Searcharxiv⌕ Search

arXiv subjects

Jiashuo Zhang

Publications and source records attributed to Jiashuo Zhang.

At least 19 recordsLinked to original sources

Causally Constrained and Uncertainty-Aware Cardiopulmonary Risk Prediction

Routine chest computed tomography (CT) captures pulmonary, cardiac, and vascular phenotypes that jointly stratify lung cancer and cardiovascular disease, the two leading causes of death in lung-screening populations. Yet clinical risk calculators evaluate each disease in isolation and omit CT findings, whereas unconstrained machine learning overfits small hospital cohorts, learns treatment- and illness-confounded risk directions, and conceals uncertainty when patient records are incomplete. Here we present the Causal Envelope Agent (CE-Agent), a two-stage framework that compiles peer-reviewed epidemiological evidence into explicit constraints on risk-factor weights before any outcome is seen, while estimating local baseline risk and newly added CT phenotypes from cohort data. At deployment, closed-form uncertainty propagation accounts for missing clinical or CT inputs within milliseconds per patient, running over two orders of magnitude faster than Monte Carlo sampling and returning calibrated risk intervals that direct chart review and selective imaging to borderline patients. Across 53,452 lung-screening participants from 33 US hospitals and 6,814 multi-ethnic adults, CE-Agent outperforms cardiovascular calculators in lung screening, exceeds the Pooled Cohort Equations on hard atherosclerotic events while correcting their 1.7-fold risk overestimation, and generalizes from a single hospital or CT scanner to external sites while lowering required sample sizes up to fivefold. Applied unchanged to lung cancer, CE-Agent matches established screening models with accurate calibration even when restricted to risk-factor directions alone, and pairing both predictions improves joint cardiopulmonary triage from a single screening visit.

cs.AI↗

Verifiable by Construction: How Well Do LLMs Substantiate Their Clinical Claims with Verbatim Quotes?

Large language models (LLMs) have been widely adopted for clinical question answering (QA). Current systems can attach citations to their answers, but these often point to broad texts, leaving time-pressed clinicians unable to verify them efficiently. An alternative is to ensure that responses are verifiable by construction: providing fine-grained verbatim quotes from reference material that substantiate claims, so users can verify an answer without opening other documents. In this paper, we evaluate the ability of current models to perform this task end-to-end: from providing citations for every factual claim, to producing verbatim quotes, to ensuring that those quotes fully substantiate the claims. To do so, we build a standardized harness over four clinical practice guidelines and evaluate fourteen LLMs on 222 synthetic clinical questions, measuring each of these stages separately. We find that most models can attach verbatim quotes to over 90% of their claims from prompting alone, apart from some lightweight models such as claude-haiku-4.5. Yet these quotes often fail to substantiate every detail of the claims they accompany. For instance, claude-opus-5 produces verbatim quotes for 98.0% of its claims, but fully substantiates only 37.1%. Our work provides insights into the current capability gap of LLMs in building verifiable clinical QA systems, along with artifacts for future research.

cs.CL↗

daVinci-Prover: Agent-Native Mid-Training for Formal Theorem Proving

Advances in large language models (LLMs) are rapidly transforming formal theorem proving. Yet further progress remains constrained by the lack of a transparent and effective data recipe for systematically improving theorem-proving models. In this work, we argue that effective agentic provers require strong mathematical skill priors to ground theorem proving and agentic trajectories to enable long-horizon, interactive proof construction. To this end, we synthesize an agent-native 33.7B-token mid-training corpus covering decomposition, search, self-verification, and revision, together with verifier-grounded agentic trajectories. We further introduce the daVinci-Prover harness, which enables models to construct proofs and interact with a Lean verifier in an isolated workspace, and use it to collect approximately 36K high-quality agent trajectories for supervised fine-tuning. Our mid-training recipe yields an average improvement of 13.7 points over the matched Qwen3-8B SFT-only baseline across MiniF2F, FATE-M, and ProofNet. When applied to the Qwen3.5 series, our approach also yields strong performance across theorem-proving benchmarks and achieves near-saturating MiniF2F performance with memory compaction. The results highlight agent-native mid-training as an effective foundation for developing stronger interactive theorem-proving agents.

cs.CL↗

SMO: Compiling Natural Language Skills into Graphs of Reusable State-Move-Outcome Bricks

Large language model agents increasingly rely on reusable natural-language skills, yet growing skill collections remain accumulations of isolated text artifacts that are difficult to compare, deduplicate, and compose. We argue that the reusable content of a skill lies not in its surface wording but in its procedural transitions: which operation to execute under a given state and which observable outcome to produce. Based on this observation, we propose State-Move-Outcome (SMO), a compilation-like framework that decomposes raw skills into reusable procedural bricks and organizes admitted bricks as an externalized symbolic structure. When compiling a skill, SMO first retrieves and composes existing bricks, dynamically connects them through Outcome-to-State conditions, and creates a candidate brick only for an uncovered procedure; the candidate becomes reusable only after validation and admission. Experiments on SkillsBench v1.1 demonstrate that SMO represents 232 rawSkills from 87 underlying tasks with 95 shared bricks, reducing the corresponding skill representations from 366,472 to 149,154 tokens--a $\mathbf{59.3}\times$ reduction and $\mathbf{2.46}\times$ compression. Across 232 skill-conditioned evaluation instances, SMO raises the average task score from $\mathbf{0.6379}$ to $\mathbf{0.7478}$ and increases 5/5 Pass instances from $\mathbf{138}$ to $\mathbf{164}$, while matching or outperforming rawSkill on $\mathbf{208}$ of $\mathbf{232}$ instances. It also reduces average runtime token consumption by $\mathbf{38.38\%}$. These results show that compiling natural-language skills into shared procedural units can jointly improve skill reuse, representation efficiency, and downstream execution.

cs.CL↗

Web3 x AI Agents: Landscape, Integrations, and Foundational Challenges

The convergence of Web3 technologies and AI agents is a rapidly evolving frontier that is reshaping decentralized ecosystems. This paper presents the first comprehensive landscape analysis of the intersection between Web3 and AI agents, examining five critical dimensions: landscape, economics, governance, security, and trust mechanisms. Through an analysis of 979 capitalized Web3-AI agent tokens at a 1 September 2026 cutoff, compared with a cohort of 275 tokens collected at 6 April 2025, we develop a taxonomy and systematically map the current market landscape (RQ1), identifying distinct patterns in project distribution and capitalization. We complement the token frame with on-chain measurement of the agent standards layer: 530,266 ERC-8004 identity registrations across 16 chains and 432,913 x402 settlements on Base. We then investigate four integration domains: AI agents in decentralized finance (RQ2), governance (RQ3), security (RQ4), and Web3 trust mechanisms for agent reliability (RQ5). Infrastructure dominates the landscape in both project count and capitalization; value is highly concentrated (Gini 0.938), and the archived cohort lost 36.4% in capitalization over seventeen months despite token population growth. The agent standards layer shows adoption at scale, but economic throughput remains limited. Across integration domains, AI agents extend DeFi trading and make continuous security auditing affordable, while agents in governance remain largely a blueprint. These findings outline critical considerations for future research toward robust, intelligent, and trustworthy decentralized systems with AI agent interactions.

cs.CY↗

Generating the wide sequence of Diffuse Galaxies with de Broglie waves of Dark Matter

Extensive Euclid satellite imaging at low surface brightness has revealed that most nearby galaxies are diffuse-looking spheroids, where the stellar radius increases for over three decades in luminosity. We argue this Diffuse Galaxy sequence results from internal stellar diffusion by Wave Dark Matter ($ψ$DM), as wave energy is transferred to star orbits over time. In particular, the soliton random motion scatters central stars onto radial orbits that become enhanced with each passage through the centre, slowly ``puffing up" the stellar profile. Heating is greater within massive galaxies as $ψ$DM fluctuations are stronger and more frequent, reproducing both the Diffuse Galaxy sequence and the rising velocity dispersion along the sequence, from Ultra-Faint to Dwarf Spheroidal and Ultra Diffuse galaxies, favouring a light boson, $m_ψ=2.88^{+0.14}_{-0.13}\times10^{-22}$eV. Winding back this diffusion, we predict the stellar content of Diffuse Galaxies, including globular clusters, formed near the centre, as anticipated by $ψ$DM simulations, where gas cools efficiently within the dense soliton. This predicted $ψ$DM evolution from compact beginnings towards diffuse spheroidal galaxies today can now be fully charted from JWST to Euclid.

astro-ph.GA↗

JWST evidence for a sharp "Cosmic Daybreak" at z = 15

Luminous young galaxies have been uncovered with relative ease by JWST, extending to z=14.5, so it is puzzling that deeper spectroscopy of fainter candidates now finds only interlopers. This redshift `ìmpasse" is underscored by the measured stellar ages of these high-z galaxies, which we show converge to zero by z=15, with a marked absence of earlier star-formation. Taken literally, such a late transition from the Dark Ages to luminous galaxies is unlike the gradual Cosmic Dawn of standard LCDM, but does confirm a key prediction of Wave Dark Matter, $ψ$DM, as a Bose-Einstein condensate. The de Broglie wave pressure resists gravity until a substantial Jeans mass of $4\times 10^9M_\odot$ is overcome at z$=$15, corresponding to a light boson $m_ψ=2.2_{-0.3}^{+0.4} \times 10^{-22}$eV, and similar to independent estimates from lensing anomalies and dwarf galaxies. Furthermore, the substantial luminosities of the highest redshift galaxies appear to converge to the initial Jeans scale of $ψ$DM, whereas LCDM predictions extend to lower luminosities and larger ages than observed. These contrasting predictions can be definitively tested as JWST observations accumulate, with diametric implications for Dark Matter as heavy particles beyond the Standard Model or ultra-light bosons motivated by the String Axiverse.

astro-ph.CO↗

Extragalactic microlensing through Ultra Diffuse Galaxies

Stellar microlensing is a powerful method to constrain compact dark matter models, uncover binary stars, and exoplanets during caustic crossing events. At cosmological distances, {\it James-Webb Space Telescope} ({\it JWST}) is routinely detecting microlensed giant stars in highly magnified galaxies behind massive lensing clusters. Here, we explore for the first time microlensing in modest redshift galaxies commonly seen through local Ultra Diffuse Galaxies (UDGs). Using the UDG NGC1052-DF2 as a proof-of-concept, we find that detecting microlensing events through UDGs is possible. Euclid is ideal for identifying samples of low-redshift star-forming galaxies seen through local galaxies for deeper cadenced follow-up, where our zeroth-order calculation estimates that $\mathcal{O}(1-10)$ events per year are expected over the whole sky under the monitoring of LSST, assuming all UDGs share similar distance and stellar content as NGC1052-DF2. Such measurements can help settle the DM issue for UDG galaxies by providing a stellar mass estimate weighted more by the low end of the stellar initial mass function (IMF) than standard stellar synthesis estimates. Note, lower efficiency is estimated for {\it JWST} of $\sim 5.6\times10^{-2}\,\textrm{yr}^{-1}$ over its five background galaxies is expected for typical {\it JWST}$\sim29$\,mag visits, and a low Vera Rubin Legacy Survey of Space and Time (LSST) detection rate of$\sim 0.03\,\textrm{yr}^{-1}$ such that NGC1052-DF2 might not be a prime target given its lack of low-redshift background galaxies.

astro-ph.GA↗

QoS-Aware RACH Preamble Slicing via Quota-Projected Branching Deep Reinforcement Learning

Quality-of-service (QoS)-aware random access requires adaptive allocation of a finite random access channel (RACH) preamble budget across heterogeneous traffic and access procedures. This paper proposes QP-BD3QN-RACH, a quota-projected branching deep reinforcement learning controller for mixed two-step (2RA) and four-step (4RA) contention-based random access. Four action branches correspond to the delay-sensitive and delay-tolerant 2RA/4RA preamble pools. A branching dueling Double DQN selects pool-specific multipliers, and deterministic quota projection converts them to nonnegative integer allocations that preserve the preamble budget. With five actions per branch, the controller represents 625 pre-projection branch-action tuples using 20 branch-action outputs. Evaluation covers five arrival loads, cross-method comparison under nominal seed 42, six-seed sensitivity of QP-BD3QN-RACH, and targeted ablations. Across the five-load grid, its mean direction-aligned differences relative to four comparators are positive: 5.74 to 8.21 percentage points for success/collision, 1.23 to 1.92 percentage points for fallback, 0.35 to 0.68 percentage points for blocking, and 0.128 to 0.456 decision intervals for successful-access delay. Load-wise results exhibit metric-dependent tradeoffs, particularly under intermediate and overload conditions.

cs.NI↗

Multi-copy Axion Transfer Function and Observational Implications of Effective de Broglie Scales

Ultra-light axions are viable fuzzy/wave-like dark matter ($ψ$DM) candidates generically predicted by the String Axiverse paradigm with multiple particle copies, whereas most of the discussions/constraints on $ψ$DM from astronomical observations to date are based on the assumption of a single particle copy. Here, we aim to complete this gap by exploring the generic multi-axion scenario motivated in the String Axiverse context, and investigate its astronomical implications in both the linear and nonlinear regimes. In the linear regime, with linear density perturbation analysis, we provide a simplified prescription for obtaining multi-copy axion transfer functions and also identify an "equivalence" among all axion copies owing to the mutual coupling to the gravitational potential. As a result of this 'equivalence', we argue the suppression to LSS is governed by an effective mass $m_{eff}^{-2}=\sum_i w_i m_i^{-2}$, with $\{ w_i\}$ being fractional contributions of different copies to the full cosmic dark matter density. In non-linear regime within galaxy halos, we show that similar notions of effective mass, with expressions provided, to govern the collective wave interference and hence determine the net stellar heating rates and the substructure-induced spread of JWST transients near critical curves. Distinctive to the multi-copy scenario, the effective mass within galaxy halos is generically anticipated to be radially decreasing following the stronger concentration of heavier copies to the galactic center. Such a spatial variation leads to radially increasing spreading scales for micro-lensed transients at different radial positions, a signature that may be tested with future JWST lensing observations.

astro-ph.CO↗

Beyond Score Prediction: LLM-Based Essay Scoring and Feedback Generation via Reinforcement Learning with Rubric Rewards

Large language models (LLMs) have been widely applied to automated essay scoring (AES) and automated feedback generation (AFG). However, existing studies rely primarily on prompt engineering or supervised fine-tuning, while systematic research on reinforcement learning (RL) post-training and automated evaluation of feedback quality remains limited. We propose RLAES, a unified LLM framework that jointly optimizes essay scoring and feedback generation through RL. To make feedback quality measurable, interpretable, and usable for training, we introduce Rubric-based Feedback Evaluation (RFE), an essay-grounded feedback evaluation framework comprising 166 fine-grained binary rubric items and an LLM-as-judge. Building on RFE, we propose Adaptive Gated Feedback Optimization (AGFO), which activates rubric-based feedback rewards on demand during RL, reducing evaluation overhead while improving feedback quality. We also propose Adjacent Contrastive Reasoning (ACR) to improve ordinal score calibration by explicitly contrasting adjacent score levels. Experimental results show that the RFE framework captures essay-feedback consistency, exhibits strong pairwise discriminative power, and closely aligns with expert preferences. On the ASAP benchmark, RLAES-AGFO achieves the best scoring performance among LLM-based methods (QWK = 0.803), while maintaining feedback quality comparable to GPT-5.5 and avoiding the feedback degradation observed under score-only RL. Code and datasets are publicly available at https://github.com/hellomuyi/RLAES.

cs.CL↗

Revisiting Performance Claims for Chest X-Ray Models Using Clinical Context

Public datasets of Chest X-Rays (CXRs) have long been a popular benchmark for developing machine learning (ML) computer vision models in healthcare. However, the reported strong average-case performance of these models do not necessarily reflect their actual utility when used in heterogeneous clinical settings, potentially masking weaker performance in medically significant scenarios. In this work we use clinical context to provide a more holistic evaluation of models for CXR diagnosis. In particular, we use discharge summaries, recorded prior to each CXR, to derive a ``pre-CXR'' probability of each CXR label, as a proxy for existing contextual knowledge available to clinicians when interpreting CXRs. We use this measure to probe model performance along two dimensions: First, using a stratified analysis, we show that models tend to have lower performance (as measured by AUROC and other metrics) among individuals with higher pre-CXR probability. Second, by controlling for pre-CXR probability via matching and re-weighting, we demonstrate that performance degrades when the correlation is broken between prior context and the current CXR label, suggesting that model performance is highly sensitive to the underlying distribution of clinical context. Specifically, cases with high pre-test probabilities present a fundamentally more difficult visual classification task, highlighting a gap in clinical utility when models are applied to high-risk cohorts.

cs.LG↗

When Large Language Models Meet UAV Projects: An Empirical Study from Developers' Perspective

In recent years, unmanned aerial vehicles (UAVs) have become increasingly popular in our daily lives and have attracted significant research interest in software engineering. At the same time, large language models (LLMs) have made notable advancements in language understanding, reasoning, and generation, making LLM applications in UAVs a promising research direction. However, existing studies have largely remained in preliminary exploration with a limited understanding of real-world practice, which causes an academia-industry gap and hinders the application of LLMs in UAVs. To address this, we conducted the first empirical study to investigate how LLMs support UAVs. To characterize common tasks and application scenarios of real-world UAV-LLM practices, we conducted a large-scale empirical study involving 997 research papers and 1,509 GitHub projects. The results classified nine common tasks (e.g., Natural Language Command Parsing) in four UAV workflows (e.g., Information Input) undertaken by LLMs in real-world UAV projects and revealed a large difference in the task distribution of research efforts and industry practices. To gain deeper insight into these differences and understand developers' perspectives on the application of LLMs in UAVs, we conducted a survey of practitioners, receiving 52 valid responses from 15 countries. The results revealed that while 40.4% of developers have attempted to apply LLMs to UAV tasks, 59.6% still face challenges integrating their UAV projects with advanced LLM capabilities. Their feedback attributes these challenges to five factors, including technological maturity, performance, safety, cost, and others, and provides practical implications for researchers and developers in conducting UAV-LLM practices.

cs.SE↗

Explainable Cross-Disease Reasoning for Cardiovascular Risk Assessment from Low-Dose Computed Tomography

Low-dose chest computed tomography (LDCT) captures pulmonary and cardiac structures in a single scan, enabling joint assessment of lung and cardiovascular health. Existing approaches typically model these domains independently and do not explicitly represent their physiological interactions. We propose an Explainable Cross-Disease Reasoning Framework for cardiovascular risk assessment from LDCT. The framework follows a constrained clinical-information pathway: it extracts pulmonary findings, grounds cross-organ mechanisms in medical knowledge, and produces a cardiovascular prediction with a natural-language rationale. It combines four components: a frozen lung-risk prior, a pulmonary perception module, an agentic reasoning module, and a cardiac subvolume feature extractor. Their outputs are fused to integrate localized cardiac evidence with mechanism-level pulmonary context. On the National Lung Screening Trial cohort, the framework achieves an AUC of 0.919 for CVD screening and up to 0.838 for CVD mortality prediction, outperforming cardiac-specific, single-disease, and foundation-model baselines. Targeted controls indicate that the gains are not explained by additional thoracic visual features alone, fixed rule propagation, or a single reasoning backend. The proposed framework thus provides an auditable approach to cross-disease cardiovascular risk assessment from LDCT.

cs.CV↗

Lensed stars in galaxy-galaxy strong lensing -- a JWST prediction for the Cosmic Horseshoe

We explore for the first time the possibility of detecting lensed star transients in galaxy-galaxy strong lensing systems upon repeated, deep imaging using the {\it James-Webb Space Telescope} ({\it JWST}). Our calculation predicts that the extremely high recent star formation rate of $\sim 140\,M_{\odot}\textrm{yr}^{-1}$ over the last 50 Myr (not accounting for image multiplicity) in the ``Cosmic Horseshoe'' lensed system ($z = 2.381$) generates many young, bright stars, of which their large abundance is expected to lead to a detection rate of $\sim 60$ transients per pointing in {\it JWST} observations with a $5σ$ limiting magnitude of $\sim 29\,m_{AB}$. With the high expected detection rate and little room for uncertainty for the lens model compared with cluster lenses, our result suggests that the Cosmic Horseshoe could be an excellent tool to test the nature of dark matter based on the spatial distribution of transients, and can be used to constrain axion mass if dark matter is constituted of ultra-light axions. We also argue that the large distance modulus of $\sim46.5\,$mag at $z \approx 2.4$ can act as a filter to screen out less massive stars as transients and allow one to better constrain the high-mass end of the stellar initial mass function based on the transient detection rate. Follow-up {\it JWST} observations of the Cosmic Horseshoe would allow one to better probe the nature of dark matter and the star formation properties, such as the initial mass function at the cosmic noon, via lensed star transients.

astro-ph.CO↗

Reliable Tests of Faint-end UV Luminosity Functions in Strong Lensing Fields

Dark matter comprises ~85% of the entire mass of the Universe, but the fundamental nature of its constituent particles remains elusive. In this thesis, I test for two competitive dark matter models: the conventional heavy particle paradigm, and dark matter being ultralight bosons of mass $\sim 10^{-22}$eV ($ψ$DM). More specifically, I test for the faint-end turnover induced by $ψ$DM models, exploiting the strong lensing power by massive galaxy clusters to probe intrinsically fainter magnitudes. A key challenge for such an analysis would be contamination by low-z galaxies sharing similar observed SEDs as high-z galaxies. As I will demonstrate, such a contamination issue is generally severe and may wash out the faint-end turnover signatures. I also show that $\sim 50\%$ of the purported $3.5\leq z\leq 5.5$ galaxies within existing photometric redshift catalogs constructed for Hubble Frontier Fields (HFF) are in fact low-z interlopers. Luckily, individual mitigation of interlopers can be achieved with the combination of deep HST and JWST observations. For fields without supplementary data, machine learning methods will be shown useful in preserving the mitigating power. Cleaner $3.5\leq z\leq 5.5$ and $6\leq z\leq 10$ samples are derived for a more reliable test in strong lensing field of MACS J0416, with which I found no evidence for faint-end turnovers, leading to a constraint on the $ψ$DM mass of $>2.97\times10^{-22}$eV at 95\% confidence. This constraint will also be interpreted in an scheme where dark matter is composed of multiple particle copies, where I argue the derived mass bound is likely on an effective de Broglie scale governing the collective behavior of the entire $ψ$DM budget under gravitational equilibrium established.

astro-ph.CO↗

KLAN: Kuaishou Landing-page Adaptive Navigator

Modern online platforms configure multiple pages to accommodate diverse user needs. This multi-page architecture inherently establishes a two-stage interaction paradigm between the user and the platform: (1) Stage I: page navigation, navigating users to a specific page and (2) Stage II: in-page interaction, where users engage with customized content within the specific page. While the majority of research has been focusing on the sequential recommendation task that improves users' feedback in Stage II, there has been little investigation on how to achieve better page navigation in Stage I. To fill this gap, we formally define the task of Personalized Landing Page Modeling (PLPM) into the field of recommender systems: Given a user upon app entry, the goal of PLPM is to proactively select the most suitable landing page from a set of candidates (e.g., functional tabs, content channels, or aggregation pages) to optimize the short-term PDR metric and the long-term user engagement and satisfaction metrics, while adhering to industrial constraints. Additionally, we propose KLAN (Kuaishou Landing-page Adaptive Navigator), a hierarchical solution framework designed to provide personalized landing pages under the formulation of PLPM. KLAN comprises three key components: (1) KLAN-ISP captures inter-day static page preference; (2) KLAN-IIT captures intra-day dynamic interest transitions and (3) KLAN-AM adaptively integrates both components for optimal navigation decisions. Extensive online experiments conducted on the Kuaishou platform demonstrate the effectiveness of KLAN, obtaining +0.205% and +0.192% improvements on in Daily Active Users (DAU) and user Lifetime (LT). Our KLAN is ultimately deployed on the online platform at full traffic, serving hundreds of millions of users. To promote further research in this important area, we will release our dataset and code upon paper acceptance.

cs.IR↗

ATLAS: Autoformalizing Theorems through Lifting, Augmentation, and Synthesis of Data

Autoformalization, the automatic translation of mathematical content from natural language into machine-verifiable formal languages, has seen significant progress driven by advances in large language models (LLMs). Nonetheless, a primary barrier to further improvements is the limited availability of parallel corpora that map informal mathematical text to its formal counterpart. To address this limitation, we propose ATLAS (Autoformalizing Theorems through Lifting, Augmentation, and Synthesis of Data), a novel data generation framework designed to produce large-scale, high-quality parallel corpora of theorem statements. Distinct from prior approaches, ATLAS begins with a concept repository, accelerates the improvement of the student model through expert iteration combined with knowledge distillation, and introduces two novel augmentation strategies that exploit the structural characteristics of formal languages. Running the proposed ATLAS framework for 10 iterations, we construct an undergraduate-level dataset of 117k theorem statements and develop the ATLAS Translator by fine-tuning Llama3.1-8B-Instruct with LoRA. This model establishes a new state of the art, demonstrating statistically significant improvements over both the Herald Translator and the Kimina-Autoformalizer across all benchmarks (p<0.05, two-sided t-test). Furthermore, we demonstrate that the full-parameter fine-tuning of a stronger base model on the ATLAS dataset leads to superior performance. The datasets, model, and code are available at https://github.com/XiaoyangLiu-sjtu/ATLAS.

cs.CL↗