arXiv subjects

Xiao Liu

Publications and source records attributed to Xiao Liu.

At least 181 records · Page 10Linked to original sources

arXiv2025

Read it in Two Steps: Translating Extremely Low-Resource Languages with Code-Augmented Grammar Books

While large language models (LLMs) have shown promise in translating extremely low-resource languages using resources like dictionaries, the effectiveness of grammar books remains debated. This paper investigates the role of grammar books in translating extremely low-resource languages by decomposing it into two key steps: grammar rule retrieval and application. To facilitate the study, we introduce ZhuangRules, a modularized dataset of grammar rules and their corresponding test sentences. Our analysis reveals that rule retrieval constitutes a primary bottleneck in grammar-based translation. Moreover, although LLMs can apply simple rules for translation when explicitly provided, they encounter difficulties in handling more complex rules. To address these challenges, we propose representing grammar rules as code functions, considering their similarities in structure and the benefit of code in facilitating LLM reasoning. Our experiments show that using code rules significantly boosts both rule retrieval and application, ultimately resulting in a 13.1% BLEU improvement in translation.

cs.CL↗

arXiv2025

Learning to Regulate: A New Event-Level Dataset of Capital Control Measures

We construct a novel event-level Capital Control Measures (CCM) dataset covering 196 countries from 1999 to 2023 by leveraging prompt-based large language models (LLMs). The dataset enables event study analysis and cross-country comparisons based on rich policy attributes, including action type, intensity, direction, implementing entity, and other multidimensional characteristics. Using a two-step prompt framework with GPT-4.1, we extract structured information from the IMF's Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER), resulting in 5,198 capital control events with 27 annotated fields and corresponding model reasoning. Secondly, to facilitate real-time classification and extension to external sources, we fine-tune an open-source Meta Llama 3.1-8B model, named CCM-Llama, trained on AREAER change logs and final status reports. The model achieves 90.09\% accuracy in category classification and 99.55\% in status prediction. Finally, we apply the CCM dataset in an empirical application: an event study on China, Australia, and the US. The results show that inward capital control measures significantly reduce fund inflows within one month, and restrictive policies tend to have stronger effects than liberalizing ones, with notable heterogeneity across countries. Our work contributes to the growing literature on the use of LLMs in economics by providing both a novel high-frequency policy dataset and a replicable framework for automated classification of capital control events from diverse and evolving information sources.

econ.GN↗

arXiv2025

Bias and Volatility: A Statistical Framework for Evaluating Large Language Model's Stereotypes and the Associated Generation Inconsistency

We present a novel statistical framework for analyzing stereotypes in large language models (LLMs) by systematically estimating the bias and variation in their generation. Current alignment evaluation metrics often overlook stereotypes' randomness caused by LLMs' inconsistent generative behavior. For instance, LLMs may display contradictory stereotypes, such as those related to gender or race, for identical professions in different contexts. Ignoring this inconsistency risks misleading conclusions in alignment assessments and undermines efforts to evaluate the potential of LLMs to perpetuate or amplify social biases and unfairness. To address this, we propose the Bias-Volatility Framework (BVF), which estimates the probability distribution of stereotypes in LLM outputs. By capturing the variation in generative behavior, BVF assesses both the likelihood and degree to which LLM outputs negatively impact vulnerable groups, enabling a quantification of aggregated discrimination risk. Additionally, we introduce a mathematical framework to decompose this risk into bias risk (from the mean of the stereotype distribution) and volatility risk (from its variation). Applying BVF to 12 widely used LLMs, we find: i) Bias risk is the dominant contributor to discrimination; ii) Most LLMs exhibit substantial pro-male stereotypes across nearly all professions; iii) Reinforcement learning from human feedback reduces bias but increases volatility; iv) Discrimination risk correlates with socio-economic factors, such as professional salaries. Finally, we highlight BVF's broader applicability for assessing how generation inconsistencies in LLMs impact behavior beyond stereotypes.

cs.CL↗

arXiv2025

Optimizing Large Language Model Training Using FP4 Quantization

The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse. To ensure stability, the framework integrates a mixed-precision training scheme and vector-wise quantization. Experimental results demonstrate that our FP4 framework achieves accuracy comparable to BF16 and FP8, with minimal degradation, scaling effectively to 13B-parameter LLMs trained on up to 100B tokens. With the emergence of next-generation hardware supporting FP4, our framework sets a foundation for efficient ultra-low precision training.

cs.LG↗

arXiv2025

Velocitune: A Velocity-based Dynamic Domain Reweighting Method for Continual Pre-training

It is well-known that a diverse corpus is critical for training large language models, which are typically constructed from a mixture of various domains. In general, previous efforts resort to sampling training data from different domains with static proportions, as well as adjusting data proportions during training. However, few methods have addressed the complexities of domain-adaptive continual pre-training. To fill this gap, we propose Velocitune, a novel framework dynamically assesses learning velocity and adjusts data proportions accordingly, favoring slower-learning domains while shunning faster-learning ones, which is guided by a scaling law to indicate the desired learning goal for each domain with less associated cost. To evaluate the effectiveness of Velocitune, we conduct experiments in a reasoning-focused dataset with CodeLlama, as well as in a corpus specialised for system command generation with Llama3 and Mistral. Velocitune achieves performance gains in both math and code reasoning tasks and command-line generation benchmarks. Further analysis reveals that key factors driving Velocitune's effectiveness include target loss prediction and data ordering.

cs.CL↗

arXiv2025

Degradation-Aware Feature Perturbation for All-in-One Image Restoration

All-in-one image restoration aims to recover clear images from various degradation types and levels with a unified model. Nonetheless, the significant variations among degradation types present challenges for training a universal model, often resulting in task interference, where the gradient update directions of different tasks may diverge due to shared parameters. To address this issue, motivated by the routing strategy, we propose DFPIR, a novel all-in-one image restorer that introduces Degradation-aware Feature Perturbations(DFP) to adjust the feature space to align with the unified parameter space. In this paper, the feature perturbations primarily include channel-wise perturbations and attention-wise perturbations. Specifically, channel-wise perturbations are implemented by shuffling the channels in high-dimensional space guided by degradation types, while attention-wise perturbations are achieved through selective masking in the attention space. To achieve these goals, we propose a Degradation-Guided Perturbation Block (DGPB) to implement these two functions, positioned between the encoding and decoding stages of the encoder-decoder architecture. Extensive experimental results demonstrate that DFPIR achieves state-of-the-art performance on several all-in-one image restoration tasks including image denoising, image dehazing, image deraining, motion deblurring, and low-light image enhancement. Our codes are available at https://github.com/TxpHome/DFPIR.

cs.CV↗

arXiv2025

Towards Adaptive Meta-Gradient Adversarial Examples for Visual Tracking

In recent years, visual tracking methods based on convolutional neural networks and Transformers have achieved remarkable performance and have been successfully applied in fields such as autonomous driving. However, the numerous security issues exposed by deep learning models have gradually affected the reliable application of visual tracking methods in real-world scenarios. Therefore, how to reveal the security vulnerabilities of existing visual trackers through effective adversarial attacks has become a critical problem that needs to be addressed. To this end, we propose an adaptive meta-gradient adversarial attack (AMGA) method for visual tracking. This method integrates multi-model ensembles and meta-learning strategies, combining momentum mechanisms and Gaussian smoothing, which can significantly enhance the transferability and attack effectiveness of adversarial examples. AMGA randomly selects models from a large model repository, constructs diverse tracking scenarios, and iteratively performs both white- and black-box adversarial attacks in each scenario, optimizing the gradient directions of each model. This paradigm minimizes the gap between white- and black-box adversarial attacks, thus achieving excellent attack performance in black-box scenarios. Extensive experimental results on large-scale datasets such as OTB2015, LaSOT, and GOT-10k demonstrate that AMGA significantly improves the attack performance, transferability, and deception of adversarial examples. Codes and data are available at https://github.com/pgao-lab/AMGA.

cs.CV↗

arXiv2025

Seed1.5-VL Technical Report

We present Seed1.5-VL, a vision-language foundation model designed to advance general-purpose multimodal understanding and reasoning. Seed1.5-VL is composed with a 532M-parameter vision encoder and a Mixture-of-Experts (MoE) LLM of 20B active parameters. Despite its relatively compact architecture, it delivers strong performance across a wide spectrum of public VLM benchmarks and internal evaluation suites, achieving the state-of-the-art performance on 38 out of 60 public benchmarks. Moreover, in agent-centric tasks such as GUI control and gameplay, Seed1.5-VL outperforms leading multimodal systems, including OpenAI CUA and Claude 3.7. Beyond visual and video understanding, it also demonstrates strong reasoning abilities, making it particularly effective for multimodal reasoning challenges such as visual puzzles. We believe these capabilities will empower broader applications across diverse tasks. In this report, we mainly provide a comprehensive review of our experiences in building Seed1.5-VL across model design, data construction, and training at various stages, hoping that this report can inspire further research. Seed1.5-VL is now accessible at https://www.volcengine.com/ (Volcano Engine Model ID: doubao-1-5-thinking-vision-pro-250428)

cs.CV↗

arXiv2025

AndroidGen: Building an Android Language Agent under Data Scarcity

Large language models have opened up a world of possibilities for various NLP tasks, sparking optimism for the future. Despite their potential, LLMs have yet to be widely used as agents on real mobile devices. The main challenge is the need for high-quality data sources. Time constraints and labor intensity often hinder human annotation. On the other hand, existing LLMs exhibit inadequate completion rates and need a robust data filtration strategy. Given these challenges, we develop a framework called AndroidGen to enhance the capabilities of LLM-based agents under data scarcity. In addition, we leverage AndroidGen to collect trajectories given human tasks and train open-source LLMs on these trajectories to develop an open-source mobile agent without manually labeled trajectories. We extensively evaluate AndroidGen with AndroidWorld, AitW, and various popular applications, demonstrating its improvements and revealing potential areas for future improvement. Code, model, and data are available at https://github.com/THUDM/AndroidGen.

cs.CL↗

arXiv2025

The Fourth Monocular Depth Estimation Challenge

This paper presents the results of the fourth edition of the Monocular Depth Estimation Challenge (MDEC), which focuses on zero-shot generalization to the SYNS-Patches benchmark, a dataset featuring challenging environments in both natural and indoor settings. In this edition, we revised the evaluation protocol to use least-squares alignment with two degrees of freedom to support disparity and affine-invariant predictions. We also revised the baselines and included popular off-the-shelf methods: Depth Anything v2 and Marigold. The challenge received a total of 24 submissions that outperformed the baselines on the test set; 10 of these included a report describing their approach, with most leading methods relying on affine-invariant predictions. The challenge winners improved the 3D F-Score over the previous edition's best result, raising it from 22.58% to 23.05%.

cs.CV↗

arXiv2025

Modeling Multivariate Degradation Data with Dynamic Covariates Under a Bayesian Framework

Degradation data are essential for determining the reliability of high-end products and systems, especially when covering multiple degradation characteristics (DCs). Modern degradation studies not only measure these characteristics but also record dynamic system usage and environmental factors, such as temperature, humidity, and ultraviolet exposures, referred to as the dynamic covariates. Most current research either focuses on a single DC with dynamic covariates or multiple DCs with fixed covariates. This paper presents a Bayesian framework to analyze data with multiple DCs, which incorporates dynamic covariates. We develop a Bayesian framework for mixed effect nonlinear general path models to describe the degradation path and use Bayesian shape-constrained P-splines to model the effects of dynamic covariates. We also detail algorithms for estimating the failure time distribution induced by our degradation model, validate the developed methods through simulation, and illustrate their use in predicting the lifespan of organic coatings in dynamic environments.

stat.ME↗

arXiv2025

Non-Hermitian optical scattering in cold atoms via four-wave mixing

Nonlinear effects could play a crucial role in addressing optical nonreciprocal behaviors in scattering media. Such behaviors are, however, typically observed within a single transmission channel and predominantly in media with fixed optical structures, which inherently restrict the tunability of a nonreciprocal response. We suggest to combine the (intrinsic) nonlinearities of a coherent multi-level medium with a tailored driving geometry that relies on two phase-mismatched standing-wave (SW) beams. This combination is essential for creating extra scattering channels over which, in addition, fully tunable optical nonreciprocal reflection can be attained. Our general approach is here adapted to four-level double-$Λ$ atoms that are found to exhibit distinct forms of nonreciprocal multi-channel scattering and quite sensitive to easily tunable parameters of two SW driving beams. The numerical results we present offer valuable insights into the field of non-Hermitian optical scattering and arise indeed from the interplay of interference among scattering processes and Bragg reflection.

physics.optics↗

arXiv2025

Enhancing Federated Learning Through Secure Cluster-Weighted Client Aggregation

Federated learning (FL) has emerged as a promising paradigm in machine learning, enabling collaborative model training across decentralized devices without the need for raw data sharing. In FL, a global model is trained iteratively on local datasets residing on individual devices, each contributing to the model's improvement. However, the heterogeneous nature of these local datasets, stemming from diverse user behaviours, device capabilities, and data distributions, poses a significant challenge. The inherent heterogeneity in federated learning gives rise to various issues, including model performance discrepancies, convergence challenges, and potential privacy concerns. As the global model progresses through rounds of training, the disparities in local data quality and quantity can impede the overall effectiveness of federated learning systems. Moreover, maintaining fairness and privacy across diverse user groups becomes a paramount concern. To address this issue, this paper introduces a novel FL framework, ClusterGuardFL, that employs dissimilarity scores, k-means clustering, and reconciliation confidence scores to dynamically assign weights to client updates. The dissimilarity scores between global and local models guide the formation of clusters, with cluster size influencing the weight allocation. Within each cluster, a reconciliation confidence score is calculated for individual data points, and a softmax layer generates customized weights for clients. These weights are utilized in the aggregation process, enhancing the model's robustness and privacy. Experimental results demonstrate the efficacy of the proposed approach in achieving improved model performance in diverse datasets.

cs.LG↗

arXiv2025

Overcoming Vocabulary Mismatch: Vocabulary-agnostic Teacher Guided Language Modeling

Using large teacher models to guide the training of smaller student models has become the prevailing paradigm for efficient and effective learning. However, vocabulary mismatches between teacher and student language models pose significant challenges in language modeling, resulting in divergent token sequences and output distributions. To overcome these limitations, we propose Vocabulary-agnostic Teacher Guided Language Modeling (VocAgnoLM), a novel approach that bridges the gap caused by vocabulary mismatch through two key methods: (1) Token-level Lexical Alignment, which aligns token sequences across mismatched vocabularies, and (2) Teacher Guided Loss, which leverages the loss of teacher model to guide effective student training. We demonstrate its effectiveness in language modeling with 1B student model using various 7B teacher models with different vocabularies. Notably, with Qwen2.5-Math-Instruct, a teacher model sharing only about 6% of its vocabulary with TinyLlama, VocAgnoLM achieves a 46% performance improvement compared to naive continual pretraining. Furthermore, we demonstrate that VocAgnoLM consistently benefits from stronger teacher models, providing a robust solution to vocabulary mismatches in language modeling.

cs.CL↗

arXiv2025

A Systematic Literature Review on Task Recommendation Systems for Crowdsourced Software Engineering

Crowdsourced Software Engineering (CSE) offers outsourcing work to software practitioners by leveraging a global online workforce. However, these software practitioners struggle to identify suitable tasks due to the variety of options available. Hence, there have been a growing number of studies on introducing recommendation systems to recommend CSE tasks to software practitioners. The goal of this study is to analyze the existing CSE task recommendation systems, investigating their extracted data, recommendation methods, key advantages and limitations, recommended task types, the use of human factors in recommendations, popular platforms, and features used to make recommendations. This SLR was conducted according to the Kitchenham and Charters' guidelines. We used both manual and automatic search strategies without putting any time limitation for searching the relevant papers. We selected 65 primary studies for data extraction, analysis, and synthesis based on our predefined inclusion and exclusion criteria. From the results of the data analysis, we classified the extracted data into four categories based on the data extraction source, categorized the proposed recommendation systems to fit into a taxonomy, and identified the key advantages and limitations of these systems. Our results revealed that human factors play a major role in CSE task recommendation. Further, we identified five popular task types recommended, popular platforms, and their features used in task recommendation. We also provided recommendations for future research directions. This SLR provides insights into current trends, gaps, and future research directions in CSE task recommendation systems

cs.SE↗

arXiv2025

AIGVE-Tool: AI-Generated Video Evaluation Toolkit with Multifaceted Benchmark

The rapid advancement in AI-generated video synthesis has led to a growth demand for standardized and effective evaluation metrics. Existing metrics lack a unified framework for systematically categorizing methodologies, limiting a holistic understanding of the evaluation landscape. Additionally, fragmented implementations and the absence of standardized interfaces lead to redundant processing overhead. Furthermore, many prior approaches are constrained by dataset-specific dependencies, limiting their applicability across diverse video domains. To address these challenges, we introduce AIGVE-Tool (AI-Generated Video Evaluation Toolkit), a unified framework that provides a structured and extensible evaluation pipeline for a comprehensive AI-generated video evaluation. Organized within a novel five-category taxonomy, AIGVE-Tool integrates multiple evaluation methodologies while allowing flexible customization through a modular configuration system. Additionally, we propose AIGVE-Bench, a large-scale benchmark dataset created with five SOTA video generation models based on hand-crafted instructions and prompts. This dataset systematically evaluates various video generation models across nine critical quality dimensions. Extensive experiments demonstrate the effectiveness of AIGVE-Tool in providing standardized and reliable evaluation results, highlighting specific strengths and limitations of current models and facilitating the advancements of next-generation AI-generated video techniques.

cs.CV↗

arXiv2025

SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models

Instruction-following is a fundamental capability of language models, requiring the model to recognize even the most subtle requirements in the instructions and accurately reflect them in its output. Such an ability is well-suited for and often optimized by preference learning. However, existing methods often directly sample multiple independent responses from the model when creating preference pairs. Such practice can introduce content variations irrelevant to whether the instruction is precisely followed (e.g., different expressions about the same semantic), interfering with the goal of teaching models to recognize the key differences that lead to improved instruction following. In light of this, we introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs free from distractions. By playing against itself, an LLM employs a tree-search strategy to refine its previous responses with respect to the instruction while minimizing unnecessary variations. Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities. Furthermore, SPaR demonstrates promising scalability, greatly enhancing models like GLM-4-9B and LLaMA3-70B. We also identify how inference scaling in tree search would impact model performance. Our code and data are publicly available at https://github.com/thu-coai/SPaR.

cs.CL↗

arXiv2025

Process-based Self-Rewarding Language Models

Large Language Models have demonstrated outstanding performance across various downstream tasks and have been widely applied in multiple scenarios. Human-annotated preference data is used for training to further improve LLMs' performance, which is constrained by the upper limit of human performance. Therefore, Self-Rewarding method has been proposed, where LLMs generate training data by rewarding their own outputs. However, the existing self-rewarding paradigm is not effective in mathematical reasoning scenarios and may even lead to a decline in performance. In this work, we propose the Process-based Self-Rewarding pipeline for language models, which introduces long-thought reasoning, step-wise LLM-as-a-Judge, and step-wise preference optimization within the self-rewarding paradigm. Our new paradigm successfully enhances the performance of LLMs on multiple mathematical reasoning benchmarks through iterative Process-based Self-Rewarding, demonstrating the immense potential of self-rewarding to achieve LLM reasoning that may surpass human capabilities.

cs.CL↗