arXiv subjects

Lingfei Wu

Publications and source records attributed to Lingfei Wu.

At least 19 recordsLinked to original sources

arXiv2026

Fu-TCM: a multimodal large language model that learns traditional Chinese medicine diagnostic reasoning

Large language models (LLMs) are advancing in multimodal understanding and reasoning, accelerating medical model development. Traditional Chinese medicine (TCM) uses evidence from the four examinations to determine disease location and nature, differentiate syndromes and guide treatment. This diagnostic process constitutes a complete chain of reasoning. We present Fu-TCM, a multimodal TCM LLM trained on data retained after cleaning and filtering an initial collection of approximately 7.12 million examples derived from classical texts, textbooks, medical images, clinical cases and publicly available datasets. The model was further optimized using Bianzheng-Grounded Policy Optimization (BGPO), which explicitly targets syndrome differentiation through rewards for response format, intermediate TCM reasoning fields and final syndrome predictions. Fu-TCM-27B led all six benchmarks spanning TCM reasoning, text and vision, achieving a macro-average accuracy of 81.01% and surpassing the strongest reference model with complete results across all six benchmarks by 7.13 percentage points. Its largest margin was 11.90 points on Fu-bianzheng. Case-level analyses further showed a positive association between intermediate-field accuracy and final syndrome-selection accuracy in most evaluated models. These findings suggest that TCM syndrome differentiation may be computationally learnable and that structured reasoning supervision can improve performance across TCM tasks.

cs.CL↗

arXiv2026

Large Teams Consistently Develop Science

Scientific progress depends on both developing established knowledge and disrupting it to make way for new ideas. Using the disruption index (D), our previous work showed that large teams tend to develop science, while small teams are more likely to disrupt it. Recent work challenges this finding, arguing that D mechanically depends on reference-list length. Using an alternative measure (oneD), it reaches the opposite conclusion: large teams are more disruptive. Combining Nobel Prize case studies, mathematical analysis, and large-scale publication data, we show that oneD misclassifies developmental contributions as disruptive and introduces the very sensitivity to reference-list length that it seeks to eliminate. The original measure, D, however, shows no evidence of the claimed systematic dependence on reference-list length. More fundamentally, even after controlling for reference-list length, larger teams remain more likely to produce strongly developmental work. Our findings reaffirm that large teams consistently develop science and offer new insights into the measurement of scientific disruption.

cs.DL↗

arXiv2026

Faster alone vs. further together: The speed and impact of researchers' responses to societal crises

Throughout the 21st century, society encountered numerous crises that demanded quick and widespread action. In the present work, we documented researchers' responses to 25 of these crises, analyzing 1,537,284 published responses from 3,262,152 researchers spanning 225 countries and territories. An analysis of publication speed and citation patterns yielded support for a proverbial claim: "to go fast, go alone; to go far, go together". Solo authors were quicker to publish responses to societal crises -- but their responses received less attention in scholarly, news, and policy venues. Growth towards large-scale collaboration was associated with increased impact; However, this team size advantage was characterized by diminishing returns and significantly varied across sampled crises. Furthermore, pluralistic machine learning modeling highlights the importance of considering additional context, such as government effectiveness and the expertise of team leaders. Taken together, results facilitate evidence-based conversations regarding how researchers optimize speed and impact in the midst of societal crises.

cs.DL↗

arXiv2026

GANDR: Claim Auditing for Verifiable Legal Answer Generation

In high-stakes domains such as legal practice, a language-model answer is only useful to the extent that a reader can verify each claim against the source the system cites. Current grounded-generation pipelines score the answer as a whole, so a correct conclusion can rest on fabricated or loosely matched citations and still score well. Closing this gap requires both a system built for per-claim verification and an evaluation that measures it. We introduce GANDR (Grounded ANswer DRafter), a two-agent system in which a Drafter writes an answer in a structured legal-reasoning format and a separate Critic, with the same view as a human verifier, audits each claim against its cited source and emits a per-claim audit trace on every round. We pair it with a strict correctness criterion requiring every citation to resolve to a passage the retriever returned. On a 185-item legal benchmark where all six systems share one backbone, one retrieval surface, and one citation instruction, GANDR ranks first on every primary metric, reaching 70.8% strict accuracy and leading the strongest baseline by 11.3 points (p<0.01). Reverting the protocol-anchored commit rule lowers strict accuracy by 22.7 points, and the strict lead stays positive on three further backbones, at +3.2 to +6.5 points. This lead traces to the Drafter configuration and the protocol-anchored commit, not to rewriting. Against two law-trained annotators the audit flags under-supported claims at F1 0.84 as a binary detector, while its four-way verdict labels agree only weakly and are advisory. Code is available upon request.

cs.CL↗

arXiv2026

AtomCite: Verification and Correction of Supplied Page-Level Citations in Multi-Page Documents

Large language models answering questions over multi-page documents are expected to cite the supporting pages, yet supplied citations are sometimes inaccurate, and current evaluations score citations at generation time or against text passages: no existing benchmark evaluates whether a system can verify and correct a page-level citation already attached to an answer. We propose AtomCite, an agentic framework that parses an answer into claims, checks each claim against the image of its cited page, and applies a deterministic repair policy. To evaluate it, we introduce DocCite, to our knowledge the first benchmark for systems that verify and correct page-level citations in document images. Built on MP-DocVQA and DUDE, it combines 928 validated injected instances with 2,468 candidate natural errors harvested from frontier- and efficiency-tier models, of which a two-annotator audit confirms 1,909 as genuine errors. Primary labels are assigned deterministically, not by LLM judges, with the human audit as a separate validation layer. Across three model families (Gemini, Claude, and GPT), AtomCite reaches around 93% binary verification accuracy on the injected benchmark, significantly outperforming every OCR-only condition, including a compute-matched control, and exceeding every prior text-based baseline given the same OCR text. Its repair policy lifts citation precision on the injected mix from a constructed 34% to 87-90% while retaining over 90% of correct claims. AtomCite also transfers: with frozen prompts and zero training, it raises the hallucination-detection scores of two open 7-8B models on five public benchmarks above the same models prompted as direct judges. Finally, the audit shows that noise in automatic labels biases measured verifier accuracy and can reverse system rankings, so evaluations relying only on synthetic or automatic labels risk mismeasuring verification capability.

cs.CL↗

arXiv2026

The Longevity of Innovation

Modern science is organized around specialization in training and teamwork. Scientists develop deep expertise within a field and combine complementary knowledge through collaboration to solve complex problems. Yet whether specialization is the most effective path to sustained innovation remains unclear. Here we introduce a quantitative framework that distinguishes generalists from specialists based on scaling patterns of disciplinary mobility while remaining independent of career age and productivity. Applying this framework to 49 million publications produced by 3 million scientists between 1900 and 2020, we examine how research style relates to innovation, learning, collaboration, and productivity. We find that scientists who move across fields are more likely to sustain innovative contributions throughout their careers, whereas those who remain within narrow fields exhibit the age-related decline in innovation. Generalists are less anchored to the literature of their training. They are more likely to pursue research independently, and, when they collaborate, they preferentially partner with other generalists. Teams with a greater share of generalists produce more innovative research, even after accounting for differences in knowledge diversity. Despite these advantages, generalists publish fewer papers on average and have become less common over time. These findings reveal a tension between the longevity of scientific careers and the longevity of scientific innovation.

cs.SI↗

arXiv2026

From "Thinking" to "Justifying": Aligning High-Stakes Explainability with Professional Communication Standards

Explainable AI (XAI) in high-stakes domains should help stakeholders trust and verify system outputs. Yet Chain-of-Thought methods reason before concluding, and logical gaps or hallucinations can yield conclusions that do not reliably align with their rationale. Thus, we propose "Result -> Justify", which constrains the output communication to present a conclusion before its structured justification. We introduce SEF (Structured Explainability Framework), operationalizing professional conventions (e.g., CREAC, BLUF) via six metrics for structure and grounding. Experiments across four tasks in three domains validate this approach: all six metrics correlate with correctness (r=0.20-0.42; p<0.001), and SEF achieves 83.9% accuracy (+5.3 over CoT). These results suggest structured justification can improve verifiability and may also improve reliability.

cs.AI↗

arXiv2025

Aging and the Narrowing of Scientific Innovation

With rising life expectancies around the world and an older scientific workforce than ever before, what does aging mean for individual scientists, and what do aging scientists mean for scientific progress as a whole? Here we examine how scientists and scholars age in terms of how their ideas and contributions relate to the evolving frontier of knowledge and how demographically aging fields relate to field-level advance. At the individual level, we examine how research experiences and choices can moderate the effects of intellectual aging. At the collective level, we explore mechanisms that link individual and collective aging. Prior research focuses on star scientists, their changing dates and rates of breakthrough success throughout history. We explore this for scientists in all fields over time, drawing upon novel deep learning measurements that allow us not only to trace positive attention through citation but also negative attention through explicit criticism with a novel, comprehensive database of over 20,000 human-validated critical citations. We find that younger scientists tend toward disruptive contributions that push the frontier, while older scientists engage in combinatorial innovation with an aging collection of components. This includes analyzing the impact of the 1994 U.S. Supreme Court ruling on mandatory retirement and examining how unexpected collaborations affect citation patterns.

cs.DL↗

arXiv2025

Innovation by Displacement

New ideas are often thought to arise from recombining existing knowledge. Yet despite rapid publication growth - and expanding opportunities for recombination - scientific breakthroughs remain rare. This gap between productivity and progress challenges recombinant growth theory as the prevailing account of innovation. We argue that the limitation of this theory lies in treating ideas solely as complements, overlooking that breakthroughs often arise when ideas act as substitutes. To test this, we integrate scientist interviews, bibliometric validation, and machine learning analysis of 41 million papers (1965 - 2024). Interviews reveal that breakthroughs are marked not by novelty (Atypicality) alone but by their ability to displace dominant ideas (Disruption). Large-scale analysis confirms that novelty and disruption represent distinct innovation mechanisms: they are negatively correlated across domains, periods, team sizes, and paper versions. Novel papers extend dominant ideas across topics and attract immediate attention; disruptive papers displace them within the same topic and generate lasting influence. Hence, progress slows not from lack of effort but because most research extends rather than overturns ideas. Applying this perspective reveals distinct roles of theories and methods in scientific change: methods more often drive breakthroughs, whereas theories tend to be novel but rarely disruptive, reinforcing the dominance of established ideas.

cs.DL↗

arXiv2025

Can Recombination Displace Dominant Scientific Ideas

The combination of diverse, pre-existing knowledge is a common explanation for scientific breakthroughs. However, a paradox exists: while scientific output and the potential for such recombination have grown exponentially, the rate of breakthrough discoveries has not. To explore this paradox, our study examines 41 million scientific articles from 1965 to 2024. We measure two key properties for each paper: atypicality, which quantifies the combination of knowledge from conceptually distant areas, and disruption. We demonstrate that these metrics capture distinct processes. Atypicality is characteristic of work that extends established concepts into new topical areas (a form of cross-topic recombination). Disruption, in contrast, signifies the replacement of a dominant idea within its own topic.

cs.DL↗

arXiv2025

Report of the 2025 Workshop on Next-Generation Ecosystems for Scientific Computing: Harnessing Community, Software, and AI for Cross-Disciplinary Team Science

This report summarizes insights from the 2025 Workshop on Next-Generation Ecosystems for Scientific Computing: Harnessing Community, Software, and AI for Cross-Disciplinary Team Science, which convened more than 40 experts from national laboratories, academia, industry, and community organizations to chart a path toward more powerful, sustainable, and collaborative scientific software ecosystems. To address urgent challenges at the intersection of high-performance computing (HPC), AI, and scientific software, participants envisioned agile, robust ecosystems built through socio-technical co-design--the intentional integration of social and technical components as interdependent parts of a unified strategy. This approach combines advances in AI, HPC, and software with new models for cross-disciplinary collaboration, training, and workforce development. Key recommendations include building modular, trustworthy AI-enabled scientific software systems; enabling scientific teams to integrate AI systems into their workflows while preserving human creativity, trust, and scientific rigor; and creating innovative training pipelines that keep pace with rapid technological change. Pilot projects were identified as near-term catalysts, with initial priorities focused on hybrid AI/HPC infrastructure, cross-disciplinary collaboration and pedagogy, responsible AI guidelines, and prototyping of public-private partnerships. This report presents a vision of next-generation ecosystems for scientific computing where AI, software, hardware, and human expertise are interwoven to drive discovery, expand access, strengthen the workforce, and accelerate scientific progress.

cs.CE↗

arXiv2025

Low-skilled Occupations Face the Highest Upskilling Pressure

Substantial scholarship has estimated the susceptibility of jobs to automation, but little has examined how job contents evolve in the information age as new technologies substitute for tasks, shifting required skills rather than eliminating entire jobs. Here we explore patterns of occupational skill change and characterize occupations and workers subject to the greatest reskilling requirements. Recent work found that changing skill requirements are greatest for STEM occupations in the 2010s. Nevertheless, analyzing 167 million online job posts covering 727 occupations, we find that skill change is greatest for low-skilled occupations when accounting for distance between skills. We further investigate the differences in skill change across employer and market size, as well as social demographic groups. We find that jobs from small employers and markets experienced larger skill upgrades to catch up with the skill demands of their large employers and markets. Female and minority workers are disproportionately employed in low-skilled jobs and face the most significant skill adjustments. While these varied skill changes could create uneven reskilling pressures across workers, they may also lead to a narrowing of gaps in job quality and prospects. We conclude by showcasing our model's potential to chart job evolution directions using skill embedding spaces.

cs.CY↗

arXiv2025

Vague Preference Policy Learning for Conversational Recommendation

Conversational recommendation systems (CRS) commonly assume users have clear preferences, leading to potential over-filtering of relevant alternatives. However, users often exhibit vague, non-binary preferences. We introduce the Vague Preference Multi-round Conversational Recommendation (VPMCR) scenario, employing a soft estimation mechanism to accommodate users' vague and dynamic preferences while mitigating over-filtering. In VPMCR, we propose Vague Preference Policy Learning (VPPL), consisting of Ambiguity-aware Soft Estimation (ASE) and Dynamism-aware Policy Learning (DPL). ASE captures preference vagueness by estimating scores for clicked and non-clicked options, using a choice-based approach and time-aware preference decay. DPL leverages ASE's preference distribution to guide the conversation and adapt to preference changes for recommendations or attribute queries. Extensive experiments demonstrate VPPL's effectiveness within VPMCR, outperforming existing methods and setting a new benchmark. Our work advances CRS by accommodating users' inherent ambiguity and relative decision-making processes, improving real-world applicability.

cs.IR↗

arXiv2025

Writing Patterns Reveal a Hidden Division of Labor in Scientific Teams

The recognition of individual contributions is central to the scientific reward system, yet coauthored papers often obscure who did what. Traditional proxies like author order assume a simplistic decline in contribution, while emerging practices such as self-reported roles are biased and limited in scope. We introduce a large-scale, behavior-based approach to identifying individual contributions in scientific papers. Using author-specific LaTeX macros as writing signatures, we analyze over 730,000 arXiv papers (1991-2023), covering over half a million scientists. Validated against self-reports, author order, disciplinary norms, and Overleaf records, our method reliably infers author-level writing activity. Section-level traces reveal a hidden division of labor: first authors focus on technical sections (e.g., Methods, Results), while last authors primarily contribute to conceptual sections (e.g., Introduction, Discussion). Our findings offer empirical evidence of labor specialization at scale and new tools to improve credit allocation in collaborative research.

cs.SI↗

arXiv2025

The Disruption Index Measures Displacement Between a Paper and Its Most Cited Reference

Initially developed to capture technical innovation and later adapted to identify scientific breakthroughs, the Disruption Index (D-index) offers the first quantitative framework for analyzing transformative research. Despite its promise, prior studies have struggled to clarify its theoretical foundations, raising concerns about potential bias. Here, we show that-contrary to the common belief that the D-index measures absolute innovation-it captures relative innovation: a paper's ability to displace its most-cited reference. In this way, the D-index reflects scientific progress as the replacement of older answers with newer ones to the same fundamental question-much like light bulbs replacing candles. We support this insight through mathematical analysis, expert surveys, and large-scale bibliometric evidence. To facilitate replication, validation, and broader use, we release a dataset of D-index values for 49 million journal articles (1800-2024) based on OpenAlex.

cs.DL↗

arXiv2025

Team Size and Its Negative Impact on the Disruption Index

As science transitions from the age of lone geniuses to an era of collaborative teams, the question of whether large teams can sustain the creativity of individuals and continue driving innovation has become increasingly important. Our previous research first revealed a negative relationship between team size and the Disruption Index-a network-based metric of innovation-by analyzing 65 million projects across papers, patents, and software over half a century. This work has sparked lively debates within the scientific community about the robustness of the Disruption Index in capturing the impact of team size on innovation. Here, we present additional evidence that the negative link between team size and disruption holds, even when accounting for factors such as reference length, citation impact, and historical time. We further show how a narrow 5-year window for measuring disruption can misrepresent this relationship as positive, underestimating the long-term disruptive potential of small teams. Like "sleeping beauties," small teams need a decade or more to see their transformative contributions to science.

cs.SI↗

arXiv2025

Towards Expressive Spectral-Temporal Graph Neural Networks for Time Series Forecasting

Time series forecasting has remained a focal point due to its vital applications in sectors such as energy management and transportation planning. Spectral-temporal graph neural network is a promising abstraction underlying most time series forecasting models that are based on graph neural networks (GNNs). However, more is needed to know about the underpinnings of this branch of methods. In this paper, we establish a theoretical framework that unravels the expressive power of spectral-temporal GNNs. Our results show that linear spectral-temporal GNNs are universal under mild assumptions, and their expressive power is bounded by our extended first-order Weisfeiler-Leman algorithm on discrete-time dynamic graphs. To make our findings useful in practice on valid instantiations, we discuss related constraints in detail and outline a theoretical blueprint for designing spatial and temporal modules in spectral domains. Building on these insights and to demonstrate how powerful spectral-temporal GNNs are based on our framework, we propose a simple instantiation named Temporal Graph Gegenbauer Convolution (TGGC), which significantly outperforms most existing models with only linear components and shows better model efficiency. Our findings pave the way for devising a broader array of provably expressive GNN-based models for time series.

cs.LG↗

arXiv2025

Is Science Inevitable?

Using large-scale citation data and a breakthrough metric, the study systematically evaluates the inevitability of scientific breakthroughs. We find that scientific breakthroughs emerge as multiple discoveries rather than singular events. Through analysis of over 40 million journal articles, we identify multiple discoveries as papers that independently displace the same reference using the Disruption Index (D-index), suggesting functional equivalence. Our findings support Merton's core argument that scientific discoveries arise from historical context rather than individual genius. The results reveal a long-tail distribution pattern of multiple discoveries across various datasets, challenging Merton's Poisson model while reinforcing the structural inevitability of scientific progress.

cs.DL↗