arXiv subjects

Chen Sun

Publications and source records attributed to Chen Sun.

At least 73 records · Page 4Linked to original sources

arXiv2025

SUPERChem: A Multimodal Reasoning Benchmark in Chemistry

Current benchmarks for evaluating the chemical reasoning capabilities of Large Language Models (LLMs) are limited by oversimplified tasks, lack of process-level evaluation, and misalignment with expert-level chemistry skills. To address these issues, we introduce SUPERChem, a benchmark of 500 expert-curated reasoning-intensive chemistry problems, covering diverse subfields and provided in both multimodal and text-only formats. Original content and an iterative curation pipeline eliminate flawed items and mitigate data contamination. Each problem is paired with an expert-authored solution path, enabling Reasoning Path Fidelity (RPF) scoring to evaluate reasoning quality beyond final-answer accuracy. Evaluations against a human baseline of 40.3% accuracy show that even the best-performing model, GPT-5 (High), reaches only 38.5%, followed closely by Gemini 2.5 Pro (37.9%) and DeepSeek-V3.1-Think (37.3%). SUPERChem elicits multi-step, multimodal reasoning, reveals model-dependent effects of visual information, and distinguishes high-fidelity reasoners from heuristic ones. By providing a challenging benchmark and a reliable evaluation framework, SUPERChem aims to facilitate the advancement of LLMs toward expert-level chemical intelligence. The dataset of the benchmark is available at https://huggingface.co/datasets/ZehuaZhao/SUPERChem.

cs.CL↗

arXiv2025

Tool-RoCo: An Agent-as-Tool Self-organization Large Language Model Benchmark in Multi-robot Cooperation

This study proposes Tool-RoCo, a novel benchmark for evaluating large language models (LLMs) in long-term multi-agent cooperation based on RoCo, a multi-robot cooperative benchmark. Recent research on LLM-based multi-agent systems has relied on predefined orchestration, while ignoring agent autonomy. Tool-RoCo treats other agents as tools and introduces cooperative tools, leveraging tool usage to evaluate multi-agent cooperation and self-organization. Tool usage means that each agent (LLM) selects a tool from a candidate set based on the current state, receives feedback, and adjusts its selection in subsequent rounds. To evaluate different autonomy levels, we propose four LLM paradigms: (1) centralized cooperation, where a single LLM allocates tools to all agents; (2) centralized self-organization, where a central LLM autonomously activates agents while keeping others inactive; (3) decentralized cooperation, where each agent has its own LLM and calls tools based on local information; and (4) self-organization, where a randomly chosen initial agent can request collaboration, activating additional agents via tool calls. Tool-RoCo includes three multi-robot tasks, SORT, PACK, and CABINET, to measure format and parameter accuracy and agent coordination through tool usage. The results using several LLMs showed that cooperative tools accounted for only 7.09% of all tools, indicating that LLM-based agents rarely invoked others as assistants. Moreover, activation tools accounted for 96.42%, suggesting that current LLMs tend to maintain active agents while seldom deactivating them for adaptive coordination. Tool-RoCo provides a systematic benchmark to evaluate LLM autonomy and cooperation in multi-agent tasks. Code and Demo: https://github.com/ColaZhang22/Tool-Roco

cs.MA↗

arXiv2025

Spacewalk-18: A Benchmark for Multimodal and Long-form Procedural Video Understanding in Novel Domains

Learning from (procedural) videos has increasingly served as a pathway for embodied agents to acquire skills from human demonstrations. To do this, video understanding models must be able to obtain structured understandings, such as the temporal segmentation of a demonstration into sequences of actions and skills, and to generalize the understandings to novel environments, tasks, and problem domains. In pursuit of this goal, we introduce Spacewalk-18, a benchmark containing two tasks: (1) step recognition and (2) video question answering, over a dataset of temporally segmented and labeled tasks in International Space Station spacewalk recordings. In tandem, the two tasks quantify a model's ability to: (1) generalize to novel domains; (2) utilize long temporal context and multimodal (e.g. visual and speech) information. Our extensive experimental analysis highlights the challenges of Spacewalk-18, but also suggests best practices for domain generalization and long-form understanding. Notably, we discover a promising adaptation via summarization technique that leads to significant performance improvement without model fine-tuning. The Spacewalk-18 benchmark is released at https://brown-palm.github.io/Spacewalk-18/.

cs.CV↗

arXiv2025

Force Prompting: Video Generation Models Can Learn and Generalize Physics-based Control Signals

Recent advances in video generation models have sparked interest in world models capable of simulating realistic environments. While navigation has been well-explored, physically meaningful interactions that mimic real-world forces remain largely understudied. In this work, we investigate using physical forces as a control signal for video generation and propose force prompts which enable users to interact with images through both localized point forces, such as poking a plant, and global wind force fields, such as wind blowing on fabric. We demonstrate that these force prompts can enable videos to respond realistically to physical control signals by leveraging the visual and motion prior in the original pretrained model, without using any 3D asset or physics simulator at inference. The primary challenge of force prompting is the difficulty in obtaining high quality paired force-video training data, both in the real world due to the difficulty of obtaining force signals, and in synthetic data due to limitations in the visual quality and domain diversity of physics simulators. Our key finding is that video generation models can generalize remarkably well when adapted to follow physical force conditioning from videos synthesized by Blender, even with limited demonstrations of few objects. Our method can generate videos which simulate forces across diverse geometries, settings, and materials. We also try to understand the source of this generalization and perform ablations that reveal two key elements: visual diversity and the use of specific text keywords during training. Our approach is trained on only around 15k training examples for a single day on four A100 GPUs, and outperforms existing methods on force adherence and physics realism, bringing world models closer to real-world physics interactions. We release all datasets, code, weights, and interactive video demos at our project page.

cs.CV↗

arXiv2025

Pulse and Polarization Structures in Axion-Converted X-rays from Pulsars

Neutron stars (NS's) with their strong magnetic fields and hot dense cores could be powerful probes of axions, a classic benchmark of feebly-coupled new particles, through abundant production of axions with the axion-nucleon coupling and subsequent conversion into X-rays due to the axion-photon coupling. In this article, we point out that the pulsation structures in both the intensity and polarization of X-rays from NS's could provide us additional information about axions and their couplings. We develop new analytical formalisms of pulsation-polarization structure applicable to a wide range of NS's in the axion scenario and argue that they hold in complicated astrophysical environments. As a case study, we apply our formalism to a representative X-ray Dim Isolated Neutron Star, RX J1856.6-3754, with an unexpected hard X-ray excess which might be axion-induced. We show with an updated fit that the axion explanation is compatible with both the intensity and pulsation data available, and combining the pulsation data does not shift the posterior by more than $1\,σ$. Yet, the preferred parameter space is close to being excluded by other astrophysical constraints. With a 75% reduction of the uncertainties in the pulsation data, we could potentially draw a definite conclusion on the axion-induced X-rays at more than $3\,σ$ level.

hep-ph↗

arXiv2025

Multi-Timescale Hierarchical Reinforcement Learning for Unified Behavior and Control of Autonomous Driving

Reinforcement Learning (RL) is increasingly used in autonomous driving (AD) and shows clear advantages. However, most RL-based AD methods overlook policy structure design. An RL policy that only outputs short-timescale vehicle control commands results in fluctuating driving behavior due to fluctuations in network outputs, while one that only outputs long-timescale driving goals cannot achieve unified optimality of driving behavior and control. Therefore, we propose a multi-timescale hierarchical reinforcement learning approach. Our approach adopts a hierarchical policy structure, where high- and low-level RL policies are unified-trained to produce long-timescale motion guidance and short-timescale control commands, respectively. Therein, motion guidance is explicitly represented by hybrid actions to capture multimodal driving behaviors on structured road and support incremental low-level extend-state updates. Additionally, a hierarchical safety mechanism is designed to ensure multi-timescale safety. Evaluation in simulator-based and HighD dataset-based highway multi-lane scenarios demonstrates that our approach significantly improves AD performance, effectively increasing driving efficiency, action consistency and safety.

cs.RO↗

arXiv2025

DiffMark: Diffusion-based Robust Watermark Against Deepfakes

Deepfakes pose significant security and privacy threats through malicious facial manipulations. While robust watermarking can aid in authenticity verification and source tracking, existing methods often lack the sufficient robustness against Deepfake manipulations. Diffusion models have demonstrated remarkable performance in image generation, enabling the seamless fusion of watermark with image during generation. In this study, we propose a novel robust watermarking framework based on diffusion model, called DiffMark. By modifying the training and sampling scheme, we take the facial image and watermark as conditions to guide the diffusion model to progressively denoise and generate corresponding watermarked image. In the construction of facial condition, we weight the facial image by a timestep-dependent factor that gradually reduces the guidance intensity with the decrease of noise, thus better adapting to the sampling process of diffusion model. To achieve the fusion of watermark condition, we introduce a cross information fusion (CIF) module that leverages a learnable embedding table to adaptively extract watermark features and integrates them with image features via cross-attention. To enhance the robustness of the watermark against Deepfake manipulations, we integrate a frozen autoencoder during training phase to simulate Deepfake manipulations. Additionally, we introduce Deepfake-resistant guidance that employs specific Deepfake model to adversarially guide the diffusion sampling process to generate more robust watermarked images. Experimental results demonstrate the effectiveness of the proposed DiffMark on typical Deepfakes. Our code will be available at https://github.com/vpsg-research/DiffMark.

cs.CV↗

arXiv2025

DRCP: Diffusion on Reinforced Cooperative Perception for Perceiving Beyond Limits

Cooperative perception enabled by Vehicle-to-Everything communication has shown great promise in enhancing situational awareness for autonomous vehicles and other mobile robotic platforms. Despite recent advances in perception backbones and multi-agent fusion, real-world deployments remain challenged by hard detection cases, exemplified by partial detections and noise accumulation which limit downstream detection accuracy. This work presents Diffusion on Reinforced Cooperative Perception (DRCP), a real-time deployable framework designed to address aforementioned issues in dynamic driving environments. DRCP integrates two key components: (1) Precise-Pyramid-Cross-Modality-Cross-Agent, a cross-modal cooperative perception module that leverages camera-intrinsic-aware angular partitioning for attention-based fusion and adaptive convolution to better exploit external features; and (2) Mask-Diffusion-Mask-Aggregation, a novel lightweight diffusion-based refinement module that encourages robustness against feature perturbations and aligns bird's-eye-view features closer to the task-optimal manifold. The proposed system achieves real-time performance on mobile platforms while significantly improving robustness under challenging conditions. Code will be released in late 2025.

cs.RO↗

arXiv2025

$100K or 100 Days: Trade-offs when Pre-Training with Academic Resources

Pre-training is notoriously compute-intensive and academic researchers are notoriously under-resourced. It is, therefore, commonly assumed that academics can't pre-train models. In this paper, we seek to clarify this assumption. We first survey academic researchers to learn about their available compute and then empirically measure the time to replicate models on such resources. We introduce a benchmark to measure the time to pre-train models on given GPUs and also identify ideal settings for maximizing training speed. We run our benchmark on a range of models and academic GPUs, spending 2,000 GPU-hours on our experiments. Our results reveal a brighter picture for academic pre-training: for example, although Pythia-1B was originally trained on 64 GPUs for 3 days, we find it is also possible to replicate this model (with the same hyper-parameters) in 3x fewer GPU-days: i.e. on 4 GPUs in 18 days. We conclude with a cost-benefit analysis to help clarify the trade-offs between price and pre-training time. We believe our benchmark will help academic researchers conduct experiments that require training larger models on more data. We fully release our codebase at: https://github.com/apoorvkh/academic-pretraining.

cs.CL↗

arXiv2025

RG-Attn: Radian Glue Attention for Multi-modality Multi-agent Cooperative Perception

Cooperative perception enhances autonomous driving by leveraging Vehicle-to-Everything (V2X) communication for multi-agent sensor fusion. However, most existing methods rely on single-modal data sharing, limiting fusion performance, particularly in heterogeneous sensor settings involving both LiDAR and cameras across vehicles and roadside units (RSUs). To address this, we propose Radian Glue Attention (RG-Attn), a lightweight and generalizable cross-modal fusion module that unifies intra-agent and inter-agent fusion via transformation-based coordinate alignment and a unified sampling/inversion strategy. RG-Attn efficiently aligns features through a radian-based attention constraint, operating column-wise on geometrically consistent regions to reduce overhead and preserve spatial coherence, thereby enabling accurate and robust fusion. Building upon RG-Attn, we propose three cooperative architectures. The first, Paint-To-Puzzle (PTP), prioritizes communication efficiency but assumes all agents have LiDAR, optionally paired with cameras. The second, Co-Sketching-Co-Coloring (CoS-CoCo), offers maximal flexibility, supporting any sensor setup (e.g., LiDAR-only, camera-only, or both) and enabling strong cross-modal generalization for real-world deployment. The third, Pyramid-RG-Attn Fusion (PRGAF), aims for peak detection accuracy with the highest computational overhead. Extensive evaluations on simulated and real-world datasets show our framework delivers state-of-the-art detection accuracy with high flexibility and efficiency. GitHub Link: https://github.com/LantaoLi/RG-Attn

cs.RO↗

arXiv2025

DriveSOTIF: Advancing Perception SOTIF Through Multimodal Large Language Models

Human drivers possess spatial and causal intelligence, enabling them to perceive driving scenarios, anticipate hazards, and react to dynamic environments. In contrast, autonomous vehicles lack these abilities, making it challenging to manage perception-related Safety of the Intended Functionality (SOTIF) risks, especially under complex or unpredictable driving conditions. To address this gap, we propose fine-tuning multimodal large language models (MLLMs) on a customized dataset specifically designed to capture perception-related SOTIF scenarios. Benchmarking results show that fine-tuned MLLMs achieve an 11.8\% improvement in close-ended VQA accuracy and a 12.0\% increase in open-ended VQA scores compared to baseline models, while maintaining real-time performance with a 0.59-second average inference time per image. We validate our approach through real-world case studies in Canada and China, where fine-tuned models correctly identify safety risks that challenge even experienced human drivers. This work represents the first application of domain-specific MLLM fine-tuning for SOTIF domain in autonomous driving. The dataset and related resources are available at github.com/s95huang/DriveSOTIF.git

cs.RO↗

arXiv2025

Know What, Know Why: Semantic Hazard Communication for Intelligent V2X Systems

In current vehicle-to-everything (V2X) communication systems, roadside units (RSUs) broadcast brief warning messages that alert nearby vehicles to avoid potential hazards. However, these messages lack contextual information on why a warning is issued, leading to excessive caution or inefficient driving behaviors. To avoid such a situation, we propose a semantic-enhanced and explainable V2X (SEE-V2X) system. In the proposed system, RSUs equipped with smart cameras detect obstructions and transmit context-aware messages to vehicles. By understanding both what the hazard is and why it occurs, drivers can make more intelligent decisions based on their specific driving situation. Furthermore, through a real-field demonstration, we show the new "see-through" feature in the proposed system, which enables drivers to visualize hidden pedestrians behind obstacles. We also perform simulations to compare traditional V2X with SEE-V2X under different traffic conditions. The results show that SEE-V2X significantly improves traffic efficiency and reduces unnecessary deceleration.

eess.SP↗

arXiv2025

Macroscopic Ferromagnetic Dynamics

In metals with finite magnetization $\vec{M}$, experiment shows that transverse polarized dc spin currents $\vec{J}_{i}$ both decay and precess on crossing a finite sample thickness. The present work uses Onsager's irreversible thermodynamics, with $\vec{M}$ and $\vec{J}_{i}$ as fundamental variables, to develop a theory with aspects of the Landau-Lifshitz theory solely for the $\vec{M}$ of (charged) electronic ferromagnets, and of the Leggett theory for the $\vec{M}$ and $\vec{J}_{i}$ of (uncharged) nuclear paramagnets. As for the ferromagnet of Landau-Lifshitz, $\partial_{t}\vec{M}$ includes a characteristic decay time $τ_{M}$. As for the nuclear paramagnet, $\partial_{t}\vec{J}_{i}$ includes a characteristic decay time $τ_{J}$, is driven by the gradient of a (vector) spin pressure, and precesses about a mean-field proportional to $\vec{M}$. The spin pressure has a coefficient $G$ proportional to a velocity squared, and $D_{0}\equiv \frac{1}{2}Gτ_{J}$ serves as an effective diffusion coefficient. These equations apply when spin currents are generated. Using the derived dynamical equations for the magnetization and for the spin current, we obtain the steady state (dc limit) solution whose transverse wavevector squared is complex, with real part from diffusion and imaginary part from precession. The ac case is also considered.

cond-mat.mes-hall↗

arXiv2025

Photocatalytic CO2 Reduction Enhanced by Synergetic Interactions among Photon Phonon and Molecule

Photocatalytic CO2 reduction is limited by inefficient CO2 activation and poor solar spectrum utilization. Here, we discovered and revealed the vibration coupling mechanism among photons, phonons, and molecules, which remarkably enhances photocatalytic catalysis of CO2 into fuels. We designed the nitrogen-doped Cu2O-based catalyst loaded onto the quartz optical substrate. The N-doping Cu2O converts linearly geometry of adsorbed CO2 molecules, which efficiently lowers the activation barrier and facilitates CO2 dissociation. Once the Cu-based catalyst is combined with a micro-pillar quartz film, the system induces vibrational strong coupling (VSC) between the asymmetric CO2 stretching mode and surface phonon polariton resonances. These resonances arise from the photothermal conversion of incident solar photons on the micro-pillars. The resonant coupling phenomena were further verified by Fourier-transform infrared spectroscopy using Synchrotron Radiation Source (SRS), which directly confirmed the interactions between molecular vibrations and photonic-phononic modes. The synergetic functions originated from this hybrid architecture achieve a CO yield of 167.7 umol h-1 g-1 under pure water conditions, which is the highest reported yield for Cu2O-based photocatalysts with 46% enhancement over non-VSC systems. This work uncovers a novel photo-thermal mechanism. It further provides a new strategy to control bond activation in photocatalytic CO2 conversion through light-vibration-matter coupling.

cond-mat.mtrl-sci↗

arXiv2025

Derivation of the Landau-Zener formula via functional equations

The Landau-Zener formula describes the diabatic transition probability of a two-level system under linear driving. Its rigorous derivation typically relies on sophisticated mathematical tools, such as special functions, Laplace transforms, or contour integrals. In this work, we present a derivation of the Landau-Zener transition probability using a fundamentally different approach via functional equations. By leveraging integrability, we prove that this transition probability satisfies a functional equation, whose solutions establish the exponential form of the formula. The coefficient in the exponent is then determined through a lowest-order perturbation calculation. This derivation is rigorous and does not involve any sophisticated mathematics. Our work provides insights into the origin of the exponential form of the Landau-Zener transition probability, and shows that the Landau-Zener formula can be viewed as a consequence of integrability, though the two-level Landau-Zener Hamiltonian itself does not satisfy the integrability conditions.

quant-ph↗

arXiv2025

Less Signals, More Understanding: Channel-Capacity Codebook Design for Digital Task-Oriented Semantic Communication

Discrete representation has emerged as a powerful tool in task-oriented semantic communication (ToSC), offering compact, interpretable, and efficient representations well-suited for low-power edge intelligence scenarios. Its inherent digital nature aligns seamlessly with hardware-friendly deployment and robust storage/transmission protocols. However, despite its strengths, current ToSC frameworks often decouple semantic-aware discrete mapping from the underlying channel characteristics and task demands. This mismatch leads to suboptimal communication performance, degraded task utility, and limited generalization under variable wireless conditions. Moreover, conventional designs frequently overlook channel-awareness in codebook construction, restricting the effectiveness of semantic symbol selection under constrained resources. To address these limitations, this paper proposes a channel-aware discrete semantic coding framework tailored for low-power edge networks. Leveraging a Wasserstein-regularized objective, our approach aligns discrete code activations with optimal input distributions, thereby improving semantic fidelity, robustness, and task accuracy. Extensive experiments on the inference tasks across diverse signal-to-noise ratio (SNR) regimes show that our method achieves notable gains in accuracy and communication efficiency. This work provides new insights into integrating discrete semantics and channel optimization, paving the way for the widespread adoption of semantic communication in future digital infrastructures.

eess.SP↗

arXiv2025

How Can Objects Help Video-Language Understanding?

Do we still need to represent objects explicitly in multimodal large language models (MLLMs)? To one extreme, pre-trained encoders convert images into visual tokens, with which objects and spatiotemporal relationships may be implicitly modeled. To the other extreme, image captions by themselves provide strong empirical performances for understanding tasks, despite missing fine-grained spatiotemporal information. To answer this question, we introduce ObjectMLLM, a framework capable of leveraging arbitrary computer vision algorithm to extract and integrate structured visual representation. Through extensive evaluations on six video question answering benchmarks, we confirm that explicit integration of object-centric representation remains necessary. Surprisingly, we observe that the simple approach of quantizing the continuous, structured object information and representing them as plain text performs the best, offering a data-efficient approach to integrate other visual perception modules into MLLM design. Our code and models are released at https://github.com/brown-palm/ObjectMLLM.

cs.CV↗

arXiv2025

What is an "Abstract Reasoner"? Revisiting Experiments and Arguments about Large Language Models

Recent work has argued that large language models (LLMs) are not "abstract reasoners", citing their poor zero-shot performance on a variety of challenging tasks as evidence. We revisit these experiments in order to add nuance to the claim. First, we show that while LLMs indeed perform poorly in a zero-shot setting, even tuning a small subset of parameters for input encoding can enable near-perfect performance. However, we also show that this finetuning does not necessarily transfer across datasets. We take this collection of empirical results as an invitation to (re-)open the discussion of what it means to be an "abstract reasoner", and why it matters whether LLMs fit the bill.

cs.CL↗