training performance. These results demonstrate that CHARM provides a simple, effective, and broadly applicable approach to building more reliable and fair reward models.

Reasoning is a fundamental cognitive process that enables logical inference, problem-solving, and decision-making. With the rapid advancement of large language models (LLMs), reasoning has emerged as a key capability that distinguishes advanced AI systems from conventional models that empower chatbots. In this survey, we categorize existing methods along two orthogonal dimensions: (1) Regimes, which define the stage at which reasoning is achieved (either at inference time or through dedicated training); and (2) Architectures, which determine the components involved in the reasoning process, distinguishing between standalone LLMs and agentic compound systems that incorporate external tools, and multi-agent collaborations. Within each dimension, we analyze two key perspectives: (1) Input level, which focuses on techniques that construct high-quality prompts that the LLM condition on; and (2) Output level, which methods that refine multiple sampled candidates to enhance reasoning quality. This categorization provides a systematic understanding of the evolving landscape of LLM reasoning, highlighting emerging trends such as the shift from inference-scaling to learning-to-reason (e.g., DeepSeek-R1), and the transition to agentic workflows (e.g., OpenAI Deep Research, Manus Agent). Additionally, we cover a broad spectrum of learning algorithms, from supervised fine-tuning to reinforcement learning such as PPO and GRPO, and the training of reasoners and verifiers. We also examine key designs of agentic workflows, from established patterns like generator-evaluator and LLM debate to recent innovations. ...

Model merging has recently emerged as a lightweight alternative to ensembling, combining multiple fine-tuned models into a single set of parameters with no additional training overhead. Yet, existing merging methods fall short of matching the full accuracy of separately fine-tuned endpoints. We present MASS (MoErging through Adaptive Subspace Selection), a new approach that closes this gap by unifying multiple fine-tuned models while retaining near state-of-the-art performance across tasks. Building on the low-rank decomposition of per-task updates, MASS stores only the most salient singular components for each task and merges them into a shared model. At inference time, a non-parametric, data-free router identifies which subspace (or combination thereof) best explains an input's intermediate features and activates the corresponding task-specific block. This procedure is fully training-free and introduces only a two-pass inference overhead plus a ~2 storage factor compared to a single pretrained model, irrespective of the number of tasks. We evaluate MASS on CLIP-based image classification using ViT-B-16, ViT-B-32 and ViT-L-14 for benchmarks of 8, 14 and 20 tasks respectively, establishing a new state-of-the-art. Most notably, MASS recovers up to ~98% of the average accuracy of individual fine-tuned models, making it a practical alternative to ensembling at a fraction of the storage cost.

Existing rule-based explanations for Graph Neural Networks (GNNs) provide global interpretability but often optimize and assess fidelity in an intermediate, uninterpretable concept space, overlooking grounding quality for end users in the final subgraph explanations. This gap yields explanations that may appear faithful yet be unreliable in practice. To this end, we propose LogicXGNN, a post-hoc framework that constructs logical rules over reliable predicates explicitly designed to capture the GNN's message-passing structure, thereby ensuring effective grounding. We further introduce data-grounded fidelity ($\textit{Fid}_{\mathcal{D}}$), a realistic metric that evaluates explanations in their final-graph form, along with complementary utility metrics such as coverage and validity. Across extensive experiments, LogicXGNN improves $\textit{Fid}_{\mathcal{D}}$ by over 20% on average relative to state-of-the-art methods while being 10-100 $\times$ faster. With strong scalability and utility performance, LogicXGNN produces explanations that are faithful to the model's logic and reliably grounded in observable data. Our code is available at https://github.com/allengeng123/LogicXGNN/.

A representation of the cause-effect mechanism is needed to enable artificial intelligence to represent how the world works. Bayesian Networks (BNs) have proven to be an effective and versatile tool for this task. BNs require constructing a structure of dependencies among variables and learning the parameters that govern these relationships. These tasks, referred to as structural learning and parameter learning, are actively investigated by the research community, with several algorithms proposed and no single method having established itself as standard. A wide range of software, tools, and packages have been developed for BNs analysis and made available to academic researchers and industry practitioners. As a consequence of having no one-size-fits-all solution, moving the first practical steps and getting oriented into this field is proving to be challenging to outsiders and beginners. In this paper, we review the most relevant tools and software for BNs structural and parameter learning to date, providing our subjective recommendations directed to an audience of beginners. In addition, we provide an extensive easy-to-consult overview table summarizing all software packages and their main features. By improving the reader understanding of which available software might best suit their needs, we improve accessibility to the field and make it easier for beginners to take their first step into it.

Score-based generative models achieve state-of-the-art sampling performance by denoising a distribution perturbed by Gaussian noise. In this paper, we focus on a single deterministic denoising step, and compare the optimal denoiser for the quadratic loss, we name ''full-denoising'', to the alternative ''half-denoising'' introduced by Hyv{ä}rinen (2025). We show that looking at the performance in terms of distance between distributions tells a more nuanced story, with different assumptions on the data leading to very different conclusions. We prove that half-denoising is better than full-denoising for regular enough densities, while full-denoising is better for singular densities such as mixtures of Dirac measures or densities supported on a low-dimensional subspace. In the latter case, we prove that full-denoising can alleviate the curse of dimensionality under a linear manifold hypothesis.

Transfer-based attacks pose a significant threat to real-world applications by directly targeting victim models with adversarial examples generated on surrogate models. While numerous approaches have been proposed to enhance adversarial transferability, existing works often overlook the intrinsic relationship between adversarial perturbations and input images. In this work, we find that adversarial perturbation often exhibits poor translation invariance for a given clean image and model, which is attributed to local invariance. Through empirical analysis, we demonstrate a positive correlation between the local invariance of adversarial perturbations w.r.t. the input image and their transferability across models. Based on this finding, we propose a general adversarial transferability boosting technique called the Local Invariance Boosting approach (LI-Boost). Extensive experiments on the standard ImageNet dataset demonstrate that LI-Boost can significantly enhance various transfer-based attacks (e.g., gradient-based, input transformation-based, model-related, advanced objective function, ensemble, etc.) on CNNs, ViTs, defense mechanisms, commercial vision API systems, and vision-language models. Our approach provides a promising direction for future research on improving adversarial transferability across models. Our code is available at https://github.com/Trustworthy-AI-Group/TransferAttack.

Modern robotics applications require an inverse kinematics (IK) solver that is fast, robust and consistent, and that provides all possible solutions. Currently, the Franka robot arm is the most widely used manipulator in robotics research. With 7 DOFs, the IK of this robot is not only complex due to its 1-DOF redundancy, but also due to the link offsets at the wrist and elbow. Due to this complexity, none of the Franka IK solvers available in the literature provide satisfactory results when used in real-world applications. Therefore, in this paper we introduce GeoFIK (Geometric Franka IK), an analytical IK solver that allows the use of different joint variables to resolve the redundancy. The approach uses screw theory to describe the entire geometry of the robot, allowing the computation of the Jacobian matrix prior to computation of joint angles. All singularities are identified and handled. As an example of how the geometric elements obtained by the IK can be exploited, a solver with the swivel angle as the free variable is provided. Several experiments are carried out to validate the speed, robustness and reliability of the GeoFIK against two state-of-the-art solvers.

We present Fillerbuster, a unified model that completes unknown regions of a 3D scene with a multi-view latent diffusion transformer. Casual captures are often sparse and miss surrounding content behind objects or above the scene. Existing methods are not suitable for this challenge as they focus on making known pixels look good with sparse-view priors, or on creating missing sides of objects from just one or two photos. In reality, we often have hundreds of input frames and want to complete areas that are missing and unobserved from the input frames. Our solution is to train a generative model that can consume a large context of input frames while generating unknown target views and recovering image poses when camera parameters are unknown. We show results where we complete partial captures on two existing datasets. We also present an uncalibrated scene completion task where our unified model predicts both poses and creates new content. We open-source our framework for integration into popular reconstruction platforms like Nerfstudio or Gsplat. We present a flexible, unified inpainting framework to predict many images and poses together, where all inputs are jointly inpainted, and it could be extended to predict more modalities such as depth.

LLMs' sources of knowledge are data snapshots containing factual information about entities collected at different timestamps and from different media types (e.g. wikis, social media, etc.). Such unstructured knowledge is subject to change due to updates through time from past to present. Equally important are the inconsistencies and inaccuracies occurring in different information sources. Consequently, the model's knowledge about an entity may be perturbed while training over the sequence of snapshots or at inference time, resulting in inconsistent and inaccurate model performance. In this work, we study the appropriateness of Large Language Models (LLMs) as repositories of factual knowledge. We consider twenty-four state-of-the-art LLMs that are either closed-, partially (weights), or fully (weight and training data) open-source. We evaluate their reliability in responding to time-sensitive factual questions in terms of accuracy and consistency when prompts are perturbed. We further evaluate the effectiveness of state-of-the-art methods to improve LLMs' accuracy and consistency. We then propose ENtity-Aware Fine-tuning (ENAF), a soft neurosymbolic approach aimed at providing structured representation of entities during fine-tuning to reduce inconsistencies and improve response stability under prompt variations.

The paper discusses the role of WordNet-based semantic classification in the formalization of constructions, and more specifically in the semantic annotation of schematic fillers, in the Italian Constructicon. We outline how the Italian Constructicon project uses Open Multilingual WordNet topics to represent semantic features and constraints of constructions.

4D millimeter-wave (mmWave) radars are sensors that provide robustness against adverse weather conditions (rain, snow, fog, etc.), and as such they are increasingly used for odometry and SLAM (Simultaneous Location and Mapping). However, the noisy and sparse nature of the returned scan data proves to be a challenging obstacle for existing registration algorithms, especially those originally intended for more accurate sensors such as LiDAR. Following the success of 3D Gaussian Splatting for vision, in this paper we propose a summarized representation for radar scenes based on global simultaneous optimization of 3D Gaussians as opposed to voxel-based approaches, and leveraging its inherent Probability Density Function (PDF) for registration. Moreover, we propose optimizing multiple registration hypotheses for better protection against local optima of the PDF. We evaluate our modeling and registration system against state of the art techniques, finding that our system provides richer models and more accurate registration results. Finally, we evaluate the effectiveness of our system in a real Radar-Inertial Odometry task. Experiments using publicly available 4D radar datasets show that our Gaussian approach is comparable to existing registration algorithms, outperforming them in several sequences. Copyright 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Recent advances in text-to-vision generation excel in visual fidelity but struggle with compositional generalization and semantic alignment. Existing datasets are noisy and weakly compositional, limiting models' understanding of complex scenes, while scalable solutions for dense, high-quality annotations remain a challenge. We introduce Generate Any Scene, a data engine that systematically enumerates scene graphs representing the combinatorial array of possible visual scenes. Generate Any Scene dynamically constructs scene graphs of varying complexity from a structured taxonomy of objects, attributes, and relations. Given a sampled scene graph, Generate Any Scene translates it into a caption for text-to-image or text-to-video generation; it also translates it into a set of visual question answers that allow automatic evaluation and reward modeling of semantic alignment. Using Generate Any Scene, we first design a self-improving framework where models iteratively enhance their performance using generated data. Stable Diffusion v1.5 achieves an average 4% improvement over baselines and surpassing fine-tuning on CC3M. Second, we also design a distillation algorithm to transfer specific strengths from proprietary models to their open-source counterparts. Using fewer than 800 synthetic captions, we fine-tune Stable Diffusion v1.5 and achieve a 10% increase in TIFA score on compositional and hard concept generation. Third, we create a reward model to align model generation with semantic accuracy at a low cost. Using GRPO algorithm, we fine-tune SimpleAR-0.5B-SFT and surpass CLIP-based methods by +5% on DPG-Bench. Finally, we apply these ideas to the downstream task of content moderation where we train models to identify challenging cases by learning from synthetic data.

Open-vocabulary 3D scene understanding is indispensable for embodied agents. Recent works leverage pretrained vision-language models (VLMs) for object segmentation and project them to point clouds to build 3D maps. Despite progress, a point cloud is a set of unordered coordinates that requires substantial storage space and does not directly convey occupancy information or spatial relation, making existing methods inefficient for downstream tasks, e.g., path planning and text-based object retrieval. To address these issues, we propose \textbf{Octree-Graph}, a novel scene representation for open-vocabulary 3D scene understanding. Specifically, a Chronological Group-wise Segment Merging (CGSM) strategy and an Instance Feature Aggregation (IFA) algorithm are first designed to get 3D instances and corresponding semantic features. Subsequently, an adaptive-octree structure is developed that stores semantics and depicts the occupancy of an object adjustably according to its shape. Finally, the Octree-Graph is constructed where each adaptive-octree acts as a graph node, and edges describe the spatial relations among nodes. Extensive experiments on various tasks are conducted on several widely-used datasets, demonstrating the versatility and effectiveness of our method. Code is available \href{https://github.com/yifeisu/OV-Octree-Graph}{here}.

Reconstructing controllable Gaussian splats for articulated objects from monocular video is especially challenging due to its inherently insufficient constraints. Existing methods address this by relying on dense masks and manually defined control signals, limiting their real-world applications. In this paper, we propose an annotation-free method, FreeGaussian, which mathematically disentangles camera egomotion and articulated movements via flow derivatives. By establishing a connection between 2D flows and 3D Gaussian dynamic flow, our method enables optimization and continuity of dynamic Gaussian motions from flow priors without any control signals. Furthermore, we introduce a 3D spherical vector controlling scheme, which represents the state as a 3D Gaussian trajectory, thereby eliminating the need for complex 1D control signal calculations and simplifying controllable Gaussian modeling. Extensive experiments on articulated objects demonstrate the state-of-the-art visual performance and precise, part-aware controllability of our method. Code is available at: https://github.com/Tavish9/freegaussian.

While post-training compression techniques effectively reduce the memory footprint, latency, and power consumption of Large Language Models (LLMs), they often result in noticeable accuracy degradation and remain limited by hardware and kernel constraints that restrict supported compression formats - ultimately reducing flexibility across a wide range of deployment scenarios. In this work, we propose EoRA - a novel $\textbf{fine-tuning-free}$ method that augments compressed LLMs with low-rank matrices, allowing users to rapidly enhance task-specific performance and freely balance the trade-off between accuracy and computational overhead beyond the constraints of compression formats. EoRA consistently outperforms prior fine-tuning-free low rank methods in recovering the accuracy of compressed LLMs, achieving notable accuracy improvements (e.g., $\mathbf{10.84\%}$ on ARC-Challenge, $\mathbf{6.74\%}$ on MathQA, and $\mathbf{11.45\%}$ on GSM8K for LLaMA3-8B compressed to 3-bit). We also introduce an optimized CUDA kernel, accelerating inference by up to 1.4x and reducing memory overhead through quantizing EoRA. Overall, EoRA offers a prompt solution for improving the accuracy of compressed models under varying user requirements, enabling more efficient and flexible deployment of LLMs. Code is available at https://github.com/NVlabs/EoRA.

Equipping humanoid robots with the capability to understand emotional states of human interactants and express emotions appropriately according to situations is essential for affective human-robot interaction. However, enabling current vision-aware multimodal emotion recognition models for affective human-robot interaction in the real-world raises embodiment challenges: addressing the environmental noise issue and meeting real-time requirements. First, in multiparty conversation scenarios, the noises inherited in the visual observation of the robot, which may come from either 1) distracting objects in the scene or 2) inactive speakers appearing in the field of view of the robot, hinder the models from extracting emotional cues from vision inputs. Secondly, realtime response, a desired feature for an interactive system, is also challenging to achieve. To tackle both challenges, we introduce an affective human-robot interaction system called UGotMe designed specifically for multiparty conversations. Two denoising strategies are proposed and incorporated into the system to solve the first issue. Specifically, to filter out distracting objects in the scene, we propose extracting face images of the speakers from the raw images and introduce a customized active face extraction strategy to rule out inactive speakers. As for the second issue, we employ efficient data transmission from the robot to the local server to improve realtime response capability. We deploy UGotMe on a human robot named Ameca to validate its real-time inference capabilities in practical scenarios. Videos demonstrating real-world deployment are available at https://lipzh5.github.io/HumanoidVLE/.

Quality-of-Service (QoS) prediction is a critical task in the service lifecycle, enabling precise and adaptive service recommendations by anticipating performance variations over time in response to evolving network uncertainties and user preferences. However, contemporary QoS prediction methods frequently encounter data sparsity and cold-start issues, which hinder accurate QoS predictions and limit the ability to capture diverse user preferences. Additionally, these methods often assume QoS data reliability, neglecting potential credibility issues such as outliers and the presence of greysheep users and services with atypical invocation patterns. Furthermore, traditional approaches fail to leverage diverse features, including domain-specific knowledge and complex higher-order patterns, essential for accurate QoS predictions. In this paper, we introduce a real-time, trust-aware framework for temporal QoS prediction to address the aforementioned challenges, featuring an end-to-end deep architecture called the Hypergraph Convoluted Transformer Network (HCTN). HCTN combines a hypergraph structure with graph convolution over hyper-edges to effectively address high-sparsity issues by capturing complex, high-order correlations. Complementing this, the transformer network utilizes multi-head attention along with parallel 1D convolutional layers and fully connected dense blocks to capture both fine-grained and coarse-grained dynamic patterns. Additionally, our approach includes a sparsity-resilient solution for detecting greysheep users and services, incorporating their unique characteristics to improve prediction accuracy. Trained with a robust loss function resistant to outliers, HCTN demonstrated state-of-the-art performance on the large-scale WSDREAM-2 datasets for response time and throughput.

Epilepsy represents the most prevalent neurological disease in the world. One-third of people suffering from mesial temporal lobe epilepsy (MTLE) exhibit drug resistance, urging the need to develop new treatments. A key part in anti-seizure medication (ASM) development is the capability of detecting and quantifying epileptic seizures occurring in electroencephalogram (EEG) signals, which is crucial for treatment efficacy evaluation. In this study, we introduced a seizure detection pipeline based on deep learning models applied to raw EEG signals. This pipeline integrates: a new pre-processing technique which segments continuous raw EEG signals without prior distinction between seizure and seizure-free activities; a post-processing algorithm developed to reassemble EEG segments and allow the identification of seizures start/end; and finally, a new evaluation procedure based on a strict seizure events comparison between predicted and real labels. Models training have been performed using a data splitting strategy which addresses the potential for data leakage. We demonstrated the fundamental differences between a seizure classification and a seizure detection task and showed the differences in performance between the two tasks. Finally, we demonstrated the generalization capabilities across species of our best architecture, combining a Convolutional Neural Network and a Transformer encoder. The model was trained on animal EEGs and tested on human EEGs with a F1-score of 93% on a balanced Bonn dataset.

In Standard Chinese, Tone 3 (the dipping tone) becomes Tone 2 (rising tone) when followed by another Tone 3. Previous studies have noted that this sandhi process may be incomplete, in the sense that the assimilated Tone 3 is still distinct from a true Tone 2. While Mandarin Tone 3 sandhi is widely studied using carefully controlled laboratory speech (Xu, 1997) and more formal registers of Beijing Mandarin (Yuan & Y. Chen, 2014), less is known about its realization in spontaneous speech, and about the effect of contextual factors on tonal realization. The present study investigates the pitch contours of two-character words with T2-T3 and T3-T3 tone patterns in spontaneous Taiwan Mandarin conversations. Our analysis makes use of the Generative Additive Mixed Model (GAMM, Wood, 2017) to examine fundamental frequency (F0) contours as a function of normalized time. We consider various factors known to influence pitch contours, including gender, duration, word position, bigram probability, neighboring tones, speaker, and also novel predictors, word and word sense (Chuang et al., 2025). Our analyses revealed that in spontaneous Taiwan Mandarin, T3-T3 words become indistinguishable from T2-T3 words, indicating complete sandhi, once the strong effect of word (or word sense) is taken into account.

Full waveform inversion (FWI) aims to reconstruct subsurface velocity models from observed seismic wavefields and has recently benefited from advances in deep learning (DL). The performance of DL-based FWI critically depends on the diversity of training data, yet existing datasets such as OpenFWI rely on fixed or weakly varying source conditions, limiting their ability to represent realistic seismic scenarios and hindering source generalization. To address this issue, we construct a new source-variable seismic dataset, termed SVFWI, by systematically varying the frequencies and horizontal locations of multiple surface sources. SVFWI is further divided into three subsets that respectively model frequency variations, location variations, and their combined effects, providing a challenging benchmark in data-driven FWI. We further propose SG-DeepONet, a novel DeepONet-based encoder-decoder framework tailored for FWI. The branch network extracts multi-scale time-frequency features from seismic observations, the trunk network explicitly embeds source physical parameters, and an interactive decoding network enables effective nonlinear fusion and high-fidelity velocity reconstruction. Extensive experiments on SVFWI demonstrate that SG-DeepONet achieves superior inversion accuracy and robustness under varying source conditions compared with existing DL-based FWI methods.