arXiv subjects

Song Xu

Publications and source records attributed to Song Xu.

14 recordsLinked to original sources

arXiv2026

The Stability of the Backward Problem for Photoacoustic Imaging in Attenuating Media via Carleman Estimates

This paper investigates the backward problem in time for photoacoustic tomography (PAT) in attenuating media. It is well-established that photoacoustic imaging in attenuating media can be accurately modeled by spatial fractional-order damping. This inverse problem is ill-posed in the sense of Hadamard. In this work, we construct a novel class of Carleman estimates independent of spatial variables, and by virtue of these estimates, we establish conditional stability estimates for this problem for the first time. Building upon this, we propose a Tikhonov-type regularization functional and derive its associated adjoint system. Furthermore, leveraging the established conditional stability results, we derive the convergence rate of the proposed regularization approach. Finally, we validate the effectiveness of our theoretical findings through extensive numerical experiments.

math.AP↗

arXiv2025

R-Log: Incentivizing Log Analysis Capability in LLMs via Reasoning-based Reinforcement Learning

The growing complexity of log data in modern software systems has prompted the use of Large Language Models (LLMs) for automated log analysis. Current approaches typically rely on direct supervised fine-tuning (SFT) on log-label pairs. However, this exacerbates the domain discrepancy between general-purpose LLMs and specialized log data, causing overfitting. Furthermore, SFT's imbalanced loss computation often allows lengthy contexts to overwhelm critical, concise details in model answers, leading to hallucinations. To address these limitations, we propose R-Log, a novel reasoning-based paradigm that mirrors the structured, step-by-step analytical process of human engineers. This approach enhances generalizability by learning the underlying rules behind conclusions. We further employ Reinforcement Learning (RL) to optimize the model within a simulated O&M environment, thereby reducing hallucinations by directly rewarding correct outcomes. R-Log is first cold-started on a curated dataset of 2k+ reasoning trajectories, guided by 13 strategies from manual O&M practices, to establish an initial reasoning capability. This ability is then refined via RL using a joint reward function. Empirical evaluations on real-world logs show that R-Log outperforms existing methods across five log analysis tasks, particularly in unseen scenarios (by 228.05%). We also designed R-Log-fast with 5x speedup while keeping 93% of the efficacy.

cs.SE↗

arXiv2025

M-DaQ: Retrieving Samples with Multilingual Diversity and Quality for Instruction Fine-Tuning Datasets

Multilingual instruction fine-tuning (IFT) empowers large language models to generalize across diverse linguistic and cultural contexts; however, high-quality, systematically curated multilingual IFT datasets remain scarce. To address this gap, we propose M-DaQ (Multilingual Diversity and Quality), a diversity-aware sampling framework that jointly optimizes instruction-response quality and cross-lingual semantic diversity. M-DaQ leverages a fine-tuned Quality Scoring Model alongside a maximal marginal relevance-inspired selection strategy to construct balanced, high-fidelity training data. Furthermore, we present the first systematic investigation of the Superficial Alignment Hypothesis in multilingual settings. Extensive evaluations across 18 languages demonstrate that models trained on M-DaQ-curated data achieve average win rates exceeding 60% against strong baselines on Alpaca-Eval and MT-Bench. Complementary human evaluations corroborate these gains, highlighting significant improvements in cultural relevance, contextual appropriateness, and instruction-following capability. The code are publicly released to facilitate reproducibility and future research.

cs.CL↗

arXiv2025

RationAnomaly: Log Anomaly Detection with Rationality via Chain-of-Thought and Reinforcement Learning

Logs constitute a form of evidence signaling the operational status of software systems. Automated log anomaly detection is crucial for ensuring the reliability of modern software systems. However, existing approaches face significant limitations: traditional deep learning models lack interpretability and generalization, while methods leveraging Large Language Models are often hindered by unreliability and factual inaccuracies. To address these issues, we propose RationAnomaly, a novel framework that enhances log anomaly detection by synergizing Chain-of-Thought (CoT) fine-tuning with reinforcement learning. Our approach first instills expert-like reasoning patterns using CoT-guided supervised fine-tuning, grounded in a high-quality dataset corrected through a rigorous expert-driven process. Subsequently, a reinforcement learning phase with a multi-faceted reward function optimizes for accuracy and logical consistency, effectively mitigating hallucinations. Experimentally, RationAnomaly outperforms state-of-the-art baselines, achieving superior F1-scores on key benchmarks while providing transparent, step-by-step analytical outputs. We have released the corresponding resources, including code and datasets.

cs.AI↗

arXiv2021

K-PLUG: Knowledge-injected Pre-trained Language Model for Natural Language Understanding and Generation in E-Commerce

Existing pre-trained language models (PLMs) have demonstrated the effectiveness of self-supervised learning for a broad range of natural language processing (NLP) tasks. However, most of them are not explicitly aware of domain-specific knowledge, which is essential for downstream tasks in many domains, such as tasks in e-commerce scenarios. In this paper, we propose K-PLUG, a knowledge-injected pre-trained language model based on the encoder-decoder transformer that can be transferred to both natural language understanding and generation tasks. We verify our method in a diverse range of e-commerce scenarios that require domain-specific knowledge. Specifically, we propose five knowledge-aware self-supervised pre-training objectives to formulate the learning of domain-specific knowledge, including e-commerce domain-specific knowledge-bases, aspects of product entities, categories of product entities, and unique selling propositions of product entities. K-PLUG achieves new state-of-the-art results on a suite of domain-specific NLP tasks, including product knowledge base completion, abstractive product summarization, and multi-turn dialogue, significantly outperforms baselines across the board, which demonstrates that the proposed method effectively learns a diverse set of domain-specific knowledge for both language understanding and generation tasks.

cs.CL↗

arXiv2019

Diversity in Biology: definitions, quantification, and models

Diversity indices are useful single-number metrics for characterizing a complex distribution of a set of attributes across a population of interest. The utility of these different metrics or sets of metrics depend on the context and application, and whether a predictive mechanistic model exists. In this topical review, we first summarize the relevant mathematical principles underlying heterogeneity in a large population before outlining the various definitions of `diversity' and providing examples of scientific topics in which its quantification plays an important role. We then review how diversity has been a ubiquitous concept across multiple fields including ecology, immunology, cellular barcoding experiments, and socioeconomic studies. Since many of these applications involve sampling of populations, we also review how diversity in small samples is related to the diversity in the entire population. Features that arise in each of these applications are highlighted.

q-bio.PE↗

arXiv2018

Immigration-induced phase transition in a regulated multispecies birth-death process

Power-law-distributed species counts or clone counts arise in many biological settings such as multispecies cell populations, population genetics, and ecology. This empirical observation that the number of species $c_{k}$ represented by $k$ individuals scales as negative powers of $k$ is also supported by a series of theoretical birth-death-immigration (BDI) models that consistently predict many low-population species, a few intermediate-population species, and very high-population species. However, we show how a simple global population-dependent regulation in a neutral BDI model destroys the power law distributions. Simulation of the regulated BDI model shows a high probability of observing a high-population species that dominates the total population. Further analysis reveals that the origin of this breakdown is associated with the failure of a mean-field approximation for the expected species abundance distribution. We find an accurate estimate for the expected distribution $\langle c_k \rangle$ by mapping the problem to a lower-dimensional Moran process, allowing us to also straightforwardly calculate the covariances $\langle c_k c_\ell \rangle$. Finally, we exploit the concepts associated with energy landscapes to explain the failure of the mean-field assumption by identifying a phase transition in the quasi-steady-state species counts triggered by a decreasing immigration rate.

q-bio.PE↗

arXiv2016

DDVV-type inequality for Hermitian matrices

In this paper we extend DDVV-type inequalities involving the Frobenius norm of commutators from real symmetric and skew-symmetric matrices to Hermitian and skew-Hermitian matrices.

math.DG↗

arXiv2016

An effective numerical method to solve a class of nonlinear singular boundary value problems using improved differential transform method

In this work, an effective numerical method is developed to solve a class of singular boundary value problems arising in various physical models by using the improved differential transform method (IDTM). The IDTM applies the Adomian polynomials to handle the differential transforms of the nonlinearities arising in the given differential equation. The relation between the Adomian polynomials of those nonlinear functions and the coefficients of unknown truncated series solution is given by a simple formula, through which one can easily deduce the approximate solution which takes the form of a convergent series. An upper bound for the estimation of approximate error is presented. Several physical problems are discussed as illustrative examples to testify the validity and applicability of the proposed method. Comparisons are made between the present method and the other existing methods.

math.NA↗

arXiv2013

Landscape construction in non-gradient dynamics: A case from evolution

Adaptive landscape has been a fundamental concept in many branches of modern biology since Wright's first proposition in 1932. Meanwhile, the general existence of landscape remains controversial. The causes include the mixed uses of different landscape definitions with their own different aims and advantages. Sometimes the difficulty and the impossibility of the landscape construction for complex models are also equated. To clarify these confusions, based on a recent formulation of Wright's theory, the current authors construct generalized adaptive landscape in a two-loci population model with non-gradient dynamics, where the conventional gradient landscape does not exist. On the generalized landscape, a population moves along an evolutionary trajectory which always increases or conserves adaptiveness but does not necessarily follow the steepest gradient direction. Comparisons of different aspects of various landscapes lead to a conclusion that the generalized landscape is a possible direction to continue the exploration of Wright's theory for complex dynamics.

q-bio.PE↗

arXiv2013

Two-timescale evolution on a singular landscape

Under the effect of strong genetic drift, it is highly probable to observe gene fixation or gene loss in a population, shown by infinite peaks on a coherently constructed potential energy landscape. It is then important to ask what such singular peaks imply, with or without the effects of other biological factors. We studied the stochastic escape time from the infinite potential peaks in the Wright-Fisher model, where the typical two-scale diffusion dynamics was observed via computer simulations. We numerically found the average escape time for all the bi-stable cases and analytically approximated the results under weak mutations and selections by calculating the mean first passage time (MFPT) in singular potential peak. Our results showed that Kramers' classical escape formula can be extended to the models with non-Gaussian probability distributions, overcoming constraints in previous methods. The constructed landscape provides a global and coherent description for system's evolutionary dynamics, allowing new biological results to be generated.

q-bio.PE↗

arXiv2012

Non-fixation in infinite potential

Under the effects of strong genetic drift, it is highly probable to observe gene fixation or loss in a population, shown by divergent probability density functions, or infinite adaptive peaks on a landscape. It is then interesting to ask what such infinite peaks imply, with or without combining other biological factors (e.g. mutation and selection). We study the stochastic escape time from the generated infinite adaptive peaks, and show that Kramers' classical escape formula can be extended to the non-Gaussian distribution cases. The constructed landscape provides a global description for system's middle and long term behaviors, breaking the constraints in previous methods.

q-bio.PE↗

arXiv2009

Simplex Subdivisions and Nonnegativity Decision of Forms

This paper mainly studies nonnegativity decision of forms based on variable substitutions. Unlike existing research, the paper regards simplex subdivisions as new perspectives to study variable substitutions, gives some subdivisions of the simplex T_n, introduces the concept of convergence of the subdivision sequence, and presents a sufficient and necessary condition for the convergent self-similar subdivision sequence. Then the relationships between subdivisions and their corresponding substitutions are established. Moreover, it is proven that if the form F is indefinite on T_n and the sequence of the successive L-substitution sets is convergent, then the sequence of sets {SLS^(m)(F)} is negatively terminating, and an algorithm for deciding indefinite forms with a counter-example is obtained. Thus, various effective substitutions for deciding positive semi-definite forms and indefinite forms are gained, which are beyond the weighted difference substitutions characterized by "difference".

cs.SC↗

arXiv2009

The weighted difference substitutions and Nonnegativity Decision of Forms

In this paper, we study the weighted difference substitutions from geometrical views. First, we give the geometric meanings of the weighted difference substitutions, and introduce the concept of convergence of the sequence of substitution sets. Then it is proven that the sequence of the successive weighted difference substitution sets is convergent. Based on the convergence of the sequence of the successive weighted difference sets, a new, simpler method to prove that if the form F is positive definite on T_n, then the sequence of sets {SDS^m(F)} is positively terminating is presented, which is different from the one given in [11]. That is, we can decide the nonnegativity of a positive definite form by successively running the weighted difference substitutions finite times. Finally, an algorithm for deciding an indefinite form with a counter-example is obtained, and some examples are listed by using the obtained algorithm.

cs.SC↗