arXiv subjects

Dongzhan Zhou

Publications and source records attributed to Dongzhan Zhou.

At least 55 records · Page 3Linked to original sources

arXiv2026

MiST: Understanding the Role of Mid-Stage Scientific Training in Developing Chemical Reasoning Models

Large Language Models can develop reasoning capabilities through online fine-tuning with rule-based rewards. However, recent studies reveal a critical constraint: reinforcement learning succeeds only when the base model already assigns non-negligible probability to correct answers -- a property we term 'latent solvability'. This work investigates the emergence of chemical reasoning capabilities and what these prerequisites mean for chemistry. We identify two necessary conditions for RL-based chemical reasoning: 1) Symbolic competence, and 2) Latent chemical knowledge. We propose mid-stage scientific training (MiST): a set of mid-stage training techniques to satisfy these, including data-mixing with SMILES/CIF-aware pre-processing, continued pre-training on 2.9B tokens, and supervised fine-tuning on 1B tokens. These steps raise the latent-solvability score on 3B and 7B models by up to 1.8x, and enable RL to lift top-1 accuracy from 10.9 to 63.9% on organic reaction naming, and from 40.6 to 67.4% on inorganic material generation. Similar results are observed for other challenging chemical tasks, while producing interpretable reasoning traces. Our results define clear prerequisites for chemical reasoning training and highlight the broader role of mid-stage training in unlocking reasoning capabilities.

cs.LG↗

arXiv2026

Beyond Static Tools: Test-Time Tool Evolution for Scientific Reasoning

The central challenge of AI for Science is not reasoning alone, but the ability to create computational methods in an open-ended scientific world. Existing LLM-based agents rely on static, pre-defined tool libraries, a paradigm that fundamentally fails in scientific domains where tools are sparse, heterogeneous, and intrinsically incomplete. In this paper, we propose Test-Time Tool Evolution (TTE), a new paradigm that enables agents to synthesize, verify, and evolve executable tools during inference. By transforming tools from fixed resources into problem-driven artifacts, TTE overcomes the rigidity and long-tail limitations of static tool libraries. To facilitate rigorous evaluation, we introduce SciEvo, a benchmark comprising 1,590 scientific reasoning tasks supported by 925 automatically evolved tools. Extensive experiments show that TTE achieves state-of-the-art performance in both accuracy and tool efficiency, while enabling effective cross-domain adaptation of computational tools. The code and benchmark have been released at https://github.com/lujiaxuan0520/Test-Time-Tool-Evol.

cs.AI↗

arXiv2026

SciEvalKit: An Open-source Evaluation Toolkit for Scientific General Intelligence

We introduce SciEvalKit, a unified benchmarking toolkit designed to evaluate AI models for science across a broad range of scientific disciplines and task capabilities. Unlike general-purpose evaluation platforms, SciEvalKit focuses on the core competencies of scientific intelligence, including Scientific Multimodal Perception, Scientific Multimodal Reasoning, Scientific Multimodal Understanding, Scientific Symbolic Reasoning, Scientific Code Generation, Science Hypothesis Generation and Scientific Knowledge Understanding. It supports six major scientific domains, spanning from physics and chemistry to astronomy and materials science. SciEvalKit builds a foundation of expert-grade scientific benchmarks, curated from real-world, domain-specific datasets, ensuring that tasks reflect authentic scientific challenges. The toolkit features a flexible, extensible evaluation pipeline that enables batch evaluation across models and datasets, supports custom model and dataset integration, and provides transparent, reproducible, and comparable results. By bridging capability-based evaluation and disciplinary diversity, SciEvalKit offers a standardized yet customizable infrastructure to benchmark the next generation of scientific foundation models and intelligent agents. The toolkit is open-sourced and actively maintained to foster community-driven development and progress in AI4Science.

cs.AI↗

arXiv2026

SketchThinker-R1: Towards Efficient Sketch-Style Reasoning in Large Multimodal Models

Despite the empirical success of extensive, step-by-step reasoning in large multimodal models, long reasoning processes inevitably incur substantial computational overhead, i.e., in terms of higher token costs and increased response time, which undermines inference efficiency. In contrast, humans often employ sketch-style reasoning: a concise, goal-directed cognitive process that prioritizes salient information and enables efficient problem-solving. Inspired by this cognitive efficiency, we propose SketchThinker-R1, which incentivizes sketch-style reasoning ability in large multimodal models. Our method consists of three primary stages. In the Sketch-Mode Cold Start stage, we convert standard long reasoning process into sketch-style reasoning and finetune base multimodal model, instilling initial sketch-style reasoning capability. Next, we train SketchJudge Reward Model, which explicitly evaluates thinking process of model and assigns higher scores to sketch-style reasoning. Finally, we conduct Sketch-Thinking Reinforcement Learning under supervision of SketchJudge to further generalize sketch-style reasoning ability. Experimental evaluation on four benchmarks reveals that our SketchThinker-R1 achieves over 64% reduction in reasoning token cost without compromising final answer accuracy. Qualitative analysis further shows that sketch-style reasoning focuses more on key cues during problem solving.

cs.CV↗

arXiv2025

SCP: Accelerating Discovery with a Global Web of Autonomous Scientific Agents

We introduce SCP: the Science Context Protocol, an open-source standard designed to accelerate discovery by enabling a global network of autonomous scientific agents. SCP is built on two foundational pillars: (1) Unified Resource Integration: At its core, SCP provides a universal specification for describing and invoking scientific resources, spanning software tools, models, datasets, and physical instruments. This protocol-level standardization enables AI agents and applications to discover, call, and compose capabilities seamlessly across disparate platforms and institutional boundaries. (2) Orchestrated Experiment Lifecycle Management: SCP complements the protocol with a secure service architecture, which comprises a centralized SCP Hub and federated SCP Servers. This architecture manages the complete experiment lifecycle (registration, planning, execution, monitoring, and archival), enforces fine-grained authentication and authorization, and orchestrates traceable, end-to-end workflows that bridge computational and physical laboratories. Based on SCP, we have constructed a scientific discovery platform that offers researchers and agents a large-scale ecosystem of more than 1,600 tool resources. Across diverse use cases, SCP facilitates secure, large-scale collaboration between heterogeneous AI systems and human researchers while significantly reducing integration overhead and enhancing reproducibility. By standardizing scientific context and tool orchestration at the protocol level, SCP establishes essential infrastructure for scalable, multi-institution, agent-driven science.

cs.AI↗

arXiv2025

MokA: Multimodal Low-Rank Adaptation for MLLMs

In this paper, we reveal that most current efficient multimodal fine-tuning methods are hindered by a key limitation: they are directly borrowed from LLMs, often neglecting the intrinsic differences of multimodal scenarios and even affecting the full utilization of all modalities. Inspired by our empirical observation, we argue that unimodal adaptation and cross-modal adaptation are two essential parts for the effective fine-tuning of MLLMs. From this perspective, we propose Multimodal low-rank Adaptation (MokA), a multimodal-aware efficient fine-tuning strategy that takes multimodal characteristics into consideration. It compresses unimodal information by modality-specific parameters while explicitly enhancing cross-modal interaction, ensuring both unimodal and cross-modal adaptation. Extensive experiments cover three representative multimodal scenarios (audio-visual-text, visual-text, and speech-text), and multiple LLM backbones (LLaMA2/3, Qwen2, Qwen2.5-VL, etc). Consistent improvements indicate the efficacy and versatility of the proposed method. Ablation studies and efficiency evaluation are also conducted to fully asses our method. Overall, we think MokA provides a more targeted solution for efficient adaptation of MLLMs, paving the way for further exploration. The project page is at https://gewu-lab.github.io/MokA.

cs.CV↗

arXiv2025

Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows

Despite advances in scientific AI, a coherent framework for Scientific General Intelligence (SGI)-the ability to autonomously conceive, investigate, and reason across scientific domains-remains lacking. We present an operational SGI definition grounded in the Practical Inquiry Model (PIM: Deliberation, Conception, Action, Perception) and operationalize it via four scientist-aligned tasks: deep research, idea generation, dry/wet experiments, and experimental reasoning. SGI-Bench comprises over 1,000 expert-curated, cross-disciplinary samples inspired by Science's 125 Big Questions, enabling systematic evaluation of state-of-the-art LLMs. Results reveal gaps: low exact match (10--20%) in deep research despite step-level alignment; ideas lacking feasibility and detail; high code executability but low execution result accuracy in dry experiments; low sequence fidelity in wet protocols; and persistent multimodal comparative-reasoning challenges. We further introduce Test-Time Reinforcement Learning (TTRL), which optimizes retrieval-augmented novelty rewards at inference, enhancing hypothesis novelty without reference answer. Together, our PIM-grounded definition, workflow-centric benchmark, and empirical insights establish a foundation for AI systems that genuinely participate in scientific discovery.

cs.AI↗

arXiv2025

Single-Agent Scaling Fails Multi-Agent Intelligence: Towards Foundation Models with Native Multi-Agent Intelligence

Foundation models (FMs) are increasingly assuming the role of the ''brain'' of AI agents. While recent efforts have begun to equip FMs with native single-agent abilities -- such as GUI interaction or integrated tool use -- we argue that the next frontier is endowing FMs with native multi-agent intelligence. We identify four core capabilities of FMs in multi-agent contexts: understanding, planning, efficient communication, and adaptation. Contrary to assumptions about the spontaneous emergence of such abilities, we provide extensive empirical evidence, across 41 large language models and 7 challenging benchmarks, showing that scaling single-agent performance alone does not automatically yield robust multi-agent intelligence. To address this gap, we outline key research directions -- spanning dataset construction, evaluation, training paradigms, and safety considerations -- for building FMs with native multi-agent intelligence.

cs.AI↗

arXiv2025

SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines

We present a scientific reasoning foundation model that aligns natural language with heterogeneous scientific representations. The model is pretrained on a 206B-token corpus spanning scientific text, pure sequences, and sequence-text pairs, then aligned via SFT on 40M instructions, annealed cold-start bootstrapping to elicit long-form chain-of-thought, and reinforcement learning with task-specific reward shaping, which instills deliberate scientific reasoning. It supports four capability families, covering up to 103 tasks across workflows: (i) faithful translation between text and scientific formats, (ii) text/knowledge extraction, (iii) property prediction, (iv) property classification, (v) unconditional and conditional sequence generation and design. Compared with specialist systems, our approach broadens instruction coverage, improves cross-domain generalization, and enhances fidelity. We detail data curation and training and show that cross-discipline learning strengthens transfer and downstream reliability. The model, instruct tuning datasets and the evaluation code are open-sourced at https://huggingface.co/SciReason and https://github.com/open-sciencelab/SciReason.

cs.CL↗

arXiv2025

LabUtopia: High-Fidelity Simulation and Hierarchical Benchmark for Scientific Embodied Agents

Scientific embodied agents play a crucial role in modern laboratories by automating complex experimental workflows. Compared to typical household environments, laboratory settings impose significantly higher demands on perception of physical-chemical transformations and long-horizon planning, making them an ideal testbed for advancing embodied intelligence. However, its development has been long hampered by the lack of suitable simulator and benchmarks. In this paper, we address this gap by introducing LabUtopia, a comprehensive simulation and benchmarking suite designed to facilitate the development of generalizable, reasoning-capable embodied agents in laboratory settings. Specifically, it integrates i) LabSim, a high-fidelity simulator supporting multi-physics and chemically meaningful interactions; ii) LabScene, a scalable procedural generator for diverse scientific scenes; and iii) LabBench, a hierarchical benchmark spanning five levels of complexity from atomic actions to long-horizon mobile manipulation. LabUtopia supports 30 distinct tasks and includes more than 200 scene and instrument assets, enabling large-scale training and principled evaluation in high-complexity environments. We demonstrate that LabUtopia offers a powerful platform for advancing the integration of perception, planning, and control in scientific-purpose agents and provides a rigorous testbed for exploring the practical capabilities and generalization limits of embodied intelligence in future research.

cs.RO↗

arXiv2025

Reasoning via Video: The First Evaluation of Video Models' Reasoning Abilities through Maze-Solving Tasks

Video Models have achieved remarkable success in high-fidelity video generation with coherent motion dynamics. Analogous to the development from text generation to text-based reasoning in language modeling, the development of video models motivates us to ask: Can video models reason via video generation? Compared with the discrete text corpus, video grounds reasoning in explicit spatial layouts and temporal continuity, which serves as an ideal substrate for spatial reasoning. In this work, we explore the reasoning via video paradigm and introduce VR-Bench -- a comprehensive benchmark designed to systematically evaluate video models' reasoning capabilities. Grounded in maze-solving tasks that inherently require spatial planning and multi-step reasoning, VR-Bench contains 7,920 procedurally generated videos across five maze types and diverse visual styles. Our empirical analysis demonstrates that SFT can efficiently elicit the reasoning ability of video model. Video models exhibit stronger spatial perception during reasoning, outperforming leading VLMs and generalizing well across diverse scenarios, tasks, and levels of complexity. We further discover a test-time scaling effect, where diverse sampling during inference improves reasoning reliability by 10--20%. These findings highlight the unique potential and scalability of reasoning via video for spatial reasoning tasks.

cs.CV↗

arXiv2025

P1: Mastering Physics Olympiads with Reinforcement Learning

Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric. Physics is the sharpest test of this shift, which binds symbols to reality in a fundamental way, serving as the cornerstone of most modern technologies. In this work, we manage to advance physics research by developing large language models with exceptional physics reasoning capabilities, especially excel at solving Olympiad-level physics problems. We introduce P1, a family of open-source physics reasoning models trained entirely through reinforcement learning (RL). Among them, P1-235B-A22B is the first open-source model with Gold-medal performance at the latest International Physics Olympiad (IPhO 2025), and wins 12 gold medals out of 13 international/regional physics competitions in 2024/2025. P1-30B-A3B also surpasses almost all other open-source models on IPhO 2025, getting a silver medal. Further equipped with an agentic framework PhysicsMinions, P1-235B-A22B+PhysicsMinions achieves overall No.1 on IPhO 2025, and obtains the highest average score over the 13 physics competitions. Besides physics, P1 models also present great performance on other reasoning tasks like math and coding, showing the great generalibility of P1 series.

cs.LG↗

arXiv2025

Scientists' First Exam: Probing Cognitive Abilities of MLLM via Perception, Understanding, and Reasoning

Scientific discoveries increasingly rely on complex multimodal reasoning based on information-intensive scientific data and domain-specific expertise. Empowered by expert-level scientific benchmarks, scientific Multimodal Large Language Models (MLLMs) hold the potential to significantly enhance this discovery process in realistic workflows. However, current scientific benchmarks mostly focus on evaluating the knowledge understanding capabilities of MLLMs, leading to an inadequate assessment of their perception and reasoning abilities. To address this gap, we present the Scientists' First Exam (SFE) benchmark, designed to evaluate the scientific cognitive capacities of MLLMs through three interconnected levels: scientific signal perception, scientific attribute understanding, scientific comparative reasoning. Specifically, SFE comprises 830 expert-verified VQA pairs across three question types, spanning 66 multimodal tasks across five high-value disciplines. Extensive experiments reveal that current state-of-the-art GPT-o3 and InternVL-3 achieve only 34.08% and 26.52% on SFE, highlighting significant room for MLLMs to improve in scientific realms. We hope the insights obtained in SFE will facilitate further developments in AI-enhanced scientific discoveries.

cs.AI↗

arXiv2025

ChemBOMAS: Accelerated BO in Chemistry with LLM-Enhanced Multi-Agent System

Bayesian optimization (BO) is a powerful tool for scientific discovery in chemistry, yet its efficiency is often hampered by the sparse experimental data and vast search space. Here, we introduce ChemBOMAS: a large language model (LLM)-enhanced multi-agent system that accelerates BO through synergistic data- and knowledge-driven strategies. Firstly, the data-driven strategy involves an 8B-scale LLM regressor fine-tuned on a mere 1% labeled samples for pseudo-data generation, robustly initializing the optimization process. Secondly, the knowledge-driven strategy employs a hybrid Retrieval-Augmented Generation approach to guide LLM in dividing the search space while mitigating LLM hallucinations. An Upper Confidence Bound algorithm then identifies high-potential subspaces within this established partition. Across the LLM-refined subspaces and supported by LLM-generated data, BO achieves the improvement of effectiveness and efficiency. Comprehensive evaluations across multiple scientific benchmarks demonstrate that ChemBOMAS set a new state-of-the-art, accelerating optimization efficiency by up to 5-fold compared to baseline methods.

cs.LG↗

arXiv2025

DeepResearch Arena: The First Exam of LLMs' Research Abilities via Seminar-Grounded Tasks

Deep research agents have attracted growing attention for their potential to orchestrate multi-stage research workflows, spanning literature synthesis, methodological design, and empirical verification. Despite these strides, evaluating their research capability faithfully is rather challenging due to the difficulty of collecting frontier research questions that genuinely capture researchers' attention and intellectual curiosity. To address this gap, we introduce DeepResearch Arena, a benchmark grounded in academic seminars that capture rich expert discourse and interaction, better reflecting real-world research environments and reducing the risk of data leakage. To automatically construct DeepResearch Arena, we propose a Multi-Agent Hierarchical Task Generation (MAHTG) system that extracts research-worthy inspirations from seminar transcripts. The MAHTG system further translates research-worthy inspirations into high-quality research tasks, ensuring the traceability of research task formulation while filtering noise. With the MAHTG system, we curate DeepResearch Arena with over 10,000 high-quality research tasks from over 200 academic seminars, spanning 12 disciplines, such as literature, history, and science. Our extensive evaluation shows that DeepResearch Arena presents substantial challenges for current state-of-the-art agents, with clear performance gaps observed across different models.

cs.AI↗

arXiv2025

SynBrain: Enhancing Visual-to-fMRI Synthesis via Probabilistic Representation Learning

Deciphering how visual stimuli are transformed into cortical responses is a fundamental challenge in computational neuroscience. This visual-to-neural mapping is inherently a one-to-many relationship, as identical visual inputs reliably evoke variable hemodynamic responses across trials, contexts, and subjects. However, existing deterministic methods struggle to simultaneously model this biological variability while capturing the underlying functional consistency that encodes stimulus information. To address these limitations, we propose SynBrain, a generative framework that simulates the transformation from visual semantics to neural responses in a probabilistic and biologically interpretable manner. SynBrain introduces two key components: (i) BrainVAE models neural representations as continuous probability distributions via probabilistic learning while maintaining functional consistency through visual semantic constraints; (ii) A Semantic-to-Neural Mapper acts as a semantic transmission pathway, projecting visual semantics into the neural response manifold to facilitate high-fidelity fMRI synthesis. Experimental results demonstrate that SynBrain surpasses state-of-the-art methods in subject-specific visual-to-fMRI encoding performance. Furthermore, SynBrain adapts efficiently to new subjects with few-shot data and synthesizes high-quality fMRI signals that are effective in improving data-limited fMRI-to-image decoding performance. Beyond that, SynBrain reveals functional consistency across trials and subjects, with synthesized signals capturing interpretable patterns shaped by biological neural variability. Our code is available at https://github.com/MichaelMaiii/SynBrain.

cs.LG↗

arXiv2025

MOOSE-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses

Scientific discovery plays a pivotal role in advancing human society, and recent progress in large language models (LLMs) suggests their potential to accelerate this process. However, it remains unclear whether LLMs can autonomously generate novel and valid hypotheses in chemistry. In this work, we investigate whether LLMs can discover high-quality chemistry hypotheses given only a research background-comprising a question and/or a survey-without restriction on the domain of the question. We begin with the observation that hypothesis discovery is a seemingly intractable task. To address this, we propose a formal mathematical decomposition grounded in a fundamental assumption: that most chemistry hypotheses can be composed from a research background and a set of inspirations. This decomposition leads to three practical subtasks-retrieving inspirations, composing hypotheses with inspirations, and ranking hypotheses - which together constitute a sufficient set of subtasks for the overall scientific discovery task. We further develop an agentic LLM framework, MOOSE-Chem, that is a direct implementation of this mathematical decomposition. To evaluate this framework, we construct a benchmark of 51 high-impact chemistry papers published and online after January 2024, each manually annotated by PhD chemists with background, inspirations, and hypothesis. The framework is able to rediscover many hypotheses with high similarity to the groundtruth, successfully capturing the core innovations-while ensuring no data contamination since it uses an LLM with knowledge cutoff date prior to 2024. Finally, based on LLM's surprisingly high accuracy on inspiration retrieval, a task with inherently out-of-distribution nature, we propose a bold assumption: that LLMs may already encode latent scientific knowledge associations not yet recognized by humans.

cs.CL↗

arXiv2025

MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search

Large language models (LLMs) have shown promise in automating scientific hypothesis generation, yet existing approaches primarily yield coarse-grained hypotheses lacking critical methodological and experimental details. We introduce and formally define the new task of fine-grained scientific hypothesis discovery, which entails generating detailed, experimentally actionable hypotheses from coarse initial research directions. We frame this as a combinatorial optimization problem and investigate the upper limits of LLMs' capacity to solve it when maximally leveraged. Specifically, we explore four foundational questions: (1) how to best harness an LLM's internal heuristics to formulate the fine-grained hypothesis it itself would judge as the most promising among all the possible hypotheses it might generate, based on its own internal scoring-thus defining a latent reward landscape over the hypothesis space; (2) whether such LLM-judged better hypotheses exhibit stronger alignment with ground-truth hypotheses; (3) whether shaping the reward landscape using an ensemble of diverse LLMs of similar capacity yields better outcomes than defining it with repeated instances of the strongest LLM among them; and (4) whether an ensemble of identical LLMs provides a more reliable reward landscape than a single LLM. To address these questions, we propose a hierarchical search method that incrementally proposes and integrates details into the hypothesis, progressing from general concepts to specific experimental configurations. We show that this hierarchical process smooths the reward landscape and enables more effective optimization. Empirical evaluations on a new benchmark of expert-annotated fine-grained hypotheses from recent literature show that our method consistently outperforms strong baselines.

cs.CL↗