Xiao Liang

Chung-Pang Wang, Changwei Chen, Xiao Liang et al.

Autonomous surgical systems must adapt to highly dynamic environments where tissue properties and visual cues evolve rapidly. Central to such adaptability is feedback: the ability to sense, interpret, and respond to changes during execution. While feedback mechanisms have been explored in surgical robotics, ranging from tool and tissue tracking to error detection, existing methods remain limited in handling the topological and perceptual challenges of tissue dissection. In this work, we propose a feedback-enabled framework for autonomous tissue dissection that explicitly reasons about topological changes from endoscopic images after each dissection action. This structured feedback guides subsequent actions, enabling the system to localize dissection progress and adapt policies online. To improve the reliability of such feedback, we introduce visibility metrics that quantify tissue exposure and formulate optimal controller designs that actively manipulate tissue to maximize visibility. Finally, we integrate these feedback mechanisms with both planning-based and learning-based dissection methods, and demonstrate experimentally that they significantly enhance autonomy, reduce errors, and improve robustness in complex surgical scenarios.

Xun Wang, Bingqing Ke, Xuanping Li et al. · deepmind

Video search has become the main routine for users to discover videos relevant to a text query on large short-video sharing platforms. During training a query-video bi-encoder model using online search logs, we identify a modality bias phenomenon that the video encoder almost entirely relies on text matching, neglecting other modalities of the videos such as vision, audio. This modality imbalanceresults from a) modality gap: the relevance between a query and a video text is much easier to learn as the query is also a piece of text, with the same modality as the video text; b) data bias: most training samples can be solved solely by text matching. Here we share our practices to improve the first retrieval stage including our solution for the modality imbalance issue. We propose MBVR (short for Modality Balanced Video Retrieval) with two key components: manually generated modality-shuffled (MS) samples and a dynamic margin (DM) based on visual relevance. They can encourage the video encoder to pay balanced attentions to each modality. Through extensive experiments on a real world dataset, we show empirically that our method is both effective and efficient in solving modality bias problem. We have also deployed our MBVR in a large video platform and observed statistically significant boost over a highly optimized baseline in an A/B test and manual GSB evaluations.

Hengyuan Zhang, Zhihao Zhang, Mingyang Wang et al.

Mechanistic Interpretability (MI) has emerged as a vital approach to demystify the opaque decision-making of Large Language Models (LLMs). However, existing reviews primarily treat MI as an observational science, summarizing analytical insights while lacking a systematic framework for actionable intervention. To bridge this gap, we present a practical survey structured around the pipeline: "Locate, Steer, and Improve." We formally categorize Localizing (diagnosis) and Steering (intervention) methods based on specific Interpretable Objects to establish a rigorous intervention protocol. Furthermore, we demonstrate how this framework enables tangible improvements in Alignment, Capability, and Efficiency, effectively operationalizing MI as an actionable methodology for model optimization. The curated paper list of this work is available at https://github.com/rattlesnakey/Awesome-Actionable-MI-Survey.

Yunxiang Li, Hua-Chieh Shao, Xiao Liang et al.

Recently, the diffusion model has emerged as a superior generative model that can produce high quality and realistic images. However, for medical image translation, the existing diffusion models are deficient in accurately retaining structural information since the structure details of source domain images are lost during the forward diffusion process and cannot be fully recovered through learned reverse diffusion, while the integrity of anatomical structures is extremely important in medical images. For instance, errors in image translation may distort, shift, or even remove structures and tumors, leading to incorrect diagnosis and inadequate treatments. Training and conditioning diffusion models using paired source and target images with matching anatomy can help. However, such paired data are very difficult and costly to obtain, and may also reduce the robustness of the developed model to out-of-distribution testing data. We propose a frequency-guided diffusion model (FGDM) that employs frequency-domain filters to guide the diffusion model for structure-preserving image translation. Based on its design, FGDM allows zero-shot learning, as it can be trained solely on the data from the target domain, and used directly for source-to-target domain translation without any exposure to the source-domain data during training. We evaluated it on three cone-beam CT (CBCT)-to-CT translation tasks for different anatomical sites, and a cross-institutional MR imaging translation task. FGDM outperformed the state-of-the-art methods (GAN-based, VAE-based, and diffusion-based) in metrics of Frechet Inception Distance (FID), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Measure (SSIM), showing its significant advantages in zero-shot medical image translation.

Jiaxin Cheng, Xiao Liang, Xingjian Shi et al.

Layout-to-image generation refers to the task of synthesizing photo-realistic images based on semantic layouts. In this paper, we propose LayoutDiffuse that adapts a foundational diffusion model pretrained on large-scale image or text-image datasets for layout-to-image generation. By adopting a novel neural adaptor based on layout attention and task-aware prompts, our method trains efficiently, generates images with both high perceptual quality and layout alignment, and needs less data. Experiments on three datasets show that our method significantly outperforms other 10 generative models based on GANs, VQ-VAE, and diffusion models.

Jiawen Xie, Pengyu Cheng, Xiao Liang et al.

Although dominant in natural language processing, transformer-based models remain challenged by the task of long-sequence processing, because the computational cost of self-attention operations in transformers swells quadratically with the input sequence length. To alleviate the complexity of long-sequence processing, we propose a simple framework to enable the offthe-shelf pre-trained transformers to process much longer sequences, while the computation and memory costs remain growing linearly with the input sequence lengths. More specifically, our method divides each long-sequence input into a batch of chunks, then aligns the interchunk information during the encoding steps, and finally selects the most representative hidden states from the encoder for the decoding process. To extract inter-chunk semantic information, we align the start and end token embeddings among chunks in each encoding transformer block. To learn an effective hidden selection policy, we design a dual updating scheme inspired by reinforcement learning, which regards the decoders of transformers as environments, and the downstream performance metrics as the rewards to evaluate the hidden selection actions. Our empirical results on real-world long-text summarization and reading comprehension tasks demonstrate effective improvements compared to prior longsequence processing baselines.

Xiao Liang, Howard Morgan, Ti Bai et al.

Cone-beam CT (CBCT)-based online adaptive radiotherapy calls for accurate auto-segmentation to reduce the time cost for physicians to edit contours. However, deep learning (DL)-based direct segmentation of CBCT images is a challenging task, mainly due to the poor image quality and lack of well-labelled large training datasets. Deformable image registration (DIR) is often used to propagate the manual contours on the planning CT (pCT) of the same patient to CBCT. In this work, we undertake solving the problems mentioned above with the assistance of DIR. Our method consists of three main components. First, we use deformed pCT contours derived from multiple DIR methods between pCT and CBCT as pseudo labels for initial training of the DL-based direct segmentation model. Second, we use deformed pCT contours from another DIR algorithm as influencer volumes to define the region of interest for DL-based direct segmentation. Third, the initially trained DL model is further fine-tuned using a smaller set of true labels. We found that DL-based direct segmentation on CBCT trained with pseudo labels and without influencer volumes shows poor performance compared to DIR-based segmentation. However, adding deformed pCT contours as influencer volumes in the direct segmentation network dramatically improves segmentation performance, reaching the accuracy level of DIR-based segmentation. The DL model with influencer volumes can be further improved through fine-tuning using a smaller set of true labels. Experiments showed that 7 out of 19 structures have an at least 0.2 Dice similarity coefficient increase compared to DIR-based segmentation. A DL-based direct CBCT segmentation model can be improved to outperform DIR-based segmentation models by using deformed pCT contours as pseudo labels and influencer volumes for initial training, and by using a smaller set of true labels for model fine tuning.

Sibo Tian, Minghui Zheng, Xiao Liang

Predicting human motion plays a crucial role in ensuring a safe and effective human-robot close collaboration in intelligent remanufacturing systems of the future. Existing works can be categorized into two groups: those focusing on accuracy, predicting a single future motion, and those generating diverse predictions based on observations. The former group fails to address the uncertainty and multi-modal nature of human motion, while the latter group often produces motion sequences that deviate too far from the ground truth or become unrealistic within historical contexts. To tackle these issues, we propose TransFusion, an innovative and practical diffusion-based model for 3D human motion prediction which can generate samples that are more likely to happen while maintaining a certain level of diversity. Our model leverages Transformer as the backbone with long skip connections between shallow and deep layers. Additionally, we employ the discrete cosine transform to model motion sequences in the frequency space, thereby improving performance. In contrast to prior diffusion-based models that utilize extra modules like cross-attention and adaptive layer normalization to condition the prediction on past observed motion, we treat all inputs, including conditions, as tokens to create a more lightweight model compared to existing approaches. Extensive experimental studies are conducted on benchmark datasets to validate the effectiveness of our human motion prediction model.

Harsh Modi, Xiao Liang, Minghui Zheng

This paper introduces a motion planning framework to plan morphology and trajectory for morphing quadrotors under extremely constrained environments. We develop a novel obstacle avoidance cost function for nonlinear model predictive control (MPC) that enables navigation through extremely narrow gaps under limited perception from a 2D LiDAR. Classical artificial potential field-based costs typically have a high cost in narrow passages, artificially blocking the navigable path. In contrast, we propose a smooth exponential obstacle cost that preserves low traversal cost within narrow gaps while maintaining strong collision avoidance behavior. The formulation avoids hard activation thresholds and introduces a cost reduction factor to reduce the cost within narrow passages. Direct use of 2D LiDAR measurements in MPC allows navigation around arbitrarily shaped obstacles. The method is embedded within an acados-based nonlinear MPC framework. Simulation and experimental results demonstrate successful traversal of narrow corridors where typical repulsive cost functions would fail. The approach provides a computationally efficient and practical solution for navigating through tight spaces while maintaining safety from the obstacles. While we are implementing the framework on the morphing quadrotors, the cost function formulation is general-purpose for any mobile robot application, and is not limited to the morphing quadrotors. The implementation code is available at \href{https://github.com/harshjmodi1996/morphocopter_mpc}{Github Repo} and a short video is available at \href{https://zh.engr.tamu.edu/wp-content/uploads/sites/310/2026/03/MPC_MorphoCopter_video.mp4}{Video Link}.

Open-H-Embodiment Consortium, Nigel Nelson, Juo-Tung Chen et al.

Autonomous medical robots hold promise to improve patient outcomes, reduce provider workload, democratize access to care, and enable superhuman precision. However, autonomous medical robotics has been limited by a fundamental data problem: existing medical robotic datasets are small, single-embodiment, and rarely shared openly, restricting the development of foundation models that the field needs to advance. We introduce Open-H-Embodiment, the largest open dataset of medical robotic video with synchronized kinematics to date, spanning more than 49 institutions and multiple robotic platforms including the CMR Versius, Intuitive Surgical's da Vinci, da Vinci Research Kit (dVRK), Rob Surgical BiTrack, Virtual Incision's MIRA, Moon Surgical Maestro, and a variety of custom systems, spanning surgical manipulation, robotic ultrasound, and endoscopy procedures. We demonstrate the research enabled by this dataset through two foundation models. GR00T-H is the first open foundation vision-language-action model for medical robotics, which is the only evaluated model to achieve full end-to-end task completion on a structured suturing benchmark (25% of trials vs. 0% for all others) and achieves 64% average success across a 29-step ex vivo suturing sequence. We also train Cosmos-H-Surgical-Simulator, the first action-conditioned world model to enable multi-embodiment surgical simulation from a single checkpoint, spanning nine robotic platforms and supporting in silico policy evaluation and synthetic data generation for the medical domain. These results suggest that open, large-scale medical robot data collection can serve as critical infrastructure for the research community, enabling advances in robot learning, world modeling, and beyond.

Xiao Liang, Yuxuan An, Di Wang et al.

Medical Vision-Language Models (VLMs) are prone to hallucinations, compromising clinical reliability. While reinforcement learning methods like Group Relative Policy Optimization (GRPO) offer a low-cost alignment solution, their reliance on sparse, outcome-based rewards inadvertently encourages models to "overthink" -- generating verbose, convoluted, and unverifiable Chain-of-Thought reasoning to justify answers. This focus on outcomes obscures factual errors and poses significant safety risks. To address this, we propose CheXPO-v2, a novel alignment framework that shifts from outcome to process supervision. Our core innovation is a Knowledge Graph Consistency Reward mechanism driven by Entity-Relation Matching. By explicitly parsing reasoning steps into structured "Disease, Relation, Anatomy" triplets, we provide fine-grained supervision that penalizes incoherent logic and hallucinations at the atomic level. Integrating this with a hard-example mining strategy, our approach significantly outperforms GRPO and state-of-the-art models on benchmarks like MIMIC-CXR-VQA. Crucially, CheXPO-v2 achieves new state-of-the-art accuracy using only 5k samples, demonstrating exceptional data efficiency while producing clinically sound and verifiable reasoning. The project source code is publicly available at: https://github.com/ecoxial2007/CheX-Phi4MM.

Nikhil U. Shinde, Xiao Liang, Florian Richter et al.

The ability to predict future states is crucial to informed decision-making while interacting with dynamic environments. With cameras providing a prevalent and information-rich sensing modality, the problem of predicting future states from image sequences has garnered a lot of attention. Current state-of-the-art methods typically train large parametric models for their predictions. Though often able to predict with accuracy these models often fail to provide interpretable confidence metrics around their predictions. Additionally these methods are reliant on the availability of large training datasets to converge to useful solutions. In this paper, we focus on the problem of predicting future images of an image sequence with interpretable confidence bounds from very little training data. To approach this problem, we use non-parametric models to take a probabilistic approach to image prediction. We generate probability distributions over sequentially predicted images, and propagate uncertainty through time to generate a confidence metric for our predictions. Gaussian Processes are used for their data efficiency and ability to readily incorporate new training data online. Our methods predictions are evaluated on a smooth fluid simulation environment. We showcase the capabilities of our approach on real world data by predicting pedestrian flows and weather patterns from satellite imagery.

Zhong-Zhi Li, Duzhen Zhang, Ming-Liang Zhang et al.

Achieving human-level intelligence requires refining the transition from the fast, intuitive System 1 to the slower, more deliberate System 2 reasoning. While System 1 excels in quick, heuristic decisions, System 2 relies on logical reasoning for more accurate judgments and reduced biases. Foundational Large Language Models (LLMs) excel at fast decision-making but lack the depth for complex reasoning, as they have not yet fully embraced the step-by-step analysis characteristic of true System 2 thinking. Recently, reasoning LLMs like OpenAI's o1/o3 and DeepSeek's R1 have demonstrated expert-level performance in fields such as mathematics and coding, closely mimicking the deliberate reasoning of System 2 and showcasing human-like cognitive abilities. This survey begins with a brief overview of the progress in foundational LLMs and the early development of System 2 technologies, exploring how their combination has paved the way for reasoning LLMs. Next, we discuss how to construct reasoning LLMs, analyzing their features, the core methods enabling advanced reasoning, and the evolution of various reasoning LLMs. Additionally, we provide an overview of reasoning benchmarks, offering an in-depth comparison of the performance of representative reasoning LLMs. Finally, we explore promising directions for advancing reasoning LLMs and maintain a real-time \href{https://github.com/zzli2022/Awesome-Slow-Reason-System}{GitHub Repository} to track the latest developments. We hope this survey will serve as a valuable resource to inspire innovation and drive progress in this rapidly evolving field.

Fangxun Shu, Yongjie Ye, Yue Liao et al.

We introduce SAIL-RL, a reinforcement learning (RL) post-training framework that enhances the reasoning capabilities of multimodal large language models (MLLMs) by teaching them when and how to think. Existing approaches are limited by outcome-only supervision, which rewards correct answers without ensuring sound reasoning, and by uniform thinking strategies, which often lead to overthinking on simple tasks and underthinking on complex ones. SAIL-RL addresses these challenges with a dual reward system: the Thinking Reward, which evaluates reasoning quality through factual grounding, logical coherence, and answer consistency, and the Judging Reward, which adaptively determines whether deep reasoning or direct answering is appropriate. Experiments on the state-of-the-art SAIL-VL2 show that SAIL-RL improves reasoning and multimodal understanding benchmarks at both 4B and 8B scales, achieving competitive performance against commercial closed-source models such as GPT-4o, and substantially reduces hallucinations, establishing it as a principled framework for building more reliable and adaptive MLLMs. The code will be available at https://github.com/BytedanceDouyinContent/SAIL-RL.

Yusuke Takimoto, Hiroyuki Sato, Hikari Takehara et al.

Differentiable rendering (DR) enables various computer graphics and computer vision applications through gradient-based optimization with derivatives of the rendering equation. Most rasterization-based approaches are built on general-purpose automatic differentiation (AD) libraries and DR-specific modules handcrafted using CUDA. Such a system design mixes DR algorithm implementation and algorithm building blocks, resulting in hardware dependency and limited performance. In this paper, we present a practical hardware-agnostic differentiable renderer called Dressi, which is based on a new full AD design. The DR algorithms of Dressi are fully written in our Vulkan-based AD for DR, Dressi-AD, which supports all primitive operations for DR. Dressi-AD and our inverse UV technique inside it bring hardware independence and acceleration by graphics hardware. Stage packing, our runtime optimization technique, can adapt hardware constraints and efficiently execute complex computational graphs of DR with reactive cache considering the render pass hierarchy of Vulkan. HardSoftRas, our novel rendering process, is designed for inverse rendering with a graphics pipeline. Under the limited functionalities of the graphics pipeline, HardSoftRas can propagate the gradients of pixels from the screen space to far-range triangle attributes. Our experiments and applications demonstrate that Dressi establishes hardware independence, high-quality and robust optimization with fast speed, and photorealistic rendering.

Tao Shi, Xiao Liang, Yaoyuan Liang et al.

Emotion recognition in conversations (ERC) is a rapidly evolving task within the natural language processing community, which aims to detect the emotions expressed by speakers during a conversation. Recently, a growing number of ERC methods have focused on leveraging supervised contrastive learning (SCL) to enhance the robustness and generalizability of learned features. However, current SCL-based approaches in ERC are impeded by the constraint of large batch sizes and the lack of compatibility with most existing ERC models. To address these challenges, we propose an efficient and model-agnostic SCL framework named Supervised Sample-Label Contrastive Learning with Soft-HGR Maximal Correlation (SSLCL), which eliminates the need for a large batch size and can be seamlessly integrated with existing ERC models without introducing any model-specific assumptions. Specifically, we introduce a novel perspective on utilizing label representations by projecting discrete labels into dense embeddings through a shallow multilayer perceptron, and formulate the training objective to maximize the similarity between sample features and their corresponding ground-truth label embeddings, while minimizing the similarity between sample features and label embeddings of disparate classes. Moreover, we innovatively adopt the Soft-HGR maximal correlation as a measure of similarity between sample features and label embeddings, leading to significant performance improvements over conventional similarity measures. Additionally, multimodal cues of utterances are effectively leveraged by SSLCL as data augmentations to boost model performances. Extensive experiments on two ERC benchmark datasets, IEMOCAP and MELD, demonstrate the compatibility and superiority of our proposed SSLCL framework compared to existing state-of-the-art SCL methods. Our code is available at \url{https://github.com/TaoShi1998/SSLCL}.

Liu Liu, Yukai Lin, Xiao Liang et al.

This technical report introduces CyberLoc, an image-based visual localization pipeline for robust and accurate long-term pose estimation under challenging conditions. The proposed method comprises four modules connected in a sequence. First, a mapping module is applied to build accurate 3D maps of the scene, one map for each reference sequence if there exist multiple reference sequences under different conditions. Second, a single-image-based localization pipeline (retrieval--matching--PnP) is performed to estimate 6-DoF camera poses for each query image, one for each 3D map. Third, a consensus set maximization module is proposed to filter out outlier 6-DoF camera poses, and outputs one 6-DoF camera pose for a query. Finally, a robust pose refinement module is proposed to optimize 6-DoF query poses, taking candidate global 6-DoF camera poses and their corresponding global 2D-3D matches, sparse 2D-2D feature matches between consecutive query images and SLAM poses of the query sequence as input. Experiments on the 4seasons dataset show that our method achieves high accuracy and robustness. In particular, our approach wins the localization challenge of ECCV 2022 workshop on Map-based Localization for Autonomous Driving (MLAD-ECCV2022).

Anjali Balagopal, Michael Dohopolski, Young Suk Kwon et al.

Background and purpose: Radiation-induced erectile dysfunction (RiED) is commonly seen in prostate cancer patients. Clinical trials have been developed in multiple institutions to investigate whether dose-sparing to the internal-pudendal-arteries (IPA) will improve retention of sexual potency. The IPA is usually not considered a conventional organ-at-risk (OAR) due to segmentation difficulty. In this work, we propose a deep learning (DL)-based auto-segmentation model for the IPA that utilizes CT and MRI or CT alone as the input image modality to accommodate variation in clinical practice. Materials and methods: 86 patients with CT and MRI images and noisy IPA labels were recruited in this study. We split the data into 42/14/30 for model training, testing, and a clinical observer study, respectively. There were three major innovations in this model: 1) we designed an architecture with squeeze-and-excite blocks and modality attention for effective feature extraction and production of accurate segmentation, 2) a novel loss function was used for training the model effectively with noisy labels, and 3) modality dropout strategy was used for making the model capable of segmentation in the absence of MRI. Results: The DSC, ASD, and HD95 values for the test dataset were 62.2%, 2.54mm, and 7mm, respectively. AI segmented contours were dosimetrically equivalent to the expert physician's contours. The observer study showed that expert physicians' scored AI contours (mean=3.7) higher than inexperienced physicians' contours (mean=3.1). When inexperienced physicians started with AI contours, the score improved to 3.7. Conclusion: The proposed model achieved good quality IPA contours to improve uniformity of segmentation and to facilitate introduction of standardized IPA segmentation into clinical trials and practice.

Ti Bai, Muhan Lin, Xiao Liang et al.

Medical image registration is a fundamental and vital task which will affect the efficacy of many downstream clinical tasks. Deep learning (DL)-based deformable image registration (DIR) methods have been investigated, showing state-of-the-art performance. A test time optimization (TTO) technique was proposed to further improve the DL models' performance. Despite the substantial accuracy improvement with this TTO technique, there still remained some regions that exhibited large registration errors even after many TTO iterations. To mitigate this challenge, we firstly identified the reason why the TTO technique was slow, or even failed, to improve those regions' registration results. We then proposed a two-levels TTO technique, i.e., image-specific optimization (ISO) and region-specific optimization (RSO), where the region can be interactively indicated by the clinician during the registration result reviewing process. For both efficiency and accuracy, we further envisioned a three-step DL-based image registration workflow. Experimental results showed that our proposed method outperformed the conventional method qualitatively and quantitatively.

Ben Chen, Siyuan Wang, Yufei Ma et al.

Generative Retrieval (GR) has emerged as a promising paradigm for modern search systems. Compared to multi-stage cascaded architecture, it offers advantages such as end-to-end joint optimization and high computational efficiency. OneSearch, as a representative industrial-scale deployed generative search framework, has brought significant commercial and operational benefits. However, its inadequate understanding of complex queries, inefficient exploitation of latent user intents, and overfitting to narrow historical preferences have limited its further performance improvement. To address these challenges, we propose \textbf{OneSearch-V2}, a latent reasoning enhanced self-distillation generative search framework. It contains three key innovations: (1) a thought-augmented complex query understanding module, which enables deep query understanding and overcomes the shallow semantic matching limitations of direct inference; (2) a reasoning-internalized self-distillation training pipeline, which uncovers users' potential yet precise e-commerce intentions beyond log-fitting through implicit in-context learning; (3) a behavior preference alignment optimization system, which mitigates reward hacking arising from the single conversion metric, and addresses personal preference via direct user feedback. Extensive offline evaluations demonstrate OneSearch-V2's strong query recognition and user profiling capabilities. Online A/B tests further validate its business effectiveness, yielding +3.98\% item CTR, +3.05\% buyer conversion rate, and +2.11\% order volume. Manual evaluation further confirms gains in search experience quality, with +1.65\% in page good rate and +1.37\% in query-item relevance. More importantly, OneSearch-V2 effectively mitigates common search system issues such as information bubbles and long-tail sparsity, without incurring additional inference costs or serving latency.

Kareem Eltouny, Seyedomid Sajedi, Xiao Liang

Visual inspection is predominantly used to evaluate the state of civil structures, but recent developments in unmanned aerial vehicles (UAVs) and artificial intelligence have increased the speed, safety, and reliability of the inspection process. In this study, we develop a semantic segmentation network based on vision transformers and Laplacian pyramids scaling networks for efficiently parsing high-resolution visual inspection images. The massive amounts of collected high-resolution images during inspections can slow down the investigation efforts. And while there have been extensive studies dedicated to the use of deep learning models for damage segmentation, processing high-resolution visual data can pose major computational difficulties. Traditionally, images are either uniformly downsampled or partitioned to cope with computational demands. However, the input is at risk of losing local fine details, such as thin cracks, or global contextual information. Inspired by super-resolution architectures, our vision transformer model learns to resize high-resolution images and masks to retain both the valuable local features and the global semantics without sacrificing computational efficiency. The proposed framework has been evaluated through comprehensive experiments on a dataset of bridge inspection report images using multiple metrics for pixel-wise materials detection.

Hengyuan Zhang, Xinrong Chen, Zunhai Su et al.

Layer-wise mixed-precision quantization (LMPQ) enables effective compression under extreme low-bit settings by allocating higher precision to sensitive layers. However, existing methods typically treat all intra-layer weight modules uniformly and rely on a single numerical property when estimating sensitivity, overlooking their distinct operational roles and structural characteristics. To address this, we propose NSDS, a novel calibration-free LMPQ framework driven by Numerical and Structural Dual-Sensitivity. Specifically, it first mechanistically decomposes each layer into distinct operational roles and quantifies their sensitivity from both numerical and structural perspectives. These dual-aspect scores are then aggregated into a unified layer-wise metric through a robust aggregation scheme based on MAD-Sigmoid and Soft-OR to guide bit allocation. Extensive experiments demonstrate that NSDS consistently achieves superior performance compared to various baselines across diverse models and downstream tasks, without relying on any calibration data.

Hongyuan Dong, Zijian Kang, Weijie Yin et al.

In this paper, we introduce SAIL-VL (ScAlable Vision Language Model TraIning via High QuaLity Data Curation), an open-source vision language model (VLM) series achieving state-of-the-art (SOTA) performance in 2B and 8B parameters. The following three key improvements contribute to SAIL-VL's leading performance: (1) Scalable high-quality visual understanding data construction: We implement a data construction pipeline to enable hundred-million-scale high-quality recaption data annotation. The resulted dataset SAIL-Caption is validated to be of the highest data quality compared with opensource datasets. (2) Scalable Pretraining with High-Quality Visual Understanding Data: We scale SAIL-VL's pretraining budget up to 655B tokens and show that even a 2B VLM benefits from scaled up training data sizes, exhibiting logarithmic data size scaling laws in benchmark performance. (3) Scalable SFT via data quantity and complexity scaling: We curate a high-quality SFT dataset collection with leading data quantity scaling effectiveness and demonstrate that training with progressively higher-complexity data surpasses baseline one-stage training by a large margin. SAIL-VL series models achieve the highest average score in 18 widely used VLM benchmarks in our evaluation, with the 2B model takes the top position over VLMs of comparable sizes on OpenCompass 2024 (https://rank.opencompass.org.cn/leaderboard-multimodal), demonstrating robust visual comprehension abilities. SAIL-VL series models are released at HuggingFace (https://huggingface.co/BytedanceDouyinContent).

Chang Liu, Sibo Tian, Xiao Liang et al.

Disassembly automation has long been pursued to address the growing demand for efficient and proper recovery of valuable components from the end-of-life (EoL) electronic products. Existing approaches have demonstrated promising and regimented performance by decomposing the disassembly process into different subtasks. However, each subtask typically requires extensive data preparation, model training, and system management. Moreover, these approaches are often task- and component-specific, making them poorly suited to handle the variability and uncertainty of EoL products and limiting their generalization capabilities. All these factors restrict the practical deployment of current robotic disassembly systems and leave them highly reliant on human labor. With the recent development of foundation models in robotics, vision-language-action (VLA) models have shown impressive performance on standard robotic manipulation tasks, but their applicability to complex, contact-rich, and long-horizon industrial practices like disassembly, which requires sequential and precise manipulation, remains limited. To address this challenge, we propose SELF-VLA, an agentic VLA framework that integrates explicit disassembly skills. Experimental studies demonstrate that our framework significantly outperforms current state-of-the-art end-to-end VLA models on two contact-rich disassembly tasks. The video illustration can be found via https://zh.engr.tamu.edu/wp-content/uploads/sites/310/2026/03/IROS-VLA-Video.mp4.

Davood Soleymanzadeh, Ivan Lopez-Sanchez, Hao Su et al.

State-of-the-art generalist manipulation policies have enabled the deployment of robotic manipulators in unstructured human environments. However, these frameworks struggle in cluttered environments primarily because they utilize auxiliary modules for low-level motion planning and control. Motion planning remains challenging due to the high dimensionality of the robot's configuration space and the presence of workspace obstacles. Neural motion planners have enhanced motion planning efficiency by offering fast inference and effectively handling the inherent multi-modality of the motion planning problem. Despite such benefits, current neural motion planners often struggle to generalize to unseen, out-of-distribution planning settings. This paper reviews and analyzes the state-of-the-art neural motion planners, highlighting both their benefits and limitations. It also outlines a path toward establishing generalist neural motion planners capable of handling domain-specific challenges. For a list of the reviewed papers, please refer to https://davoodsz.github.io/planning-manip-survey.github.io/.

Bochao Liu, Zhipeng Qian, Yang Zhao et al.

Operating and maintaining (O&M) large-scale online engine systems (search, recommendation, advertising) demands substantial human effort for release monitoring, alert response, and root cause analysis. While LLM-based agents are a natural fit for these tasks, the deployment bottleneck is not reasoning capability but orchestration: selecting, for each operational event, the relevant data (metrics, logs, change events) and the applicable operational knowledge (handbook rules and practitioner experience). Feeding all signals indiscriminately causes dilution and hallucination, while manually curating the event-to-(data, knowledge) mapping is intractable under dozens of daily releases. We present Bian Que, an agentic framework with three contributions: (i) a \emph{unified operational paradigm} abstracting day-to-day O&M into three canonical patterns: release interception, proactive inspection, and alert root cause analysis; (ii) \emph{Flexible Skill Arrangement}, where each Skill specifies which data and knowledge to retrieve for a given business-module context and can be automatically generated and updated by LLMs or iteratively refined through natural-language instructions from on-call engineers; (iii) a \emph{unified self-evolving mechanism} in which one correction signal drives two parallel pathways, case-memory-to-knowledge distillation and targeted Skill refinement. Deployed on the e-commerce search engine of KuaiShou, the major short-video platform in China, Bian Que reduces alert volume by 75%, achieves 80% root-cause analysis accuracy, and cuts mean time to resolution by over 50%. Our framework achieves 99.0% pass rate on offline evaluations. Our code is available at https://github.com/benchen4395/BianQue_Assistant.

Sibo Tian, Minghui Zheng, Xiao Liang

Human motion prediction is a cornerstone of human-robot collaboration (HRC), as robots need to infer the future movements of human workers based on past motion cues to proactively plan their motion, ensuring safety in close collaboration scenarios. The diffusion model has demonstrated remarkable performance in predicting high-quality motion samples with reasonable diversity, but suffers from a slow generative process which necessitates multiple model evaluations, hindering real-world applications. To enable real-time prediction, in this work, we propose training a one-step multi-layer perceptron-based (MLP-based) diffusion model for motion prediction using knowledge distillation and Bayesian optimization. Our method contains two steps. First, we distill a pretrained diffusion-based motion predictor, TransFusion, directly into a one-step diffusion model with the same denoiser architecture. Then, to further reduce the inference time, we remove the computationally expensive components from the original denoiser and use knowledge distillation once again to distill the obtained one-step diffusion model into an even smaller model based solely on MLPs. Bayesian optimization is used to tune the hyperparameters for training the smaller diffusion model. Extensive experimental studies are conducted on benchmark datasets, and our model can significantly improve the inference speed, achieving real-time prediction without noticeable degradation in performance.

Kareem Eltouny, Seyedomid Sajedi, Xiao Liang

Visual inspection is the predominant technique for evaluating the condition of civil infrastructure. The recent advances in unmanned aerial vehicles (UAVs) and artificial intelligence have made the visual inspections faster, safer, and more reliable. Camera-equipped UAVs are becoming the new standard in the industry by collecting massive amounts of visual data for human inspectors. Meanwhile, there has been significant research on autonomous visual inspections using deep learning algorithms, including semantic segmentation. While UAVs can capture high-resolution images of buildings' façades, high-resolution segmentation is extremely challenging due to the high computational memory demands. Typically, images are uniformly downsized at the price of losing fine local details. Contrarily, breaking the images into multiple smaller patches can cause a loss of global contextual in-formation. We propose a hybrid strategy that can adapt to different inspections tasks by managing the global and local semantics trade-off. The framework comprises a compound, high-resolution deep learning architecture equipped with an attention-based segmentation model and learnable downsampler-upsampler modules designed for optimal efficiency and in-formation retention. The framework also utilizes vision transformers on a grid of image crops aiming for high precision learning without downsizing. An augmented inference technique is used to boost the performance and re-duce the possible loss of context due to grid cropping. Comprehensive experiments have been performed on 3D physics-based graphics models synthetic environments in the Quake City dataset. The proposed framework is evaluated using several metrics on three segmentation tasks: component type, component damage state, and global damage (crack, rebar, spalling).

Jiurun Song, Xiao Liang, Minghui Zheng

In human-robot collaboration (HRC), robots must adapt online to dynamic task constraints and evolving human intent. While physical corrections provide a natural, low-latency channel for operators to convey motion-level adjustments, extracting task-level semantic intent from such brief interactions remains challenging. Existing foundation-model-based approaches primarily rely on vision and language inputs and lack mechanisms to interpret physical feedback. Meanwhile, traditional physical human-robot interaction (pHRI) methods leverage physical corrections for trajectory guidance but struggle to infer task-level semantics. To bridge this gap, we propose TATIC, a unified framework that utilizes torque-based contact force estimation and a task-aware Temporal Convolutional Network (TCN) to jointly infer discrete task-level intent and estimate continuous motion-level parameters from brief physical corrections. Task-aligned feature canonicalization ensures robust generalization across diverse layouts, while an intent-driven adaptation scheme translates inferred human intent into robot motion adaptations. Experiments achieve a 0.904 Macro-F1 score in intent recognition and demonstrate successful hardware validation in collaborative disassembly (see experimental video at https://youtu.be/xF8A52qwEc8).

Yichang Feng, Xiao Liang, Minghui Zheng

This paper introduces a new algorithm for trajectory optimization, Decoupled Reduced-space and Adaptive Feasibility-repair Trajectory Optimization (DRAFTO). It first constructs a constrained objective that accounts for smoothness, safety, joint limits, and task requirements. Then, it optimizes the coefficients, which are the coordinates of a set of basis functions for trajectory parameterization. To reduce the number of repeated constrained optimizations while handling joint-limit feasibility, the optimization is decoupled into a reduced-space Gauss-Newton (GN) descent for the main iterations and constrained quadratic programming for initialization and terminal feasibility repair. The two-phase acceptance rule with a non-monotone policy is applied to the GN model, which uses a hinge-squared penalty for inequality constraints, to ensure globalizability. The results of our benchmark tests against optimization-based planners, such as CHOMP, TrajOpt, GPMP2, and FACTO, and sampling-based planners, such as RRT-Connect, RRT*, and PRM, validate the high efficiency and reliability across diverse scenarios and tasks. The experiment involving grabbing an object from a drawer further demonstrates the potential for implementation in complex manipulation tasks. The supplemental video is available at https://youtu.be/XisFI37YyTQ.

Xiao Liang, Zhong-Zhi Li, Zhenghao Lin et al.

Large language models (LLMs) have demonstrated strong reasoning capabilities through step-by-step chain-of-thought (CoT) reasoning. Nevertheless, at the limits of model capability, CoT often proves insufficient, and its strictly sequential nature constrains test-time scalability. A potential alternative is divide-and-conquer (DAC) reasoning, which decomposes a complex problem into subproblems to facilitate more effective exploration of the solution. Although promising, our analysis reveals a fundamental misalignment between general-purpose post-training and DAC-style inference, which limits the model's capacity to fully leverage this potential. To bridge this gap and fully unlock LLMs' reasoning capabilities on the most challenging tasks, we propose an end-to-end reinforcement learning (RL) framework to enhance their DAC-style reasoning capacity. At each step, the policy decomposes a problem into a group of subproblems, solves them sequentially, and addresses the original one conditioned on the subproblem solutions, with both decomposition and solution integrated into RL training. Under comparable training, our DAC-style framework endows the model with a higher performance ceiling and stronger test-time scalability, surpassing CoT by 8.6% in Pass@1 and 6.3% in Pass@32 on competition-level benchmarks.

Zhong-Zhi Li, Xiao Liang, Zihao Tang et al.

Large Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate redundant reasoning processes while preserving the model's reasoning capability. We validate our approach across models on DeepSeek-R1-Distill-7B and DeepSeek-R1-Distill-14B and on a diverse set of benchmarks with varying difficulty levels. Our method significantly reduces the number of output tokens by nearly 40% while maintaining the accuracy of the reasoning. Our code and data will be available soon.

Weijie Yin, Dingkang Yang, Hongyuan Dong et al.

Vision Transformers (ViTs) are essential as foundation backbones in establishing the visual comprehension capabilities of Multimodal Large Language Models (MLLMs). Although most ViTs achieve impressive performance through image-text pair-based contrastive learning or self-supervised mechanisms, they struggle to engage in connector-based co-training directly with LLMs due to potential parameter initialization conflicts and modality semantic gaps. To address the above challenges, this paper proposes SAILViT, a gradual feature learning-enhanced ViT for facilitating MLLMs to break through performance bottlenecks in complex multimodal interactions. SAILViT achieves coarse-to-fine-grained feature alignment and world knowledge infusion with gradual feature refinement, which better serves target training demands. We perform thorough empirical analyses to confirm the powerful robustness and generalizability of SAILViT across different dimensions, including parameter sizes, model architectures, training strategies, and data scales. Equipped with SAILViT, existing MLLMs show significant and consistent performance improvements on the OpenCompass benchmark across extensive downstream tasks. SAILViT series models are released at https://huggingface.co/BytedanceDouyinContent.

Xiao Liang, Chenxi Liu, Zhi Ma et al.

Medical Vision-Language Models (MedVLMs) show immense promise in clinical applicability. However, their reliability is hindered by hallucinations, where models often fail to derive answers from visual evidence, instead relying on learned textual priors. Existing mitigation strategies for MedVLMs have distinct limitations: training-based methods rely on costly expert annotations, limiting scalability, while training-free interventions like contrastive decoding, though data-efficient, apply a global, untargeted correction whose effects in complex real-world clinical settings can be unreliable. To address these challenges, we introduce Anatomical Region-Guided Contrastive Decoding (ARCD), a plug-and-play strategy that mitigates hallucinations by providing targeted, region-specific guidance. Our module leverages an anatomical mask to direct a three-tiered contrastive decoding process. By dynamically re-weighting at the token, attention, and logits levels, it verifiably steers the model's focus onto specified regions, reinforcing anatomical understanding and suppressing factually incorrect outputs. Extensive experiments across diverse datasets, including chest X-ray, CT, brain MRI, and ocular ultrasound, demonstrate our method's effectiveness in improving regional understanding, reducing hallucinations, and enhancing overall diagnostic accuracy.

Yi Su, Xinchen Luo, Hongtao Cheng et al.

Quantized inference has demonstrated substantial system-level benefits in large language models while preserving model quality. In contrast, reliably applying low-precision quantization to recommender systems remains challenging in industrial settings. This difficulty arises from differences in training paradigms, architectural patterns, and computational characteristics, which lead to distinct numerical behaviors in weights and activations. Traditional recommender models often exhibit high-magnitude and high-variance weights and activations, making them more sensitive to quantization-induced perturbations. In addition, recommendation workloads frequently suffer from limited hardware utilization, limiting the practical gains of low-precision computation. In this work, we revisit low-precision inference in the context of generative recommendation. Through empirical distribution analysis, we show that the weight and activation statistics of OneRec-V2 are significantly more controlled and closer to those of large language models than traditional recommendation models. Moreover, OneRec-V2 exhibits a more compute-intensive inference pattern with substantially higher hardware utilization, enabling more end-to-end throughput gains with low-precision computation. Leveraging this property, we develop a FP8 post training quantization framework and integrate it into an optimized inference infrastructure. The proposed joint optimization achieves a 49\% reduction in end-to-end inference latency and a 92\% increase in throughput. Extensive online A/B testing further confirms that FP8 inference introduces no degradation in core metrics. These results suggest that as recommender systems evolve toward the paradigms of large language models, algorithm-level and system-level optimization techniques established in the LLM domain can be effectively adapted to large-scale recommendation workloads.

Weijie Yin, Yongjie Ye, Fangxun Shu et al.

We introduce SAIL-VL2, an open-suite vision-language foundation model (LVM) for comprehensive multimodal understanding and reasoning. As the successor to SAIL-VL, SAIL-VL2 achieves state-of-the-art performance at the 2B and 8B parameter scales across diverse image and video benchmarks, demonstrating strong capabilities from fine-grained perception to complex reasoning. Its effectiveness is driven by three core innovations. First, a large-scale data curation pipeline with scoring and filtering strategies enhances both quality and distribution across captioning, OCR, QA, and video data, improving training efficiency. Second, a progressive training framework begins with a powerful pre-trained vision encoder (SAIL-ViT), advances through multimodal pre-training, and culminates in a thinking-fusion SFT-RL hybrid paradigm that systematically strengthens model capabilities. Third, architectural advances extend beyond dense LLMs to efficient sparse Mixture-of-Experts (MoE) designs. With these contributions, SAIL-VL2 demonstrates competitive performance across 106 datasets and achieves state-of-the-art results on challenging reasoning benchmarks such as MMMU and MathVista. Furthermore, on the OpenCompass leaderboard, SAIL-VL2-2B ranks first among officially released open-source models under the 4B parameter scale, while serving as an efficient and extensible foundation for the open-source multimodal community.

Hengyuan Zhang, Xinrong Chen, Yingmin Qiu et al.

Parameter-efficient fine-tuning (PEFT) methods, particularly Low-Rank Adaptation (LoRA), offer an efficient way to adapt large language models with reduced computational costs. However, their performance is limited by the small number of trainable parameters. Recent work combines LoRA with the Mixture-of-Experts (MoE), i.e., LoRA-MoE, to enhance capacity, but two limitations remain in hindering the full exploitation of its potential: 1) the influence of downstream tasks when assigning expert numbers, and 2) the uniform rank assignment across all LoRA experts, which restricts representational diversity. To mitigate these gaps, we propose GuiLoMo, a fine-grained layer-wise expert numbers and ranks allocation strategy with GuidedSelection Vectors (GSVs). GSVs are learned via a prior bilevel optimization process to capture both model- and task-specific needs, and are then used to allocate optimal expert numbers and ranks. Experiments on three backbone models across diverse benchmarks show that GuiLoMo consistently achieves superior or comparable performance to all baselines. Further analysis offers key insights into how expert numbers and ranks vary across layers and tasks, highlighting the benefits of adaptive expert configuration. Our code is available at https://github.com/Liar406/Gui-LoMo.git.

Jiuzhou Lei, Chang Liu, Yu She et al.

Vision-based policies have achieved a good performance in robotic manipulation due to the accessibility and richness of visual observations. However, purely visual sensing becomes insufficient in contact-rich and force-sensitive tasks where force/torque (F/T) signals provide critical information about contact dynamics, alignment, and interaction quality. Although various strategies have been proposed to integrate vision and F/T signals, including auxiliary prediction objectives, mixture-of-experts architectures, and contact-aware gating mechanisms, a comparison of these approaches remains lacking. In this work, we provide a comparison study of different F/T-vision integration strategies within diffusion-based manipulation policies. In addition, we propose an adaptive integration strategy that ignores F/T signals during non-contact phases while adaptively leveraging both vision and torque information during contact. Experimental results demonstrate that our method outperforms the strongest baseline by 14% in success rate, highlighting the importance of contact-aware multimodal fusion for robotic manipulation.

Yi Luo, Haochen Zhao, Xiao Liang et al.

Drug synergy prediction is a critical task in the development of effective combination therapies for complex diseases, including cancer. Although existing methods have shown promising results, they often operate as black-box predictors that rely predominantly on statistical correlations between drug characteristics and results. To address this limitation, we propose CausalDDS, a novel framework that disentangles drug molecules into causal and spurious substructures, utilizing the causal substructure representations for predicting drug synergy. By focusing on causal sub-structures, CausalDDS effectively mitigates the impact of redundant features introduced by spurious substructures, enhancing the accuracy and interpretability of the model. In addition, CausalDDS employs a conditional intervention mechanism, where interventions are conditioned on paired molecular structures, and introduces a novel optimization objective guided by the principles of sufficiency and independence. Extensive experiments demonstrate that our method outperforms baseline models, particularly in cold start and out-of-distribution settings. Besides, CausalDDS effectively identifies key substructures underlying drug synergy, providing clear insights into how drug combinations work at the molecular level. These results underscore the potential of CausalDDS as a practical tool for predicting drug synergy and facilitating drug discovery.

Yang Wang, Hai Yu, Wei He et al.

This paper investigates the control problem of dual-arm unmanned aerial manipulator systems (DAUAMs). Strong coupling between the dual-arm and the multirotor platform, together with unmodeled dynamics and external disturbances, poses significant challenges to stable and accurate operation. An adaptive event-triggered control scheme with neural network-based approximation is proposed to address these issues while explicitly considering communication constraints. First, a dynamic model of the DAUAM system is derived, and a command-filter-based backstepping framework with error compensation is constructed. Then, a neural network is employed to approximate external frictions, and an event-triggered mechanism is designed to reduce the transmission frequency of control updates, thereby alleviating communication and energy burdens. Lyapunov-based analysis shows that all closed-loop signals remain bounded and that the tracking error converges to a neighborhood of the desired trajectory within a fixed time. Finally, experiments on a self-built DAUAM platform demonstrate that the proposed approach achieves accurate trajectory tracking.

Eric Hanchen Jiang, Levina Li, Rui Sun et al.

Large Language Models (LLMs) have shown remarkable performance in completing various tasks. However, solving complex problems often requires the coordination of multiple agents, raising a fundamental question: how to effectively select and interconnect these agents. In this paper, we propose \textbf{Agent Q-Mix}, a reinforcement learning framework that reformulates topology selection as a cooperative Multi-Agent Reinforcement Learning (MARL) problem. Our method learns decentralized communication decisions using QMIX value factorization, where each agent selects from a set of communication actions that jointly induce a round-wise communication graph. At its core, Agent Q-Mix combines a topology-aware GNN encoder, GRU memory, and per-agent Q-heads under a Centralized Training with Decentralized Execution (CTDE) paradigm. The framework optimizes a reward function that balances task accuracy with token cost. Across seven core benchmarks in coding, reasoning, and mathematics, Agent Q-Mix achieves the highest average accuracy compared to existing methods while demonstrating superior token efficiency and robustness against agent failure. Notably, on the challenging Humanity's Last Exam (HLE) using Gemini-3.1-Flash-Lite as a backbone, Agent Q-Mix achieves 20.8\% accuracy, outperforming Microsoft Agent Framework (19.2\%) and LangGraph (19.2\%), followed by AutoGen and Lobster by OpenClaw. These results underscore the effectiveness of learned, decentralized topology optimization in pushing the boundaries of multi-agent reasoning.

Liu Qifeng, Zhao Dawei, Dong Yabo et al.

3D object detection from point clouds plays a critical role in autonomous driving. Currently, the primary methods for point cloud processing are voxel-based and pillarbased approaches. Voxel-based methods offer high accuracy through fine-grained spatial segmentation but suffer from slower inference speeds. Pillar-based methods enhance inference speed but still fall short of voxel-based methods in accuracy. To address these issues, we propose a novel point cloud processing method, PointSlice, which slices point clouds along the horizontal plane and includes a dedicated detection network. The main contributions of PointSlice are: (1) A new point cloud processing technique that converts 3D point clouds into multiple sets of 2D (x-y) data slices. The model only learns 2D data distributions, treating the 3D point cloud as separate batches of 2D data, which reduces the number of model parameters and enhances inference speed; (2) The introduction of a Slice Interaction Network (SIN). To maintain vertical relationships across slices, we incorporate SIN into the 2D backbone network, which improves the model's 3D object perception capability. Extensive experiments demonstrate that PointSlice achieves high detection accuracy and inference speed. On the Waymo dataset, PointSlice is 1.13x faster and has 0.79x fewer parameters than the state-of-the-art voxel-based method (SAFDNet), with only a 1.2 mAPH accuracy reduction. On the nuScenes dataset, we achieve a state-of-the-art detection result of 66.74 mAP. On the Argoverse 2 dataset, PointSlice is 1.10x faster, with 0.66x fewer parameters and a 1.0 mAP accuracy reduction. The code will be available at https://github.com/qifeng22/PointSlice2.

Xiao Liang, Zijian Zhao, Weichao Zeng et al.

Learning musical structures and composition patterns is necessary for both music generation and understanding, but current methods do not make uniform use of learned features to generate and comprehend music simultaneously. In this paper, we propose PianoBART, a pre-trained model that uses BART for both symbolic piano music generation and understanding. We devise a multi-level object selection strategy for different pre-training tasks of PianoBART, which can prevent information leakage or loss and enhance learning ability. The musical semantics captured in pre-training are fine-tuned for music generation and understanding tasks. Experiments demonstrate that PianoBART efficiently learns musical patterns and achieves outstanding performance in generating high-quality coherent pieces and comprehending music. Our code and supplementary material are available at https://github.com/RS2002/PianoBart.

Yeshu Li, Jonathan Cui, Yilun Sheng et al.

Whole brain segmentation is an important neuroimaging task that segments the whole brain volume into anatomically labeled regions-of-interest. Convolutional neural networks have demonstrated good performance in this task. Existing solutions, usually segment the brain image by classifying the voxels, or labeling the slices or the sub-volumes separately. Their representation learning is based on parts of the whole volume whereas their labeling result is produced by aggregation of partial segmentation. Learning and inference with incomplete information could lead to sub-optimal final segmentation result. To address these issues, we propose to adopt a full volume framework, which feeds the full volume brain image into the segmentation network and directly outputs the segmentation result for the whole brain volume. The framework makes use of complete information in each volume and can be implemented easily. An effective instance in this framework is given subsequently. We adopt the $3$D high-resolution network (HRNet) for learning spatially fine-grained representations and the mixed precision training scheme for memory-efficient training. Extensive experiment results on a publicly available $3$D MRI brain dataset show that our proposed model advances the state-of-the-art methods in terms of segmentation performance. Source code is publicly available at https://github.com/microsoft/VoxHRNet.

Shengyu Zhao, Jonathan Cui, Yilun Sheng et al.

Numerous task-specific variants of conditional generative adversarial networks have been developed for image completion. Yet, a serious limitation remains that all existing algorithms tend to fail when handling large-scale missing regions. To overcome this challenge, we propose a generic new approach that bridges the gap between image-conditional and recent modulated unconditional generative architectures via co-modulation of both conditional and stochastic style representations. Also, due to the lack of good quantitative metrics for image completion, we propose the new Paired/Unpaired Inception Discriminative Score (P-IDS/U-IDS), which robustly measures the perceptual fidelity of inpainted images compared to real images via linear separability in a feature space. Experiments demonstrate superior performance in terms of both quality and diversity over state-of-the-art methods in free-form image completion and easy generalization to image-to-image translation. Code is available at https://github.com/zsyzzsoft/co-mod-gan.

Xinyao Zhang, Chang Liu, Xiao Liang et al.

Precise segmentation of irregular and densely arranged components is essential for robotic disassembly and material recovery in electronic waste (e-waste) recycling. This study evaluates the impact of model architecture and scale on segmentation performance by comparing SAM2, a transformer-based vision model, with the lightweight YOLOv8 network. Both models were trained and tested on a newly collected dataset of 1,456 annotated RGB images of laptop components including logic boards, heat sinks, and fans, captured under varying illumination and orientation conditions. Data augmentation techniques, such as random rotation, flipping, and cropping, were applied to improve model robustness. YOLOv8 achieved higher segmentation accuracy (mAP50 = 98.8%, mAP50-95 = 85%) and stronger boundary precision than SAM2 (mAP50 = 8.4%). SAM2 demonstrated flexibility in representing diverse object structures but often produced overlapping masks and inconsistent contours. These findings show that large pre-trained models require task-specific optimization for industrial applications. The resulting dataset and benchmarking framework provide a foundation for developing scalable vision algorithms for robotic e-waste disassembly and circular manufacturing systems.

Xiao Liang, Tao Shi, Yaoyuan Liang et al.

Video grounding aims to localize the target moment in an untrimmed video corresponding to a given sentence query. Existing methods typically select the best prediction from a set of predefined proposals or directly regress the target span in a single-shot manner, resulting in the absence of a systematical prediction refinement process. In this paper, we propose DiffusionVG, a novel framework with diffusion models that formulates video grounding as a conditional generation task, where the target span is generated from Gaussian noise inputs and interatively refined in the reverse diffusion process. During training, DiffusionVG progressively adds noise to the target span with a fixed forward diffusion process and learns to recover the target span in the reverse diffusion process. In inference, DiffusionVG can generate the target span from Gaussian noise inputs by the learned reverse diffusion process conditioned on the video-sentence representations. Without bells and whistles, our DiffusionVG demonstrates superior performance compared to existing well-crafted models on mainstream Charades-STA, ActivityNet Captions and TACoS benchmarks.

Xumeng Wen, Zihan Liu, Shun Zheng et al.

Recent advancements in long chain-of-thought (CoT) reasoning, particularly through the Group Relative Policy Optimization algorithm used by DeepSeek-R1, have led to significant interest in the potential of Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Models (LLMs). While RLVR promises to improve reasoning by allowing models to learn from free exploration, there remains debate over whether it truly enhances reasoning abilities or simply boosts sampling efficiency. This paper systematically investigates the impact of RLVR on LLM reasoning. We revisit Pass@K experiments and demonstrate that RLVR can extend the reasoning boundary for both mathematical and coding tasks. This is supported by our introduction of a novel evaluation metric, CoT-Pass@K, which captures reasoning success by accounting for both the final answer and intermediate reasoning steps. Furthermore, we present a theoretical framework explaining RLVR's incentive mechanism, demonstrating how it can encourage correct reasoning even when rewards are based solely on answer correctness. Our analysis of RLVR's training dynamics reveals that it incentivizes correct reasoning early in the process, with substantial improvements in reasoning quality confirmed through extensive evaluations. These findings provide strong evidence of RLVR's potential to enhance LLM reasoning, offering valuable insights into its mechanisms and performance improvements.

Xiao Liang, Zhongzhi Li, Yeyun Gong et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a key paradigm for post-training Large Language Models (LLMs), particularly for complex reasoning tasks. However, vanilla RLVR training has been shown to improve Pass@1 performance at the expense of policy entropy, leading to reduced generation diversity and limiting the Pass@k performance, which typically represents the upper bound of LLM reasoning capability. In this paper, we systematically analyze the policy's generation diversity from the perspective of training problems and find that augmenting and updating training problems helps mitigate entropy collapse during training. Based on these observations, we propose an online Self-play with Variational problem Synthesis (SvS) strategy for RLVR training, which uses the policy's correct solutions to synthesize variational problems while ensuring their reference answers remain identical to the originals. This self-improving strategy effectively maintains policy entropy during training and substantially improves Pass@k compared with standard RLVR, sustaining prolonged improvements and achieving absolute gains of 18.3% and 22.8% in Pass@32 performance on the competition-level AIME24 and AIME25 benchmarks. Experiments on 12 reasoning benchmarks across varying model sizes from 3B to 32B consistently demonstrate the generalizability and robustness of SvS.

Yiyao Yu, Yuxiang Zhang, Dongdong Zhang et al.

Large Language Models (LLMs) have made notable progress in mathematical reasoning, yet often rely on single-paradigm reasoning, limiting their effectiveness across diverse tasks. We introduce Chain-of-Reasoning (CoR), a novel unified framework integrating multiple reasoning paradigms--Natural Language Reasoning (NLR), Algorithmic Reasoning (AR), and Symbolic Reasoning (SR)--to enable synergistic collaboration. CoR generates multiple potential answers via different reasoning paradigms and synthesizes them into a coherent final solution. We propose a Progressive Paradigm Training (PPT) strategy for models to progressively master these paradigms, leading to CoR-Math-7B. Experimental results demonstrate that CoR-Math-7B significantly outperforms current SOTA models, achieving up to a 41.0% absolute improvement over GPT-4o in theorem proving and a 15.0% improvement over RL-based methods on the MATH benchmark in arithmetic tasks. These results show the enhanced mathematical comprehension ability of our model, enabling zero-shot generalization across tasks.