Chi Chen

Chi Chen, Ruoyu Qin, Fuwen Luo et al. · tsinghua

Recently, Multimodal Large Language Models (MLLMs) that enable Large Language Models (LLMs) to interpret images through visual instruction tuning have achieved significant success. However, existing visual instruction tuning methods only utilize image-language instruction data to align the language and image modalities, lacking a more fine-grained cross-modal alignment. In this paper, we propose Position-enhanced Visual Instruction Tuning (PVIT), which extends the functionality of MLLMs by integrating an additional region-level vision encoder. This integration promotes a more detailed comprehension of images for the MLLM. In addition, to efficiently achieve a fine-grained alignment between the vision modules and the LLM, we design multiple data generation strategies to construct an image-region-language instruction dataset. Finally, we present both quantitative experiments and qualitative analysis that demonstrate the superiority of the proposed model. Code and data will be released at https://github.com/PVIT-official/PVIT.

Shuxin Zheng, Jiyan He, Chang Liu et al. · microsoft-research

Advances in deep learning have greatly improved structure prediction of molecules. However, many macroscopic observations that are important for real-world applications are not functions of a single molecular structure, but rather determined from the equilibrium distribution of structures. Traditional methods for obtaining these distributions, such as molecular dynamics simulation, are computationally expensive and often intractable. In this paper, we introduce a novel deep learning framework, called Distributional Graphormer (DiG), in an attempt to predict the equilibrium distribution of molecular systems. Inspired by the annealing process in thermodynamics, DiG employs deep neural networks to transform a simple distribution towards the equilibrium distribution, conditioned on a descriptor of a molecular system, such as a chemical graph or a protein sequence. This framework enables efficient generation of diverse conformations and provides estimations of state densities. We demonstrate the performance of DiG on several molecular tasks, including protein conformation sampling, ligand structure sampling, catalyst-adsorbate sampling, and property-guided structure generation. DiG presents a significant advancement in methodology for statistically understanding molecular systems, opening up new research opportunities in molecular science.

Ziyue Wang, Chi Chen, Peng Li et al. · tsinghua

Large Language Models (LLMs) demonstrate impressive reasoning ability and the maintenance of world knowledge not only in natural language tasks, but also in some vision-language tasks such as open-domain knowledge-based visual question answering (OK-VQA). As images are invisible to LLMs, researchers convert images to text to engage LLMs into the visual question reasoning procedure. This leads to discrepancies between images and their textual representations presented to LLMs, which consequently impedes final reasoning performance. To fill the information gap and better leverage the reasoning capability, we design a framework that enables LLMs to proactively ask relevant questions to unveil more details in the image, along with filters for refining the generated information. We validate our idea on OK-VQA and A-OKVQA. Our method continuously boosts the performance of baselines methods by an average gain of 2.15% on OK-VQA, and achieves consistent improvements across different LLMs.

Shihan Wu, Xuecheng Liu, Shaoxuan Xie et al.

Despite the critical role of bimanual manipulation in endowing robots with human-like dexterity, large-scale and diverse datasets remain scarce due to the significant hardware heterogeneity across bimanual robotic platforms. To bridge this gap, we introduce RoboCOIN, a large-scale multi-embodiment bimanual manipulation dataset comprising over 180,000 demonstrations collected from 15 distinct robotic platforms. Spanning 16 diverse environments-including residential, commercial, and industrial settings-the dataset features 421 bimanual tasks systematically categorized by 39 bimanual collaboration actions and 432 objects. A key innovation of our work is the hierarchical capability pyramid, which provides granular annotations ranging from trajectory-level concepts to segment-level subtasks and frame-level kinematics. Furthermore, we present CoRobot, an efficient data processing pipeline powered by the Robot Trajectory Markup Language (RTML), designed to facilitate quality assessment, automated annotation, and unified multi-embodiment and data management. Extensive experiments demonstrate the effectiveness of RoboCOIN in enhancing the performance of various bimanual manipulation models across a wide spectrum of robotic embodiments. The entire dataset and codebase are fully open-sourced, providing a valuable resource for advancing research in bimanual and multi-embodiment manipulation.

Shiqi Dai, Zizhi Ma, Zhicong Luo et al.

While Multimodal Large Language Models (MLLMs) have exhibited remarkable general intelligence across diverse domains, their potential in low-altitude applications dominated by Unmanned Aerial Vehicles (UAVs) remains largely underexplored. Existing MLLM benchmarks rarely cover the unique challenges of low-altitude scenarios, while UAV-related evaluations mainly focus on specific tasks such as localization or navigation, without a unified evaluation of MLLMs'general intelligence. To bridge this gap, we present MM-UAVBench, a comprehensive benchmark that systematically evaluates MLLMs across three core capability dimensions-perception, cognition, and planning-in low-altitude UAV scenarios. MM-UAVBench comprises 19 sub-tasks with over 5.7K manually annotated questions, all derived from real-world UAV data collected from public datasets. Extensive experiments on 16 open-source and proprietary MLLMs reveal that current models struggle to adapt to the complex visual and cognitive demands of low-altitude scenarios. Our analyses further uncover critical bottlenecks such as spatial bias and multi-view understanding that hinder the effective deployment of MLLMs in UAV scenarios. We hope MM-UAVBench will foster future research on robust and reliable MLLMs for real-world UAV intelligence.

Fanhu Zeng, Zhicong Luo, Zefan Wang et al.

Visual reasoning through reinforcement learning with verifiable rewards (RLVR) has achieved remarkable progress. However, when dealing with multi-source inputs, existing approaches tend to treat them as a mere accumulation of information, lacking explicit mechanisms to distinguish whether integrating additional sources yields information gain or introduces interference. Therefore, they struggle to effectively model dynamic interaction when integrating multiple sources, particularly when they differ significantly in physical properties and semantics, e.g., infrared and depth, leading to inferior performance to mono-source reasoning when a certain source holds the dominant signal. To address this issue, we propose MARS, a novel mono-anchored multi-source reasoning framework that models each visual modality as an independent information source. Specifically, by treating mono-source rewards as dynamic anchors, our method explicitly incorporates the information gain introduced by multi-source fusion into advantage normalization and adaptively emphasizes mutual promotion between sources while suppressing potential noise or conflicts during RLVR. From theoretical analysis, our method effectively quantifies information gain introduced by multi-source integration in gradient estimation, enabling consistent modality regulation. Empirical results also show impressive 3.2% and 4.9% performance gains on GRPO and DAPO across diverse datasets, confirming effectiveness of our method.

Junbo Cui, Bokai Xu, Chongyi Wang et al.

Recent progress in multimodal large language models (MLLMs) has brought AI capabilities from static offline data processing to real-time streaming interaction, yet they still remain far from human-level multimodal interaction. The key bottlenecks are no longer modality coverage or latency alone, but the interaction paradigm itself. First, perception and response are still separated into alternating phases, preventing models from incorporating new inputs for timely adjustment during generation. Second, most current models remain reactive, responding only to explicit user requests instead of acting proactively in the evolving multimodal environment. We present MiniCPM-o 4.5, our latest effort towards human-like multimodal interaction, which mitigates these gaps by real-time full-duplex omni-modal interaction. It can see, listen, and speak simultaneously in real-time, while also exhibiting proactive behaviors such as issuing reminders or comments based on its continuous understanding of the live scene. The key technique behind MiniCPM-o 4.5 is Omni-Flow, a unified streaming framework that aligns omni-modal inputs and outputs along a shared temporal axis. This formulation converts conventional turn-based interaction into a full-duplex, time-aligned process, enabling simultaneous perception and response and allowing proactive behavior to arise within the same framework. With a total of 9B parameters, MiniCPM-o 4.5 approaches Gemini 2.5 Flash in vision-language capabilities, delivering state-of-the-art open-source performance at its scale. It also surpasses Qwen3-Omni-30B-A3B in omni-modal understanding and delivers better speech generation, with significantly higher computation efficiency. Driven by its efficient architecture design and inference optimization, the model can perform real-time full-duplex omni-modal interaction on edge devices with less than 12GB RAM cost.

Wenxuan Li, Qin Zou, Chi Chen et al.

3D object detection in driving scenarios faces the challenge of complex road environments, which can lead to the loss or incompleteness of key features, thereby affecting perception performance. To address this issue, we propose an advanced detection framework called Co-Fix3D. Co-Fix3D integrates Local and Global Enhancement (LGE) modules to refine Bird's Eye View (BEV) features. The LGE module uses Discrete Wavelet Transform (DWT) for pixel-level local optimization and incorporates an attention mechanism for global optimization. To handle varying detection difficulties, we adopt multi-head LGE modules, enabling each module to focus on targets with different levels of detection complexity, thus further enhancing overall perception capability. Experimental results show that on the nuScenes dataset's LiDAR benchmark, Co-Fix3D achieves 69.4\% mAP and 73.5\% NDS, while on the multimodal benchmark, it achieves 72.3\% mAP and 74.7\% NDS. The source code is publicly available at \href{https://github.com/rubbish001/Co-Fix3d}{https://github.com/rubbish001/Co-Fix3d}.

Xuanle Zhao, Xianzhen Luo, Qi Shi et al.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in chart understanding tasks. However, interpreting charts with textual descriptions often leads to information loss, as it fails to fully capture the dense information embedded in charts. In contrast, parsing charts into code provides lossless representations that can effectively contain all critical details. Although existing open-source MLLMs have achieved success in chart understanding tasks, they still face two major challenges when applied to chart-to-code tasks: (1) Low executability and poor restoration of chart details in the generated code and (2) Lack of large-scale and diverse training data. To address these challenges, we propose \textbf{ChartCoder}, the first dedicated chart-to-code MLLM, which leverages Code LLMs as the language backbone to enhance the executability of the generated code. Furthermore, we introduce \textbf{Chart2Code-160k}, the first large-scale and diverse dataset for chart-to-code generation, and propose the \textbf{Snippet-of-Thought (SoT)} method, which transforms direct chart-to-code generation data into step-by-step generation. Experiments demonstrate that ChartCoder, with only 7B parameters, surpasses existing open-source MLLMs on chart-to-code benchmarks, achieving superior chart restoration and code excitability. Our code is available at https://github.com/thunlp/ChartCoder.

Junming Lin, Zheng Fang, Chi Chen et al.

The rapid development of Multimodal Large Language Models (MLLMs) has expanded their capabilities from image comprehension to video understanding. However, most of these MLLMs focus primarily on offline video comprehension, necessitating extensive processing of all video frames before any queries can be made. This presents a significant gap compared to the human ability to watch, listen, think, and respond to streaming inputs in real time, highlighting the limitations of current MLLMs. In this paper, we introduce StreamingBench, the first comprehensive benchmark designed to evaluate the streaming video understanding capabilities of MLLMs. StreamingBench assesses three core aspects of streaming video understanding: (1) real-time visual understanding, (2) omni-source understanding, and (3) contextual understanding. The benchmark consists of 18 tasks, featuring 900 videos and 4,500 human-curated QA pairs. Each video features five questions presented at different time points to simulate a continuous streaming scenario. We conduct experiments on StreamingBench with 13 open-source and proprietary MLLMs and find that even the most advanced proprietary MLLMs like Gemini 1.5 Pro and GPT-4o perform significantly below human-level streaming video understanding capabilities. We hope our work can facilitate further advancements for MLLMs, empowering them to approach human-level video comprehension and interaction in more realistic scenarios.

Chi Chen, Tianle Jiang, Xiaodong Wei et al.

The three-dimensional (3D) microstructures of polycrystalline materials exert a critical influence on their mechanical and physical properties. Realistic, controllable construction of these microstructures is a key step toward elucidating structure-property relationships, yet remains a formidable challenge. Herein, we propose PolyCrysDiff, a framework based on conditional latent diffusion that enables the end-to-end generation of computable 3D polycrystalline microstructures. Comprehensive qualitative and quantitative evaluations demonstrate that PolyCrysDiff faithfully reproduces target grain morphologies, orientation distributions, and 3D spatial correlations, while achieving an $R^2$ over 0.972 on grain attributes (e.g., size and sphericity) control, thereby outperforming mainstream approaches such as Markov random field (MRF)- and convolutional neural network (CNN)-based methods. The computability and physical validity of the generated microstructures are verified through a series of crystal plasticity finite element method (CPFEM) simulations. Leveraging PolyCrysDiff's controllable generative capability, we systematically elucidate how grain-level microstructural characteristics affect the mechanical properties of polycrystalline materials. This development is expected to pave a key step toward accelerated, data-driven optimization and design of polycrystalline materials.

Wenxuan Li, Qin Zou, Shoubing Chen et al.

In autonomous driving, 3D object detection is essential for accurate perception and reliable decision-making. However, object motion and ego-motion often induce cross-frame spatiotemporal inconsistencies in BEV-based detectors, leading to temporal BEV feature misalignment and degraded spatiotemporal consistency. To address these challenges, we propose Co-Fusion4D, a unified framework that explicitly preserves cross-frame spatiotemporal consistency and suppresses temporal feature drift. Co-Fusion4D adopts a current-frame-centric strategy, treating the current frame as the primary source of information while selectively incorporating historical frames after spatiotemporal filtering and alignment. This dominant-complementary mechanism effectively mitigates cumulative alignment errors, suppresses noisy feature propagation, and exploits reliable temporal cues for a more consistent BEV representation. In addition, Co-Fusion4D integrates a Dual Attention Fusion (DAF) module to further enhance spatiotemporal feature interaction. DAF jointly leverages intra-frame spatial attention and inter-frame temporal attention to adaptively align and fuse multi-frame features, emphasizing motion-consistent regions while suppressing spurious correlations. By departing from conventional uniform fusion paradigms, this design substantially improves the temporal stability and discriminative capability of BEV representations. Extensive experiments on the nuScenes benchmark demonstrate that Co-Fusion4D achieves state-of-the-art performance, with 74.9% mAP and 75.6% NDS, without relying on test-time augmentation or external data.

Xuanle Zhao, Xuexin Liu, Haoyue Yang et al.

Although multimodal large language models (MLLMs) show promise in generating chart rendering code, editing charts via code presents a greater challenge. This task demands MLLMs to integrate chart understanding and reasoning capacities, which are labor-intensive. While many MLLMs claim such editing capabilities, current evaluations rely on limited case studies, highlighting the urgent need for a comprehensive evaluation framework. In this work, we propose \textsc{ChartEdit}, a novel benchmark designed for chart editing tasks, featuring $1405$ diverse editing instructions applied to $233$ real-world charts, each manually annotated and validated for accuracy. Utilizing \textsc{ChartEdit}, we evaluate the performance of 10 mainstream MLLMs across two types of experiments at both the code and chart levels. The results suggest that large-scale models can generate code to produce images that partially match the reference images. However, their ability to generate accurate edits according to the instructions remains limited. The state-of-the-art (SOTA) model achieves a score of only $59.96$, highlighting significant challenges in precise modification. In contrast, small-scale models, including chart-domain models, struggle both with following editing instructions and generating overall chart images, underscoring the need for further development in this area. Code is available at https://github.com/xxlllz/ChartEdit.

Xinyu Ma, Ziyang Ding, Zhicong Luo et al.

Human experts excel at fine-grained visual discrimination by leveraging domain knowledge to refine perceptual features, a capability that remains underdeveloped in current Multimodal Large Language Models (MLLMs). Despite possessing vast expert-level knowledge, MLLMs struggle to integrate reasoning into visual perception, often generating direct responses without deeper analysis. To bridge this gap, we introduce knowledge-intensive visual grounding (KVG), a novel visual grounding task that requires both fine-grained perception and domain-specific knowledge integration. To address the challenges of KVG, we propose DeepPerception, an MLLM enhanced with cognitive visual perception capabilities. Our approach consists of (1) an automated data synthesis pipeline that generates high-quality, knowledge-aligned training samples, and (2) a two-stage training framework combining supervised fine-tuning for cognitive reasoning scaffolding and reinforcement learning to optimize perception-cognition synergy. To benchmark performance, we introduce KVG-Bench a comprehensive dataset spanning 10 domains with 1.3K manually curated test cases. Experimental results demonstrate that DeepPerception significantly outperforms direct fine-tuning, achieving +8.08\% accuracy improvements on KVG-Bench and exhibiting +4.60\% superior cross-domain generalization over baseline approaches. Our findings highlight the importance of integrating cognitive processes into MLLMs for human-like visual perception and open new directions for multimodal reasoning research. The data, codes, and models are released at https://github.com/thunlp/DeepPerception.

Oliver Knitter, Sang Hyub Kim, Maximilian Wurzer et al.

We present an experimental study of energy-to-solution (ETS) of hybrid quantum-classical applications, enabled by direct instrumentation of power consumption of a Forte Enterprise trapped-ion quantum processor. We apply this methodology to a hybrid quantum-classical pipeline for quantum fine-tuning of foundational AI models, and validate the approach end-to-end on quantum hardware. Despite noise and limited qubit counts, the resulting models achieve accuracy competitive with and exceeding classical baselines such as logistic regression and support vector classifiers. Our results show that QPU energy consumption scales approximately linearly with qubit number for shallow circuits, while classical simulation exhibits exponential scaling, indicating a break-even for ETS around 34 qubits. The classification error improvement of the best quantum fine-tuned model over the best classical fine-tuned model considered in this study is around 24%. We further contextualize these findings with comparisons to tensor network methods. This work establishes energy-to-solution as a measurable and scalable metric for evaluating quantum applications and provides experimental evidence of favorable energy-accuracy trade-offs.

You Li, Heyu Huang, Chi Chen et al.

The recent advancement of Multimodal Large Language Models (MLLMs) has significantly improved their fine-grained perception of single images and general comprehension across multiple images. However, existing MLLMs still face challenges in achieving precise grounding in complex multi-image scenarios. To address this, we first explore a Chain-of-Thought (CoT) framework that integrates single-image grounding with multi-image comprehension. While partially effective, it remains unstable and struggles to capture abstract visual information due to its non-end-to-end nature. Therefore, we introduce Migician, the first multi-image grounding model capable of performing free-form and accurate grounding across multiple images. To support this, we present the MGrounding-630k dataset, which comprises data for several multi-image grounding tasks derived from existing datasets, along with newly generated free-form grounding instruction-following data. Furthermore, we propose MIG-Bench, a comprehensive benchmark specifically designed for evaluating multi-image grounding capabilities. Experimental results demonstrate that our model achieves significantly superior multi-image grounding capabilities, outperforming the best existing MLLMs by 24.94% and even surpassing much larger 70B models. Our code, model, dataset, and benchmark are fully open-sourced at https://migician-vg.github.io/.

Xinyue Lou, You Li, Jinan Xu et al.

The rapid development of Multimodal Large Reasoning Models (MLRMs) has demonstrated broad application potential, yet their safety and reliability remain critical concerns that require systematic exploration. To address this gap, we conduct a comprehensive and systematic safety evaluation of 11 MLRMs across 5 benchmarks and unveil prevalent safety degradation phenomena in most advanced models. Moreover, our analysis reveals distinct safety patterns across different benchmarks: significant safety degradation is observed across jailbreak robustness benchmarks, whereas safety-awareness benchmarks demonstrate less pronounced degradation. In particular, the long thought process in some scenarios even enhances safety performance. Therefore, it is a potential approach to address safety issues in MLRMs by leveraging the intrinsic reasoning capabilities of the model to detect unsafe intent. To operationalize this insight, we construct a multimodal tuning dataset that incorporates a safety-oriented thought process. Experimental results from fine-tuning existing MLRMs with this dataset effectively enhances the safety on both jailbreak robustness and safety-awareness benchmarks. This study provides a new perspective for developing safe MLRMs. Our dataset is available at https://github.com/xinyuelou/Think-in-Safety.

You Li, Chi Chen, Yanghao Li et al.

Latent visual reasoning aims to mimic human's imagination process by meditating through hidden states of Multimodal Large Language Models. While recognized as a promising paradigm for visual reasoning, the underlying mechanisms driving its effectiveness remain unclear. Motivated to demystify the true source of its efficacy, we investigate the validity of latent reasoning using Causal Mediation Analysis. We model the process as a causal chain: the input as the treatment, the latent tokens as the mediator, and the final answer as the outcome. Our findings uncover two critical disconnections: (a) Input-Latent Disconnect: dramatic perturbations on the input result in negligible changes to the latent tokens, suggesting that latent tokens do not effectively attend to the input sequence. (b) Latent-Answer Disconnect: perturbations on the latent tokens yield minimal impact on the final answer, indicating the limited causal effect latent tokens imposing on the outcome. Furthermore, extensive probing analysis reveals that latent tokens encode limited visual information and exhibit high similarity. Consequently, we challenge the necessity of latent reasoning and propose a straightforward alternative named CapImagine, which teaches the model to explicitly imagine using text. Experiments on vision-centric benchmarks show that CapImagine significantly outperforms complex latent-space baselines, highlighting the superior potential of visual reasoning through explicit imagination.

Tianyu Yu, Zefan Wang, Chongyi Wang et al. · tsinghua

Multimodal Large Language Models (MLLMs) are undergoing rapid progress and represent the frontier of AI development. However, their training and inference efficiency have emerged as a core bottleneck in making MLLMs more accessible and scalable. To address the challenges, we present MiniCPM-V 4.5, an 8B parameter model designed for high efficiency and strong performance. We introduce three core improvements in model architecture, data strategy and training method: a unified 3D-Resampler model architecture for highly compact encoding over images and videos, a unified learning paradigm for document knowledge and text recognition without heavy data engineering, and a hybrid reinforcement learning strategy for proficiency in both short and long reasoning modes. Comprehensive experimental results in OpenCompass evaluation show that MiniCPM-V 4.5 surpasses widely used proprietary models such as GPT-4o-latest, and significantly larger open-source models such as Qwen2.5-VL 72B. Notably, the strong performance is achieved with remarkable efficiency. For example, on the widely adopted VideoMME benchmark, MiniCPM-V 4.5 achieves state-of-the-art performance among models under 30B size, using just 46.7\% GPU memory cost and 8.7\% inference time of Qwen2.5-VL 7B.

Shichu Sun, Yichen Zhang, Haolin Song et al.

Visual encoding followed by token condensing has become the standard architectural paradigm in multi-modal large language models (MLLMs). Many recent MLLMs increasingly favor global native- resolution visual encoding over slice-based methods. To investigate this trend, we systematically compare their behavior on vision-language understanding and attention patterns, revealing that global encoding enhances overall capability but at the expense of greater computational overhead. To address this issue, we present LLaVA-UHD v3, an MLLM centered upon our proposed Progressive Visual Compression (PVC) method, which can be seamlessly integrated into standard Vision Transformer (ViT) to enable efficient native-resolution encoding. The PVC approach consists of two key modules: (i) refined patch embedding, which supports flexible patch-size scaling for fine-grained visual model- ing, (ii) windowed token compression, hierarchically deployed across ViT layers to progressively aggregate local token representations. Jointly modulated by these two modules, a widely pretrained ViT can be reconfigured into an efficient architecture while largely preserving generality. Evaluated across extensive benchmarks, the transformed ViT, termed ViT-UHD, demonstrates competitive performance with MoonViT while reducing TTFT (time-to-first-token) by 2.4x, when developed within an identical MLLM architecture. Building upon ViT-UHD, LLaVA-UHD v3 also achieves competitive performance to Qwen2-VL, while further reducing TTFT by 1.9x. We will release all code and checkpoints to support future research on efficient MLLMs.

Fuwen Luo, Shengfeng Lou, Chi Chen et al.

Video temporal understanding is crucial for multimodal large language models (MLLMs) to reason over events in videos. Despite recent advances in general video understanding, current MLLMs still struggle with fine-grained temporal reasoning. While reinforcement learning (RL) has been explored to address this issue recently, existing RL approaches remain limited in effectiveness. In this work, we propose MUSEG, a novel RL-based method that enhances temporal understanding by introducing timestamp-aware multi-segment grounding. MUSEG enables MLLMs to align queries with multiple relevant video segments, promoting more comprehensive temporal reasoning. To facilitate effective learning, we design a customized RL training recipe with phased rewards that progressively guides the model toward temporally grounded reasoning. Extensive experiments on temporal grounding and time-sensitive video QA tasks demonstrate that MUSEG significantly outperforms existing methods and generalizes well across diverse temporal understanding scenarios. View our project at https://github.com/THUNLP-MT/MUSEG.

Sang Hyub Kim, Oliver Knitter, Jonathan Mei et al.

We study parity features as representations that can be evaluated entirely classically once the binary or quantized input representation and parity words are fixed, particularly when labels depend on higher-order feature interactions or when discrete inference interfaces support perturbation robustness. A parity feature is a signed product over selected bits of a binary input: once the participating bits are known, evaluation requires no quantum resources. Reaching a useful parity representation requires solving two challenges. When the input is parity-ready (a meaningful binary string), the challenge is basis discovery: selecting useful parity words from a combinatorial search space. Otherwise, the challenge is encoding: constructing a binary vector on which parity computation is meaningful. We use hybrid quantum-classical training pipelines to address these: learnable Pauli word selection for basis discovery, learned projection encodings for continuous embeddings, and sPQC-Parity for discrete inputs. On three native-binary parity tasks with 5-10 qubits, the learned parity basis improves mean accuracy by 23.9% to 41.7% over logistic-regression and support-vector baselines. A model comparison shows that the improvement comes primarily from discovering the right parity basis, rather than from quantum moment computation at inference. On five continuous text benchmarks, learned projection recovers much of the loss introduced by dimensionality reduction and fixed binarization, exceeding the full continuous baseline on CR, SST-2, and SST-5. On three encoding-limited discrete datasets, when compared with PCA-bin as the baseline, sPQC-Parity reaches 94.6% improvement on mushroom, 3.0% on splice, and matches PCA-bin on promoter. We also analyze inference robustness under binary or quantized inference, where rounding gives exact invariance below half the quantization step.

Chi Chen, Xin Peng, Zhenchang Xing et al.

Context based API recommendation is an important way to help developers find the needed APIs effectively and efficiently. For effective API recommendation, we need not only a joint view of both structural and textual code information, but also a holistic view of correlated API usage in control and data flow graph as a whole. Unfortunately, existing API recommendation methods exploit structural or textual code information separately. In this work, we propose a novel API recommendation approach called APIRec-CST (API Recommendation by Combining Structural and Textual code information). APIRec-CST is a deep learning model that combines the API usage with the text information in the source code based on an API Context Graph Network and a Code Token Network that simultaneously learn structural and textual features for API recommendation. We apply APIRec-CST to train a model for JDK library based on 1,914 open-source Java projects and evaluate the accuracy and MRR (Mean Reciprocal Rank) of API recommendation with another 6 open-source projects. The results show that our approach achieves respectively a top-1, top-5, top-10 accuracy and MRR of 60.3%, 81.5%, 87.7% and 69.4%, and significantly outperforms an existing graph-based statistical approach and a tree-based deep learning approach for API recommendation. A further analysis shows that textual code information makes sense and improves the accuracy and MRR. We also conduct a user study in which two groups of students are asked to finish 6 programming tasks with or without our APIRec-CST plugin. The results show that APIRec-CST can help the students to finish the tasks faster and more accurately and the feedback on the usability is overwhelmingly positive.

Haoqiang Zhou, Chi Chen, Yongfeng Zhi et al.

Distributed stochastic optimization enables multi-agent collaboration in applications such as distributed learning and sensor networks, but also raises critical privacy concerns due to the involvement of sensitive data. While existing privacy-preserving approaches often face limitations in balancing accuracy with efficiency, we propose a novel distributed stochastic gradient descent algorithm that integrates Paillier homomorphic encryption with heterogeneous and time-varying random stepsizes. The proposed algorithm provides inherent privacy protection against both internal honest-but-curious agents and external eavesdroppers, without relying on any trusted neighbors. Furthermore, we incorporate an attenuation factor to effectively mitigate quantization error induced by the encryption process, ensuring almost sure convergence to the optimal solution while maintaining privacy preservation. Numerical simulations demonstrate the effectiveness and efficiency of the proposed approach.

Shiji Zhao, Ranjie Duan, Fengxiang Wang et al.

Multimodal Large Language Models (MLLMs) have achieved impressive performance and have been put into practical use in commercial applications, but they still have potential safety mechanism vulnerabilities. Jailbreak attacks are red teaming methods that aim to bypass safety mechanisms and discover MLLMs' potential risks. Existing MLLMs' jailbreak methods often bypass the model's safety mechanism through complex optimization methods or carefully designed image and text prompts. Despite achieving some progress, they have a low attack success rate on commercial closed-source MLLMs. Unlike previous research, we empirically find that there exists a Shuffle Inconsistency between MLLMs' comprehension ability and safety ability for the shuffled harmful instruction. That is, from the perspective of comprehension ability, MLLMs can understand the shuffled harmful text-image instructions well. However, they can be easily bypassed by the shuffled harmful instructions from the perspective of safety ability, leading to harmful responses. Then we innovatively propose a text-image jailbreak attack named SI-Attack. Specifically, to fully utilize the Shuffle Inconsistency and overcome the shuffle randomness, we apply a query-based black-box optimization method to select the most harmful shuffled inputs based on the feedback of the toxic judge model. A series of experiments show that SI-Attack can improve the attack's performance on three benchmarks. In particular, SI-Attack can obviously improve the attack success rate for commercial MLLMs such as GPT-4o or Claude-3.5-Sonnet.

Zhong Zhang, Yaxi Lu, Yikun Fu et al. · tsinghua

The recent progress of large language model agents has opened new possibilities for automating tasks through graphical user interfaces (GUIs), especially in mobile environments where intelligent interaction can greatly enhance usability. However, practical deployment of such agents remains constrained by several key challenges. Existing training data is often noisy and lack semantic diversity, which hinders the learning of precise grounding and planning. Models trained purely by imitation tend to overfit to seen interface patterns and fail to generalize in unfamiliar scenarios. Moreover, most prior work focuses on English interfaces while overlooks the growing diversity of non-English applications such as those in the Chinese mobile ecosystem. In this work, we present AgentCPM-GUI, an 8B-parameter GUI agent built for robust and efficient on-device GUI interaction. Our training pipeline includes grounding-aware pre-training to enhance perception, supervised fine-tuning on high-quality Chinese and English trajectories to imitate human-like actions, and reinforcement fine-tuning with GRPO to improve reasoning capability. We also introduce a compact action space that reduces output length and supports low-latency execution on mobile devices. AgentCPM-GUI achieves state-of-the-art performance on five public benchmarks and a new Chinese GUI benchmark called CAGUI, reaching $96.9\%$ Type-Match and $91.3\%$ Exact-Match. To facilitate reproducibility and further research, we publicly release all code, model checkpoint, and evaluation data.

Fuwen Luo, Chi Chen, Zihao Wan et al. · tsinghua

Multimodal large language models (MLLMs) have demonstrated promising results in a variety of tasks that combine vision and language. As these models become more integral to research and applications, conducting comprehensive evaluations of their capabilities has grown increasingly important. However, most existing benchmarks fail to consider that, in certain situations, images need to be interpreted within a broader context. In this work, we introduce a new benchmark, named as CODIS, designed to assess the ability of models to use context provided in free-form text to enhance visual comprehension. Our findings indicate that MLLMs consistently fall short of human performance on this benchmark. Further analysis confirms that these models struggle to effectively extract and utilize contextual information to improve their understanding of images. This underscores the pressing need to enhance the ability of MLLMs to comprehend visuals in a context-dependent manner. View our project website at https://thunlp-mt.github.io/CODIS.

Chi Chen, Yiyang Du, Zheng Fang et al. · tsinghua

Recent developments in Multimodal Large Language Models (MLLMs) have shown rapid progress, moving towards the goal of creating versatile MLLMs that understand inputs from various modalities. However, existing methods typically rely on joint training with paired multimodal instruction data, which is resource-intensive and challenging to extend to new modalities. In this paper, we propose a new paradigm through the model composition of existing MLLMs to create a new model that retains the modal understanding capabilities of each original model. Our basic implementation, NaiveMC, demonstrates the effectiveness of this paradigm by reusing modality encoders and merging LLM parameters. Furthermore, we introduce DAMC to address parameter interference and mismatch issues during the merging process, thereby enhancing the model performance. To facilitate research in this area, we propose MCUB, a benchmark for assessing ability of MLLMs to understand inputs from diverse modalities. Experiments on this benchmark and four other multimodal understanding tasks show significant improvements over baselines, proving that model composition can create a versatile model capable of processing inputs from multiple modalities.

Xingyu Su, Xiaojie Zhu, Yang Li et al.

Amidst the surge in deep learning-based password guessing models, challenges of generating high-quality passwords and reducing duplicate passwords persist. To address these challenges, we present PagPassGPT, a password guessing model constructed on Generative Pretrained Transformer (GPT). It can perform pattern guided guessing by incorporating pattern structure information as background knowledge, resulting in a significant increase in the hit rate. Furthermore, we propose D&C-GEN to reduce the repeat rate of generated passwords, which adopts the concept of a divide-and-conquer approach. The primary task of guessing passwords is recursively divided into non-overlapping subtasks. Each subtask inherits the knowledge from the parent task and predicts succeeding tokens. In comparison to the state-of-the-art model, our proposed scheme exhibits the capability to correctly guess 12% more passwords while producing 25% fewer duplicates.

Houzhe Wang, Xiaojie Zhu, Chi Chen et al.

With recent legislation on the right to be forgotten, machine unlearning has emerged as a crucial research area. It facilitates the removal of a user's data from federated trained machine learning models without the necessity for retraining from scratch. However, current machine unlearning algorithms are confronted with challenges of efficiency and validity. To address the above issues, we propose a new framework, named Goldfish. It comprises four modules: basic model, loss function, optimization, and extension. To address the challenge of low validity in existing machine unlearning algorithms, we propose a novel loss function. It takes into account the loss arising from the discrepancy between predictions and actual labels in the remaining dataset. Simultaneously, it takes into consideration the bias of predicted results on the removed dataset. Moreover, it accounts for the confidence level of predicted results. Additionally, to enhance efficiency, we adopt knowledge a distillation technique in the basic model and introduce an optimization module that encompasses the early termination mechanism guided by empirical risk and the data partition mechanism. Furthermore, to bolster the robustness of the aggregated model, we propose an extension module that incorporates a mechanism using adaptive distillation temperature to address the heterogeneity of user local data and a mechanism using adaptive weight to handle the variety in the quality of uploaded models. Finally, we conduct comprehensive experiments to illustrate the effectiveness of proposed approach.

Xiaoying Zhang, Da Peng, Yipeng Zhang et al.

Recent progress in (multimodal) large language models ((M)LLMs) has shifted focus from pre-training to inference-time computation and post-training optimization, largely due to concerns over the availability of high-quality human data. However, these strategies alone are insufficient to drive substantial model improvements. We argue that effective model advancement requires strong synergy among pre-training, inference-time computation, and post-training optimization. In this paper, we introduce Self-Improving cognition (SIcog), a self-learning framework for constructing next-generation foundation MLLMs by imparting multimodal knowledge and enhancing systematic cognitive capabilities through multimodal pre-training with self-generated data. Specifically, we propose Chain-of-Description for step-by-step visual understanding and integrate structured Chain-of-Thought (CoT) reasoning to support in-depth multimodal reasoning. SIcog first equips a base model with systematic perception and reasoning using minimal external supervision. The enhanced models then generate candidate image captions and CoT reasoning responses for unlabeled images and image-question pairs across diverse tasks, which are filtered through a semantic-similarity-guided self-consistency mechanism. These high-quality, self-generated samples enable large-scale multimodal pre-training, creating a self-improvement loop. Experiments demonstrate SIcog's effectiveness in developing MLLMs with enhanced multimodal cognition. Using only 213K self-generated pre-training samples, SIcog achieves significant improvements, including +3.6% on MMStar and +3.5% on AI2D, outperforming previous pre-training approaches. When combined with post-training techniques for CoT reasoning, SIcog yields +9% gains on MMVet and +8.5% on ScienceQA.

Ziyue Wang, Chi Chen, Yiqi Zhu et al. · tsinghua

With the bloom of Large Language Models (LLMs), Multimodal Large Language Models (MLLMs) that incorporate LLMs with pre-trained vision models have recently demonstrated impressive performance across diverse vision-language tasks. However, they fall short to comprehend context involving multiple images. A primary reason for this shortcoming is that the visual features for each images are encoded individually by frozen encoders before feeding into the LLM backbone, lacking awareness of other images and the multimodal instructions. We term this issue as prior-LLM modality isolation and propose a two phase paradigm, browse-and-concentrate, to enable in-depth multimodal context fusion prior to feeding the features into LLMs. This paradigm initially "browses" through the inputs for essential insights, and then revisits the inputs to "concentrate" on crucial details, guided by these insights, to achieve a more comprehensive understanding of the multimodal inputs. Additionally, we develop training strategies specifically to enhance the understanding of multi-image inputs. Our method markedly boosts the performance on 7 multi-image scenarios, contributing to increments on average accuracy by 2.13% and 7.60% against strong MLLMs baselines with 3B and 11B LLMs, respectively.

Yiyang Du, Xiaochen Wang, Chi Chen et al. · tsinghua

Recently, model merging methods have demonstrated powerful strengths in combining abilities on various tasks from multiple Large Language Models (LLMs). While previous model merging methods mainly focus on merging homogeneous models with identical architecture, they meet challenges when dealing with Multimodal Large Language Models (MLLMs) with inherent heterogeneous property, including differences in model architecture and the asymmetry in the parameter space. In this work, we propose AdaMMS, a novel model merging method tailored for heterogeneous MLLMs. Our method tackles the challenges in three steps: mapping, merging and searching. Specifically, we first design mapping function between models to apply model merging on MLLMs with different architecture. Then we apply linear interpolation on model weights to actively adapt the asymmetry in the heterogeneous MLLMs. Finally in the hyper-parameter searching step, we propose an unsupervised hyper-parameter selection method for model merging. As the first model merging method capable of merging heterogeneous MLLMs without labeled data, extensive experiments on various model combinations demonstrated that AdaMMS outperforms previous model merging methods on various vision-language benchmarks.

Zhongchen Deng, Zhechen Yang, Chi Chen et al.

Plane instance segmentation from RGB-D data is a crucial research topic for many downstream tasks. However, most existing deep-learning-based methods utilize only information within the RGB bands, neglecting the important role of the depth band in plane instance segmentation. Based on EfficientSAM, a fast version of SAM, we propose a plane instance segmentation network called PlaneSAM, which can fully integrate the information of the RGB bands (spectral bands) and the D band (geometric band), thereby improving the effectiveness of plane instance segmentation in a multimodal manner. Specifically, we use a dual-complexity backbone, with primarily the simpler branch learning D-band features and primarily the more complex branch learning RGB-band features. Consequently, the backbone can effectively learn D-band feature representations even when D-band training data is limited in scale, retain the powerful RGB-band feature representations of EfficientSAM, and allow the original backbone branch to be fine-tuned for the current task. To enhance the adaptability of our PlaneSAM to the RGB-D domain, we pretrain our dual-complexity backbone using the segment anything task on large-scale RGB-D data through a self-supervised pretraining strategy based on imperfect pseudo-labels. To support the segmentation of large planes, we optimize the loss function combination ratio of EfficientSAM. In addition, Faster R-CNN is used as a plane detector, and its predicted bounding boxes are fed into our dual-complexity network as prompts, thereby enabling fully automatic plane instance segmentation. Experimental results show that the proposed PlaneSAM sets a new SOTA performance on the ScanNet dataset, and outperforms previous SOTA approaches in zero-shot transfer on the 2D-3D-S, Matterport3D, and ICL-NUIM RGB-D datasets, while only incurring a 10% increase in computational overhead compared to EfficientSAM.

Houzhe Wang, Xiaojie Zhu, Chi Chen

With the increasing importance of data privacy and security, federated unlearning emerges as a new research field dedicated to ensuring that once specific data is deleted, federated learning models no longer retain or disclose related information. In this paper, we propose a zero-shot federated unlearning scheme, named Jellyfish. It distinguishes itself from conventional federated unlearning frameworks in four key aspects: synthetic data generation, knowledge disentanglement, loss function design, and model repair. To preserve the privacy of forgotten data, we design a zero-shot unlearning mechanism that generates error-minimization noise as proxy data for the data to be forgotten. To maintain model utility, we first propose a knowledge disentanglement mechanism that regularises the output of the final convolutional layer by restricting the number of activated channels for the data to be forgotten and encouraging activation sparsity. Next, we construct a comprehensive loss function that incorporates multiple components, including hard loss, confusion loss, distillation loss, model weight drift loss, gradient harmonization, and gradient masking, to effectively align the learning trajectories of the objectives of ``forgetting" and ``retaining". Finally, we propose a zero-shot repair mechanism that leverages proxy data to restore model accuracy within acceptable bounds without accessing users' local data. To evaluate the performance of the proposed zero-shot federated unlearning scheme, we conducted comprehensive experiments across diverse settings. The results validate the effectiveness and robustness of the scheme.

Ziyue Wang, Yurui Dong, Fuwen Luo et al.

The rapid advancing of Multimodal Large Language Models (MLLMs) has spurred interest in complex multimodal reasoning tasks in the real-world and virtual environment, which require coordinating multiple abilities, including visual perception, visual reasoning, spatial awareness, and target deduction. However, existing evaluations primarily assess the final task completion, often degrading assessments to isolated abilities such as visual grounding and visual question answering. Less attention is given to comprehensively and quantitatively analyzing reasoning process in multimodal environments, which is crucial for understanding model behaviors and underlying reasoning mechanisms beyond merely task success. To address this, we introduce MM-Escape, an extensible benchmark for investigating multimodal reasoning, inspired by real-world escape games. MM-Escape emphasizes intermediate model behaviors alongside final task completion. To achieve this, we develop EscapeCraft, a customizable and open environment that enables models to engage in free-form exploration for assessing multimodal reasoning. Extensive experiments show that MLLMs, regardless of scale, can successfully complete the simplest room escape tasks, with some exhibiting human-like exploration strategies. Yet, performance dramatically drops as task difficulty increases. Moreover, we observe that performance bottlenecks vary across models, revealing distinct failure modes and limitations in their multimodal reasoning abilities, such as repetitive trajectories without adaptive exploration, getting stuck in corners due to poor visual spatial awareness, and ineffective use of acquired props, such as the key. We hope our work sheds light on new challenges in multimodal reasoning, and uncovers potential improvements in MLLMs capabilities.

Yichen Zhang, Da Peng, Zonghao Guo et al.

A recent cutting-edge topic in multimodal modeling is to unify visual comprehension and generation within a single model. However, the two tasks demand mismatched decoding regimes and visual representations, making it non-trivial to jointly optimize within a shared feature space. In this work, we present Cheers, a unified multimodal model that decouples patch-level details from semantic representations, thereby stabilizing semantics for multimodal understanding and improving fidelity for image generation via gated detail residuals. Cheers includes three key components: (i) a unified vision tokenizer that encodes and compresses image latent states into semantic tokens for efficient LLM conditioning, (ii) an LLM-based Transformer that unifies autoregressive decoding for text generation and diffusion decoding for image generation, and (iii) a cascaded flow matching head that decodes visual semantics first and then injects semantically gated detail residuals from the vision tokenizer to refine high-frequency content. Experiments on popular benchmarks demonstrate that Cheers matches or surpasses advanced UMMs in both visual understanding and generation. Cheers also achieves 4x token compression, enabling more efficient high-resolution image encoding and generation. Notably, Cheers outperforms the Tar-1.5B on the popular benchmarks GenEval and MMBench, while requiring only 20% of the training cost, indicating effective and efficient (i.e., 4x token compression) unified multimodal modeling. We will release all code and data for future research.

Houzhe Wang, Xiaojie Zhu, Chi Chen

With the increasing importance of data privacy and security, federated unlearning has emerged as a novel research field dedicated to ensuring that federated learning models no longer retain or leak relevant information once specific data has been deleted. In this paper, to the best of our knowledge, we propose the first complete pipeline for federated unlearning, which includes a federated unlearning approach and an evaluation framework. Our proposed federated unlearning approach ensures high efficiency and model accuracy without the need to store historical data.It effectively leverages the knowledge distillation model alongside various optimization mechanisms. Moreover, we propose a framework named Skyeye to visualize the forgetting capacity of federated unlearning models. It utilizes the federated unlearning model as the classifier integrated into a Generative Adversarial Network (GAN). Afterward, both the classifier and discriminator guide the generator in generating samples. Throughout this process, the generator learns from the classifier's knowledge. The generator then visualizes this knowledge through sample generation. Finally, the model's forgetting capability is evaluated based on the relevance between the deleted data and the generated samples. Comprehensive experiments are conducted to illustrate the effectiveness of the proposed federated unlearning approach and the corresponding evaluation framework.

Cheng Zeng, Xiatian Qi, Chi Chen et al.

Transformers have been seldom employed in point cloud roof plane instance segmentation, which is the focus of this study, and existing superpoint Transformers suffer from limited performance due to the use of low-quality superpoints. To address this challenge, we establish two criteria that high-quality superpoints for Transformers should satisfy and introduce a corresponding two-stage superpoint generation process. The superpoints generated by our method not only have accurate boundaries, but also exhibit consistent geometric sizes and shapes, both of which greatly benefit the feature learning of superpoint Transformers. To compensate for the limitations of deep learning features when the training set size is limited, we incorporate multidimensional handcrafted features into the model. Additionally, we design a decoder that combines a Kolmogorov-Arnold Network with a Transformer module to improve instance prediction and mask extraction. Finally, our network's predictions are refined using traditional algorithm-based postprocessing. For evaluation, we annotated a real-world dataset and corrected annotation errors in the existing RoofN3D dataset. Experimental results show that our method achieves state-of-the-art performance on our dataset, as well as both the original and reannotated RoofN3D datasets. Moreover, our model is not sensitive to plane boundary annotations during training, significantly reducing the annotation burden. Through comprehensive experiments, we also identified key factors influencing roof plane segmentation performance: in addition to roof types, variations in point cloud density, density uniformity, and 3D point precision have a considerable impact. These findings underscore the importance of incorporating data augmentation strategies that account for point cloud quality to enhance model robustness under diverse and challenging conditions.

Yubo Sun, Chunyi Peng, Yukun Yan et al.

Visual retrieval-augmented generation (VRAG) augments vision-language models (VLMs) with external visual knowledge to ground reasoning and reduce hallucinations. Yet current VRAG systems often fail to reliably perceive and integrate evidence across multiple images, leading to weak grounding and erroneous conclusions. In this paper, we propose EVisRAG, an end-to-end framework that learns to reason with evidence-guided multi-image to address this issue. The model first observes retrieved images and records per-image evidence, then derives the final answer from the aggregated evidence. To train EVisRAG effectively, we introduce Reward-Scoped Group Relative Policy Optimization (RS-GRPO), which binds fine-grained rewards to scope-specific tokens to jointly optimize visual perception and reasoning abilities of VLMs. Experimental results on multiple visual question answering benchmarks demonstrate that EVisRAG delivers substantial end-to-end gains over backbone VLM with 27\% improvements on average. Further analysis shows that, powered by RS-GRPO, EVisRAG improves answer accuracy by precisely perceiving and localizing question-relevant evidence across multiple images and deriving the final answer from that evidence, much like a real detective.

Shiji Zhao, Chi Chen, Ranjie Duan et al.

Adversarial Training (AT) is widely recognized as an effective approach to enhance the adversarial robustness of Deep Neural Networks. As a variant of AT, Adversarial Robustness Distillation (ARD) has shown outstanding performance in enhancing the robustness of small models. However, both AT and ARD face robust fairness issue: these models tend to display strong adversarial robustness against some classes (easy classes) while demonstrating weak adversarial robustness against others (hard classes). This paper explores the underlying factors of this problem and points out the smoothness degree of soft labels for different classes significantly impacts the robust fairness from both empirical observation and theoretical analysis. Based on the above exploration, we propose Anti-Bias Soft Label Distillation (ABSLD) within the Knowledge Distillation framework to enhance the adversarial robust fairness. Specifically, ABSLD adaptively reduces the student's error risk gap between different classes, which is accomplished by adjusting the class-wise smoothness degree of teacher's soft labels during the training process, and the adjustment is managed by assigning varying temperatures to different classes. Additionally, as a label-based approach, ABSLD is highly adaptable and can be integrated with the sample-based methods. Extensive experiments demonstrate ABSLD outperforms state-of-the-art methods on the comprehensive performance of robustness and fairness.

Xingfei Wei, Qiankun Mo, Chi Chen et al.

Artificial intelligence (AI) models remain an emerging strategy to accelerate materials design and development. We demonstrate that convolutional neural network (CNN) models can characterize DNA origami nanostructures employed in programmable self-assembling, which is important in many applications such as in biomedicine. Specifically, we benchmark the performance of 9 CNN models -- viz. AlexNet, GoogLeNet, VGG16, VGG19, ResNet18, ResNet34, ResNet50, ResNet101, and ResNet152 -- to characterize the ligation number of DNA origami nanostructures in transmission electron microscopy (TEM) images. We first pre-train CNN models using a large image dataset of 720 images from our coarse-grained (CG) molecular dynamics (MD) simulations. Then, we fine-tune the pre-trained CNN models, using a small experimental TEM dataset with 146 TEM images. All CNN models were found to have similar computational time requirements, while their model sizes and performances are different. We use 20 test MD images to demonstrate that among all of the pre-trained CNN models ResNet50 and VGG16 have the highest and second highest accuracies. Among the fine-tuned models, VGG16 was found to have the highest agreement on the test TEM images. Thus, we conclude that fine-tuned VGG16 models can quickly characterize the ligation number of nanostructures in large TEM images.

Yipeng Zhang, Yifan Liu, Zonghao Guo et al.

Vision transformers (ViTs) are widely employed in multimodal large language models (MLLMs) for visual encoding. However, they exhibit inferior performance on tasks regarding fine-grained visual perception. We attribute this to the limitations of ViTs in capturing diverse multi-modal visual levels, such as low-level details. To address this issue, we present LLaVA-UHD v2, an MLLM with advanced perception abilities by introducing a well-designed vision-language projector, the Hierarchical window (Hiwin) transformer. Hiwin transformer enhances MLLM's ability to capture diverse multi-modal visual granularities, by incorporating our constructed high-resolution semantic pyramid. Specifically, Hiwin transformer comprises two key modules: (i) a visual detail injection module, which progressively injects low-level visual details into high-level language-aligned semantics features, thereby forming an inverse semantic pyramid (ISP), and (ii) a hierarchical window attention module, which leverages cross-scale windows to condense multi-level semantics from the ISP. Extensive experiments show that LLaVA-UHD v2 outperforms compared MLLMs on a wide range of benchmarks. Notably, our design achieves an average boost of 3.7% across 14 benchmarks compared with the baseline method, 9.3% on DocVQA for instance. All the data and code will be publicly available to facilitate future research.

Chi Chen, Peng Li, Maosong Sun et al.

Weakly supervised vision-and-language pre-training (WVLP), which learns cross-modal representations with limited cross-modal supervision, has been shown to effectively reduce the data cost of pre-training while maintaining decent performance on downstream tasks. However, current WVLP methods use only local descriptions of images, i.e., object tags, as cross-modal anchors to construct weakly-aligned image-text pairs for pre-training. This affects the data quality and thus the effectiveness of pre-training. In this paper, we propose to directly take a small number of aligned image-text pairs as anchors, and represent each unaligned image and text by its similarities to these anchors, i.e., relative representations. We build a WVLP framework based on the relative representations, namely RELIT, which collects high-quality weakly-aligned image-text pairs from large-scale image-only and text-only data for pre-training through relative representation-based retrieval and generation. Experiments on four downstream tasks show that RELIT achieves new state-of-the-art results under the weakly supervised setting.

Ruiqi Ma, Chi Chen, Bisheng Yang et al.

At present, the anchor-based or anchor-free models that use LiDAR point clouds for 3D object detection use the center assigner strategy to infer the 3D bounding boxes. However, in a real world scene, the LiDAR can only acquire a limited object surface point clouds, but the center point of the object does not exist. Obtaining the object by aggregating the incomplete surface point clouds will bring a loss of accuracy in direction and dimension estimation. To address this problem, we propose a corner-guided anchor-free single-stage 3D object detection model (CG-SSD ).Firstly, 3D sparse convolution backbone network composed of residual layers and sub-manifold sparse convolutional layers are used to construct bird's eye view (BEV) features for further deeper feature mining by a lite U-shaped network; Secondly, a novel corner-guided auxiliary module (CGAM) is proposed to incorporate corner supervision signals into the neural network. CGAM is explicitly designed and trained to detect partially visible and invisible corners to obtains a more accurate object feature representation, especially for small or partial occluded objects; Finally, the deep features from both the backbone networks and CGAM module are concatenated and fed into the head module to predict the classification and 3D bounding boxes of the objects in the scene. The experiments demonstrate CG-SSD achieves the state-of-art performance on the ONCE benchmark for supervised 3D object detection using single frame point cloud data, with 62.77%mAP. Additionally, the experiments on ONCE and Waymo Open Dataset show that CGAM can be extended to most anchor-based models which use the BEV feature to detect objects, as a plug-in and bring +1.17%-+14.27%AP improvement.

Xin Wang, Xin Peng, Jun Sun et al.

Code summarization is the task of generating natural language description of source code, which is important for program understanding and maintenance. Existing approaches treat the task as a machine translation problem (e.g., from Java to English) and applied Neural Machine Translation models to solve the problem. These approaches only consider a given code unit (e.g., a method) without its broader context. The lacking of context may hinder the NMT model from gathering sufficient information for code summarization. Furthermore, existing approaches use a fixed vocabulary and do not fully consider the words in code, while many words in the code summary may come from the code. In this work, we present a neural network model named ToPNN for code summarization, which uses the topics in a broader context (e.g., class) to guide the neural networks that combine the generation of new words and the copy of existing words in code. Based on the model we present an approach for generating natural language code summaries at the method level (i.e., method comments). We evaluate our approach using a dataset with 4,203,565 commented Java methods. The results show significant improvement over state-of-the-art approaches and confirm the positive effect of class topics and the copy mechanism.

Chi Chen, Maosong Sun, Yang Liu

Word alignment, which aims to align translationally equivalent words between source and target sentences, plays an important role in many natural language processing tasks. Current unsupervised neural alignment methods focus on inducing alignments from neural machine translation models, which does not leverage the full context in the target sequence. In this paper, we propose Mask-Align, a self-supervised word alignment model that takes advantage of the full context on the target side. Our model masks out each target token and predicts it conditioned on both source and the remaining target tokens. This two-step process is based on the assumption that the source token contributing most to recovering the masked target token should be aligned. We also introduce an attention variant called leaky attention, which alleviates the problem of unexpected high cross-attention weights on special tokens such as periods. Experiments on four language pairs show that our model outperforms previous unsupervised neural aligners and obtains new state-of-the-art results.