arXiv.itsnick.co

an independent field guide to 23 million papers from arXiv & OpenAlex — not affiliated with arXiv.org

Author

Jiaming Luo

44 papers · 2017–2026

2017: 12018: 02019: 12020: 22021: 12022: 32023: 32024: 72025: 162026: 10
2017peak 16 in 20252026

Frequent co-authors

Papers

showing 1–25 of 44 RSS

Gemma 3 Technical Report

Gemma Team, Aishwarya Kamath, Johan Ferret +213 more

cs.CL · · cited 1,943× · arXiv ↗ · more like this ·

We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achieved by increasing the ratio of local to global attention layers, and keeping the span on local attention short. The Gemma 3 models are trained with distillation and achieve superior performance to Gemma 2 for both pre-trained and instruction finetuned versions. In particular, our novel post-training recipe significantly improves the math, chat, instruction-following and multilingual abilities, making Gemma3-4B-IT competitive with Gemma2-27B-IT and Gemma3-27B-IT comparable to Gemini-1.5-Pro…

Prompting PaLM for Translation: Assessing Strategies and Performance

David Vilar, Markus Freitag, Colin Cherry +3 more

cs.CL · · cited 245× · arXiv ↗ · more like this ·

Large language models (LLMs) that have been trained on multilingual but not parallel text exhibit a remarkable ability to translate between languages. We probe this ability in an in-depth study of the pathways language model (PaLM), which has demonstrated the strongest machine translation (MT) performance among similarly-trained LLMs to date. We investigate various strategies for choosing translation examples for few-shot prompting, concluding that example quality is the most important factor. Using optimized prompts, we revisit previous assessments of PaLM's MT capabilities with more recent test sets, modern MT metrics, and human evaluation, and find that its performance, while impressive, still lags that of state-of-the-art supervised systems. We conclude by providing an analysis of PaLM's MT output which reveals some interesting properties and prospects for future work.

Out-of-Distribution Detection and Selective Generation for Conditional Language Models

Jie Ren, Jiaming Luo, Yao Zhao +4 more

cs.CL · · cited 237× · arXiv ↗ · more like this ·

Machine learning algorithms typically assume independent and identically distributed samples in training and at test time. Much work has shown that high-performing ML classifiers can degrade significantly and provide overly-confident, wrong classification predictions, particularly for out-of-distribution (OOD) inputs. Conditional language models (CLMs) are predominantly trained to classify the next token in an output sequence, and may suffer even worse degradation on OOD inputs as the prediction is done auto-regressively over many steps. Furthermore, the space of potential low-quality outputs is larger as arbitrary text can be generated and it is important to know when to trust the generated output. We present a highly accurate and lightweight OOD detection method for CLMs, and demonstrate its effectiveness on abstractive summarization and translation. We also show how our method can be…

Neural Decipherment via Minimum-Cost Flow: from Ugaritic to Linear B

Jiaming Luo, Yuan Cao, Regina Barzilay

cs.CL · · cited 29× · arXiv ↗ · more like this ·

In this paper we propose a novel neural approach for automatic decipherment of lost languages. To compensate for the lack of strong supervision signal, our model design is informed by patterns in language change documented in historical linguistics. The model utilizes an expressive sequence-to-sequence model to capture character-level correspondences between cognates. To effectively train the model in an unsupervised manner, we innovate the training procedure by formalizing it as a minimum-cost flow problem. When applied to the decipherment of Ugaritic, we achieve a 5.5% absolute improvement over state-of-the-art results. We also report the first automatic results in deciphering Linear B, a syllabic language related to ancient Greek, where our model correctly translates 67.3% of cognates.

Deciphering Undersegmented Ancient Scripts Using Phonetic Prior

Jiaming Luo, Frederik Hartmann, Enrico Santus +2 more

cs.CL · · cited 25× · arXiv ↗ · more like this ·

Most undeciphered lost languages exhibit two characteristics that pose significant decipherment challenges: (1) the scripts are not fully segmented into words; (2) the closest known language is not determined. We propose a decipherment model that handles both of these challenges by building on rich linguistic constraints reflecting consistent patterns in historical sound change. We capture the natural phonological geometry by learning character embeddings based on the International Phonetic Alphabet (IPA). The resulting generative framework jointly models word segmentation and cognate alignment, informed by phonological constraints. We evaluate the model on both deciphered languages (Gothic, Ugaritic) and an undeciphered one (Iberian). The experiments show that incorporating phonetic geometry leads to clear and consistent gains. Additionally, we propose a measure for language closeness…

Unsupervised Learning of Morphological Forests

Jiaming Luo, Karthik Narasimhan, Regina Barzilay

cs.CL · · cited 11× · arXiv ↗ · more like this ·

This paper focuses on unsupervised modeling of morphological families, collectively comprising a forest over the language vocabulary. This formulation enables us to capture edgewise properties reflecting single-step morphological derivations, along with global distributional properties of the entire forest. These global properties constrain the size of the affix set and encourage formation of tight morphological families. The resulting objective is solved using Integer Linear Programming (ILP) paired with contrastive estimation. We train the model by alternating between optimizing the local log-linear model and the global ILP objective. We evaluate our system on three tasks: root detection, clustering of morphological families and segmentation. Our experiments demonstrate that our model yields consistent gains in all three tasks compared with the best published results.

The impact of parental absence on the mental health of middle school students in rural areas of Western China

Xiaohong Ren, Cen Lin, Lu Pan +5 more

Sociology and Political Science · Frontiers in Public Health · · cited 10× · DOI ↗ · more like this ·

Background Extensive research has established the association between parental absence and adolescent psychological well-being, particularly in the Chinese context. However, studies specifically examining the dual impact of parental separation and migration on psychological outcomes among adolescents in Western China remain relatively limited. Aim This study aims to systematically examine the association between various parental absence situations and mental health outcomes in early adolescence, with the objective of informing targeted interventions and policy formulations to optimize psychosocial support systems for vulnerable youth population. Methods The Wilcoxon rank-sum test was employed to analyze continuous and ordinal variables that exhibited non-normal distributions. To investigate the associations between various patterns of parental absence and psychological outcomes…

To Diverge or Not to Diverge: A Morphosyntactic Perspective on Machine Translation vs Human Translation

Jiaming Luo, Colin Cherry, George Foster

cs.CL · · cited 8× · arXiv ↗ · more like this ·

We conduct a large-scale fine-grained comparative analysis of machine translations (MT) against human translations (HT) through the lens of morphosyntactic divergence. Across three language pairs and two types of divergence defined as the structural difference between the source and the target, MT is consistently more conservative than HT, with less morphosyntactic diversity, more convergent patterns, and more one-to-one alignments. Through analysis on different decoding algorithms, we attribute this discrepancy to the use of beam search that biases MT towards more convergent patterns. This bias is most amplified when the convergent pattern appears around 50% of the time in training data. Lastly, we show that for a majority of morphosyntactic divergences, their presence in HT is correlated with decreased MT performance, presenting a greater challenge for MT systems.

Cross-lagged panel analysis of the relationship between social networking sites use (SNSU) and sleep problems among university students

Xiaoling Li, Hai‐Lei Li, Jiaming Luo

Sociology and Political Science · BMC Public Health · · cited 6× · DOI ↗ · more like this ·

BACKGROUND: Sleep remains a cornerstone for sociopsychological well-being, but it is in decline, especially among today's youth. Simultaneously, engagement with social media is escalating. Research has identified a link between social networking sites use and sleep problems; however, the nature and direction of the relations remain obscure. Therefore, it is imperative to pursue longitudinal research to elucidate this correlation and guide suitable intervention practices. The present study aimed to examine the reciprocal relationship between social networking sites use and sleep problems. METHODS: By adopting a three-stage cross-lagged design across four time points, assessment results from 194 university students were gathered at four-week intervals. RESULTS: The findings indicate that (1) Social networking sites use was significantly greater in females than in males at all four time…

Exploring the mental health issues of left-behind children based on propensity score matching: the parallel mediating role of psychological abuse and neglect

Cen Lin, Yuqin Song, Lu Pan +8 more

Safety Research · Child and Adolescent Psychiatry and Mental Health · · cited 3× · DOI ↗ · more like this ·

INTRODUCTION: The underdeveloped western region of China, where most of the country’s migrant workers originate, has many children who remain at home while their parents work elsewhere. These left-behind children often suffer from significant psychological issues, including depression, anxiety, stress, and sleep quality problems. The pathways through which being left behind leads to these mental health challenges are not well understood. This study employs Propensity Score Matching (PSM) to adjust for confounders and investigate the mediating effects of psychological abuse and neglect on these mental health issues. METHODS: The study surveyed 17 middle schools in western China using convenience cluster sampling. It gathered data on demographics, the Child Psychological Abuse and Neglect Scale (CPANS), the Depression Anxiety Stress Scale (DASS-21), and the Pittsburgh Sleep Quality Index…

Prediction of the effect of bike-sharing on urban carbon emission reduction: Evidence from Beijing

Huijun Zhao, Zhong Lin Wang, Jiaming Luo +1 more

Transportation · Journal of Urban Management · · cited 3× · DOI ↗ · more like this ·

Bike-sharing is regarded as a green and economical public transport mode. However, how to scientifically predict its carbon emission reduction effect is still on the way. This article builds an assessment model for emission reduction of bike-sharing. The modeling is based on the rational person hypothesis that consumers take transportation consumption expenditure as a budget constraint, and get the optimal number of trips of shared bikes for utility maximization. Then, we take Beijing as an example to carry out empirical research, whereby data on key influencing factors were collected through questionnaires and situational analysis. Finally, it is predicted that the carbon emission reduction contributed by bike-sharing in Beijing will be approximately 522,540.32–1,135,957.21 tons in 2025–2030. This study can provide a benchmark theoretical model for urban emission reduction, as well as…

Improving the Robustness of Summarization Models by Detecting and Removing Input Noise

Kundan Krishna, Yao Zhao, Jie Ren +4 more

cs.CL · · cited 2× · arXiv ↗ · more like this ·

The evaluation of abstractive summarization models typically uses test data that is identically distributed as training data. In real-world practice, documents to be summarized may contain input noise caused by text extraction artifacts or data pipeline bugs. The robustness of model performance under distribution shift caused by such noise is relatively under-studied. We present a large empirical study quantifying the sometimes severe loss in performance (up to 12 ROUGE-1 points) from different types of input noise for a range of datasets and model sizes. We then propose a light-weight method for detecting and removing such noise in the input during model inference without requiring any extra training, auxiliary models, or even prior knowledge of the type of noise. Our proposed approach effectively mitigates the loss in performance, recovering a large fraction of the performance drop…

Leveraging Domain Knowledge at Inference Time for LLM Translation: Retrieval versus Generation

Bryan Li, Jiaming Luo, Eleftheria Briakou +1 more

cs.CL · · cited 1× · arXiv ↗ · more like this ·

While large language models (LLMs) have been increasingly adopted for machine translation (MT), their performance for specialist domains such as medicine and law remains an open challenge. Prior work has shown that LLMs can be domain-adapted at test-time by retrieving targeted few-shot demonstrations or terminologies for inclusion in the prompt. Meanwhile, for general-purpose LLM MT, recent studies have found some success in generating similarly useful domain knowledge from an LLM itself, prior to translation. Our work studies domain-adapted MT with LLMs through a careful prompting setup, finding that demonstrations consistently outperform terminology, and retrieval consistently outperforms generation. We find that generating demonstrations with weaker models can close the gap with larger model's zero-shot performance. Given the effectiveness of demonstrations, we perform detailed…

Situating Vulnerability: An Ethical Governance Framework for Anthropomorphic AI Interaction

Hanhui Xu, Jiacheng Ji, Yuhan Wang +3 more

cs.CY · · cited 0× · arXiv ↗ · more like this ·

Large language models increasingly address users in the register of care: they remember personal details, adopt a first-person emotional voice, and sustain relationships across many sessions. This anthropomorphic mode of interaction generates risks that output-level metrics capture poorly, because the same empathic utterance may console one user, entrap a second, and consolidate a delusion in a third. Governance has responded in two ways. Uniform restriction refuses sensitive requests or suppresses empathic phrasing across all outputs; group-targeted protection profiles vulnerable populations and applies stricter or prohibitive rules to them. This article argues that both fail. Uniform restriction mislocates harm in the lexical form of responses rather than in the trajectory of a relationship; group-based protection depends on a list that can never be complete, treats a fluid property…

When Does Reasoning Help in Machine Translation? A Hierarchical Analysis of LRM Reasoning Traces

Yuxiang Liu, Jiaming Luo, Eleftheria Briakou +1 more

cs.CL · · cited 0× · arXiv ↗ · more like this ·

Large Reasoning Models increasingly use intermediate traces for machine translation, but it remains unclear when such reasoning helps or hurts. We analyze reasoning traces across models, languages, domains, and datasets, focusing on reasoning language, length, and structure. We find that the best reasoning language is model-specific, reasoning length has a non-monotonic relationship with quality, and traces exhibit recurring functional patterns. To uncover these patterns, we introduce Hierarchical Meta-Summarization (HMS), a scalable framework that induces coarse- and fine-grained reasoning structures without predefined taxonomies. HMS reveals a shared organization--understanding/planning, translating/drafting, and refining/verifying--alongside domain-specific variation. Our results suggest that MT reasoning should be controlled in a model-aware, length-aware, and pattern-aware manner…

Ultrafast Magnetization Induced from Raman-Active Axial Chiral Phonons

Chuankun Huang, Alexander Milner, Jiaming Luo +8 more

cond-mat.mtrl-sci · · cited 0× · arXiv ↗ · more like this ·

Axial chiral phonons, which carry angular momentum and exhibit unusually large magnetic moments, provide a new degree of freedom for controlling the time-reversal symmetry and magnetic properties of quantum materials. While infrared-active axial phonons are efficiently excited to large amplitudes by circularly polarized terahertz pulses, their application may be constrained by short lifetimes, small penetration depths, low spatial resolution, and limited availability of optical sources. Here, we report that an axial Raman-active phonon mode in CeF3 exhibits a significant magnetic moment and an exceedingly long lifetime that more closely matches the paramagnetic spin dynamics. The axial phonon population is resonantly driven by a near-infrared laser pulse with rotating linear polarization, known as an optical centrifuge, a coherent control scheme never applied to solids before. The…

Finite Comparison Algebras for Motivic Projectors and Triangular Stability

Jiaming Luo

Computational Theory and Mathematics · Mathematics · · cited 0× · DOI ↗ · more like this ·

We study a finite-stage criterion for representing auxiliary or motivated idempotents by ordinary motivic projectors. To a fixed finite list of operations on a motive M we attach a finite comparison algebra EΓ(M), a finite-dimensional ordinary comparison subring R, and a surjective comparison map R→EΓ(M); its kernel is the defect ideal recording the ordinary relations that become true after comparison. If this defect ideal is nilpotent, then every idempotent in the chosen finite comparison algebra has an ordinary idempotent representative in R. The same nilpotence gives unipotent conjugacy of representatives after passage to a common finite refinement, and hence a refinement-independent isomorphism class of the resulting Karoubi summand. We isolate filtered, support-theoretic, and Peirce triangular criteria for verifying defect nilpotence. In a block upper-triangular comparison algebra…

Physical exercise and problematic smartphone use among adolescents in Western China: a multi-center cross-sectional study on the serial mediation of self-control and meaning in life

Xingyu He, Min Liu, Ping Mo +2 more

Sociology and Political Science · BMC Public Health · · cited 0× · DOI ↗ · more like this ·

OBJECTIVE: Problematic smartphone use has become a prominent public health issue in the information era. Although research has shown that exercise is beneficial in alleviating problematic smartphone use among adolescents, the specific underlying mechanisms have not been fully elucidated. This study aims to explore the internal mechanisms between self-control and the meaning in life in adolescents' problematic smartphone use (PSU) through physical exercise. METHODS: A multi-center, large-sample cross-sectional study was conducted among students from 34 primary and secondary schools in western China. Assessments were conducted using the Physical Activity Rating Scale (PARS-3), the Brief Self-Control Scale (BSCS), Meaning in Life Questionnaire (MLQ), and the Smartphone Addiction Scale-Short Version (SAS-SV). Data were analyzed using the Mann-Whitney U test, correlation analysis, and…

A TOPOLOGICAL-ANALYTIC DICTIONARY, MODULI, AND CANONICAL BASEPOINTS FOR SELF-DUAL MEZZOPERVERSITIES ON CHEEGER SPACES

Jiaming Luo

Mathematical Physics · HAL (Le Centre pour la Communication Scientifique Directe) · · cited 0× · OpenAlex ↗ · more like this ·

<div> We establish an analytic-topological interface for self-dual mezzoperversities on Cheeger spaces, relating analytic ideal boundary conditions for L 2 de Rham complexes in the sense of Albin-Leichtnam-Mazzeo-Piazza (ALMP) to Banagl's topological self-dual sheaves and Lagrangian structures. The analytic-to-topological direction is unconditional. Conversely, for a Banagl datum admitting a suitable analytic realization, we construct a mezzoperversity whose Deligne-type complex agrees with the corresponding topological self-dual complex. For a fixed ALMP package, we classify analytic self-dual choices by homotopy classes of incidence-compatible sections of Lagrangian Grassmannian bundles, and we formulate the remaining global comparison problem in terms of realizability and homotopy-uniqueness. We further show that the scatteringat-zero canonical mezzoperversity constructed in [Luo26b…

SCATTERING AT ZERO AND CANONICAL SELF-DUAL MEZZOPERVERSITIES ON ADMISSIBLE CHEEGER SPACES

Jiaming Luo

Geometry and Topology · HAL (Le Centre pour la Communication Scientifique Directe) · · cited 0× · OpenAlex ↗ · more like this ·

<div> Let (X, g) be a compact smoothly stratified pseudomanifold equipped with an admissible iterated incomplete edge metric. We construct a canonical self-dual mezzoperversity on this class by taking the zero-energy limit of the low-energy scattering operator for the de Rham complex. For each singular stratum Y , this limit defines an endomorphism SY (0) of the Albin-Leichtnam-Mazzeo-Piazza trace bundle HY → Y . We prove that the associated involution field JY is fiberwise anti-symplectic for the Green form, smooth along the stratum, compatible with flat transport and incidence maps, stable under admissible deformations, and equivariant for duality. It follows that the fixed subbundles WY = Fix(JY ) determine a canonical mezzoperversity Wscatt. Invoking the Hodge-Fredholm theory of Albin-Leichtnam-Mazzeo-Piazza, we obtain Fredholmness of the resulting de Rham complex, identification…

A MAXIMAL CGM-TYPE L 2 DE RHAM THEORY FOR STRATIFIED PSEUDOMANIFOLDS WITH ITERATED INCOMPLETE EDGE METRICS

Jiaming Luo

Geometry and Topology · HAL (Le Centre pour la Communication Scientifique Directe) · · cited 0× · OpenAlex ↗ · more like this ·

<div> On a smoothly stratified pseudomanifold equipped with an iterated incomplete edge (iie) metric, the de Rham differential admits many closed L 2 extensions, so any general Cheeger-Goresky-MacPherson statement must be parameterized by ideal boundary conditions (mezzoperversities). Building on the framework of Albin-Leichtnam-Mazzeo-Piazza, we associate to each mezzoperversity W a Fredholm L 2 de Rham complex on X reg whose cohomology identifies with a Deligne-type intersection cohomology IH • W (X) and is invariant under changes of suitably scaled iie metrics within a fixed quasi-isometry class. We characterize Cheeger spaces (those admitting a selfdual W), for which Poincaré duality holds and, in dimension 4ℓ, the analytic signature agrees with Banagl's topological signature. We also exhibit an obstruction to a functorial canonical "middle" choice of W on the full iie class.…

CANONICAL AND HODGE-POLARIZED MEZZOPERVERSITIES FOR FUBINI-STUDY ITERATED WEDGE METRICS ON COMPLEX PROJECTIVE VARIETIES

Jiaming Luo

Geometry and Topology · HAL (Le Centre pour la Communication Scientifique Directe) · · cited 0× · OpenAlex ↗ · more like this ·

<div> Let X ⊂ CP N be a complex projective variety equipped with a fixed Thom-Mather stratification, and let gFS be the induced Fubini-Study iterated wedge metric on X reg . In [Luo26b, Luo26a] we construct, via iterated Calderón constructions, a canonical mezzoperversity Wcan and identify the L 2 de Rham cohomology of the corresponding canonical closed extension with IH • Wcan (X). In [Luo26c] we construct a polarized Hodge mezzoperversity W polHdg and prove ICW polHdg ∼ = IC m. In this paper we prove the analytic comparison Wcan = W polHdg without additional hypotheses. Consequently, for arbitrary stratification depth, the canonical L 2 de Rham cohomology of (X reg , gFS) identifies with the middle intersection cohomology IH • m(X ), thereby verifying the Cheeger-Goresky-MacPherson conjecture in this setting. </div>

On the Log Hodge Theory of Toroidal Varieties and a Partial Proof of the Absolute Hodge Conjecture

Jiaming Luo

math.AG · · cited 0× · arXiv ↗ · more like this ·

In this paper, we establish an innovative framework in logarithmic Hodge theory for toroidal varieties, introducing weighted toroidal structures and developing a systematic obstruction theory for Hodge classes. Building upon recent advances in toroidal geometry, particularly the $E_1$-degeneration results of Wei (\cite{Wei24}), we construct a categorical Hodge correspondence and prove the Absolute Hodge conjecture for projective toroidal varieties equipped with rational-weighted structures satisfying specific compatibility conditions. Our approach provides new tools for understanding the fine structure of Hodge classes and their extension properties across boundaries, with potential applications in moduli space compactifications, non-abelian Hodge theory, and tropical geometry.

Derived Stratifications and Arithmetic Intersection Theory for Varieties with Isolated Singularities

Jiaming Luo, Shirong Li

math.AG · · cited 0× · arXiv ↗ · more like this ·

In this paper, We develop the stratified de Rham theory on singular spaces using modern tools including derived geometry and stratified structures. This work unifies and extends the de Rham theory, Hodge theory, and deformation theory of singular spaces into the frameworks of stratified geometry, $p$-adic geometry, and derived geometry. Additionally, we close a gap in Ohsawa's original proof, concerning the convergence of $L^2$ harmonic forms in the Cheeger-Goresky-MacPherson conjecture for varieties with isolated singularities. Indicating that harmonic forms converge strongly and the $L^2$-cohomology coincides with intersection cohomology.

Derived Stratified-Microlocal Framework and Moduli Space Resolution for the Cheeger-Goresky-Macpherson Conjecture

Jiaming Luo

math.AG · · cited 0× · arXiv ↗ · more like this ·

In this paper, We define the stratified metric $\infty$-category $\mathbf{StratMet}_{\infty}$ and the middle perversity moduli stack $\mathscr{M}^{\mathrm{mid}}$. We construct a universal truncation complex $Ω_{X,\mathrm{FS}}^{\bullet,\mathrm{univ}}$ for a projective variety $X\subseteq\mathbb{P}^N$. By introducing the stratified singular characteristic variety $\mathrm{SSH}_{\mathrm{strat}}$, we establish a microlocal correspondence between metric asymptotic behavior and topology, proving the natural isomorphism $$H_2^*(X_{\mathrm{reg}}, ds_{\mathrm{FS}}^2) \cong IH^*(X,\mathbb{C}).$$ This framework transcends transverse singularity constraints, achieves moduli space parametrized duality, and develops new paradigms for high-codimension singular topology, quantum singularity theory, and $p$-adic Hodge theory.