arXiv subjects

Yu Rong

Publications and source records attributed to Yu Rong.

At least 55 records · Page 3Linked to original sources

arXiv2025

SeaLLMs-Audio: Large Audio-Language Models for Southeast Asia

We introduce SeaLLMs-Audio, the first large audio-language model (LALM) tailored for multiple Southeast Asian (SEA) languages-Indonesian (id), Thai (th), and Vietnamese (vi)-alongside English (en) and Chinese (zh). Trained on a large-scale audio corpus, SeaLLMs-Audio exhibits strong performance across diverse audio-centric tasks, spanning fine-grained audio understanding and voice-based interaction. Its key features include: 1) Multilingual: the model primarily supports 5 languages, namely Indonesian, Thai, Vietnamese, English, and Chinese; 2) Multimodal: the model accepts flexible input modalities, including audio only, text only, as well as audio with text; 3) Multi-task: the model supports a wide range of tasks, including audio analysis tasks such as Audio Captioning, Automatic Speech Recognition, Speech-to-Text Translation, Speech Emotion Recognition, Speech Question Answering, and Speech Summarization. It also enables voice-based dialogue, including answering factual, mathematical, and general knowledge queries. As a significant step towards advancing audio LLMs in Southeast Asia, we expect SeaLLMs-Audio to benefit both the regional research community and industry. To automate LALM evaluation for Southeast Asia, we introduce SeaBench-Audio, a benchmark spanning multiple tasks. Experiments show that SeaLLMs-Audio achieves competitive performance compared with other LALMs on SEA languages.

cs.CL↗

arXiv2025

Universally Invariant Learning in Equivariant GNNs

Equivariant Graph Neural Networks (GNNs) have demonstrated significant success across various applications. To achieve completeness -- that is, the universal approximation property over the space of equivariant functions -- the network must effectively capture the intricate multi-body interactions among different nodes. Prior methods attain this via deeper architectures, augmented body orders, or increased degrees of steerable features, often at high computational cost and without polynomial-time solutions. In this work, we present a theoretically grounded framework for constructing complete equivariant GNNs that is both efficient and practical. We prove that a complete equivariant GNN can be achieved through two key components: 1) a complete scalar function, referred to as the canonical form of the geometric graph; and 2) a full-rank steerable basis set. Leveraging this finding, we propose an efficient algorithm for constructing complete equivariant GNNs based on two common models: EGNN and TFN. Empirical results demonstrate that our model demonstrates superior completeness and excellent performance with only a few layers, thereby significantly reducing computational overhead while maintaining strong practical efficacy.

cs.LG↗

arXiv2025

Scaling Language-Centric Omnimodal Representation Learning

Recent multimodal embedding approaches leveraging multimodal large language models (MLLMs) fine-tuned with contrastive learning (CL) have shown promising results, yet the underlying reasons behind their superiority remain underexplored. This work argues that a crucial advantage of MLLM-based approaches stems from implicit cross-modal alignment achieved during generative pretraining, where the language decoder learns to exploit multimodal signals within a shared representation space for generating unimodal outputs. Through analysis of anisotropy and kernel similarity structure, we empirically confirm that latent alignment emerges within MLLM representations, allowing CL to serve as a lightweight refinement stage. Leveraging this insight, we propose a Language-Centric Omnimodal Embedding framework, termed LCO-Emb. Extensive experiments across diverse backbones and benchmarks demonstrate its effectiveness, achieving state-of-the-art performance across modalities. Furthermore, we identify a Generation-Representation Scaling Law (GRSL), showing that the representational capabilities gained through contrastive refinement scales positively with the MLLM's generative capabilities. This suggests that improving generative abilities evolves as an effective paradigm for enhancing representation quality. We provide a theoretical explanation of GRSL, which formally links the MLLM's generative quality to the upper bound on its representation performance, and validate it on a challenging, low-resource visual-document retrieval task, showing that continual generative pretraining before CL can further enhance the potential of a model's embedding capabilities. Codes, models, and resources are available at https://github.com/LCO-Embedding/LCO-Embedding.

cs.CL↗

arXiv2025

ReasonMed: A 370K Multi-Agent Generated Dataset for Advancing Medical Reasoning

Reasoning-based large language models have excelled in mathematics and programming, yet their potential in knowledge-intensive medical question answering remains underexplored and insufficiently validated in clinical contexts. To bridge this gap, we introduce ReasonMed, the largest medical reasoning dataset to date, comprising 370k high-quality examples distilled from 1.75 million initial reasoning paths generated by complementary LLMs and curated through a cost-efficient easy-medium-difficult (EMD) pipeline. ReasonMed is built through a multi-agent generation, verification, and refinement process, in which an Error Refiner improves reasoning paths by correcting error-prone steps identified by a verifier. Using ReasonMed, we investigate effective strategies for training medical reasoning models and find that integrating detailed CoT reasoning with concise answer summaries yields the most robust fine-tuning results. Models trained on ReasonMed set a new benchmark: ReasonMed-7B surpasses the prior best sub-10B models by 4.17% and even exceeds LLaMA3.1-70B on PubMedQA by 4.60%. When scaled to ReasonMed-14B, it remains highly competitive, underscoring consistent scaling potential. The codes and datasets are available at https://github.com/YuSun-Work/ReasonMed.

cs.CL↗

arXiv2025

Galaxy Group Spin Alignment with Cosmic Filament in the TNG Simulation

We investigate the alignment between the spin vectors of galaxy groups and the axes of their nearest cosmic filaments using the TNG300-1 cosmological hydrodynamical simulation. By systematically analyzing a large sample of groups, we find a robust perpendicular alignment between group spin and filament orientation. Among all examined properties, only group mass and the distance to the nearest filament significantly affect the strength of this alignment: more massive groups and those closer to filaments exhibit a stronger perpendicular signal. In contrast, the alignment is largely insensitive to group richness, the stellar mass threshold used to select member galaxies, and redshift. We further quantify the bias introduced by using member galaxies as tracers of group spin, finding a typical misalignment angle of $\sim38^\circ$ between the spin measured from all dark matter particles and that inferred from member galaxies, independent of group richness or stellar mass cut. Our results provide a clear theoretical benchmark for interpreting observational measurements of spin-filament alignment and highlight the importance of considering group mass and environment. These findings help clarify the main factors influencing spin-filament alignment and provide useful context for future observational and theoretical studies of angular momentum in the cosmic web.

astro-ph.CO↗

arXiv2025

MMR1: Enhancing Multimodal Reasoning with Variance-Aware Sampling and Open Resources

Large multimodal reasoning models have achieved rapid progress, but their advancement is constrained by two major limitations: the absence of open, large-scale, high-quality long chain-of-thought (CoT) data, and the instability of reinforcement learning (RL) algorithms in post-training. Group Relative Policy Optimization (GRPO), the standard framework for RL fine-tuning, is prone to gradient vanishing when reward variance is low, which weakens optimization signals and impairs convergence. This work makes three contributions: (1) We propose Variance-Aware Sampling (VAS), a data selection strategy guided by Variance Promotion Score (VPS) that combines outcome variance and trajectory diversity to promote reward variance and stabilize policy optimization. (2) We release large-scale, carefully curated resources containing ~1.6M long CoT cold-start data and ~15k RL QA pairs, designed to ensure quality, difficulty, and diversity, along with a fully reproducible end-to-end training codebase. (3) We open-source a family of multimodal reasoning models in multiple scales, establishing standardized baselines for the community. Experiments across mathematical reasoning benchmarks demonstrate the effectiveness of both the curated data and the proposed VAS. Comprehensive ablation studies and analyses provide further insight into the contributions of each component. In addition, we theoretically establish that reward variance lower-bounds the expected policy gradient magnitude, with VAS serving as a practical mechanism to realize this guarantee. Our code, data, and checkpoints are available at https://github.com/LengSicong/MMR1.

cs.CV↗

arXiv2025

GeoPQA: Bridging the Visual Perception Gap in MLLMs for Geometric Reasoning

Recent advancements in reinforcement learning (RL) have enhanced the reasoning abilities of large language models (LLMs), yet the impact on multimodal LLMs (MLLMs) is limited. Particularly in vision-intensive tasks like geometric reasoning, MLLMs hallucinate frequently, leading to inaccurate reasoning. We attribute this to the perceptual bottleneck in MLLMs, which caps the benefits of reasoning training. To quantify this, we design a Geo-Perception Question-Answering (GeoPQA) benchmark, targeting basic geometric concepts and spatial relationships. Experiments on GeoPQA reveal significant shortcomings of MLLMs in visual perception, which constrain RL reward signals for effective training. To address this bottleneck, we propose a two-stage RL training framework by first enhancing the visual perception of geometric structures, then fostering reasoning capabilities. Applied to Qwen2.5-VL-3B-Instruct, our two-stage training improves geometric reasoning by 9.7% and geometric problem solving by 9.1%, compared to the direct reasoning training approach. Our method also generalizes to other vision-intensive domains like figure understanding, highlighting the importance of perceptual grounding in effective MLLM reasoning.

cs.CL↗

arXiv2025

Distributed Coherent Beamforming at 60 GHz Enabled by Optically-Established Coherence

We implement and experimentally demonstrate a 60 GHz distributed system leveraging an optical time synchronization system that provides precise time and frequency alignment between independent elements of the distributed mesh. Utilizing such accurate coherence, we perform receive beamforming with interference rejection and transmit nulling. In these configurations, the system achieves a coherent gain over an incoherent network of N nodes, significantly improving the relevant signal power ratios. Our system demonstrates extended array phase coherence times, enabling advanced techniques. Results from over-the-air experiments demonstrate a 14.3 dB signal-to-interference-plus-noise improvement in interference-laden scenarios with a contributing 13.5 dB null towards interference in receive beamforming. In transmit nulling, a signal-to-noise ratio (SNR) gain of 7.9 dB is measured towards an intended receiver while maintaining an SNR reduction of 8.9 dB at another receiver. These findings represent the use of distributed coherence in the V band without the use of GPS timing.

eess.SP↗

arXiv2025

AMAZe: A Multi-Agent Zero-shot Index Advisor for Relational Databases

Index recommendation is one of the most important problems in database management system (DBMS) optimization. Given queries and certain index-related constraints, traditional methods rely on heuristic optimization or learning-based models to select effective indexes and improve query performance. However, heuristic optimization suffers from high computation time, and learning-based models lose generalisability due to training for different workloads and database schemas. With the recent rapid development of large language models (LLMs), methods using prompt tuning have been proposed to enhance the efficiency of index selection. However, such methods still can not achieve the state-of-the-art (SOTA) results, and preparing the index selection demonstrations is also resource-intensive. To address these issues, we propose AMAZe, a zero-shot LLM-based index advisor with a multi-agent framework. We decompose the index recommendation problem into sub-steps, including planning, selection, combination, revision, and reflection. A set of LLM-embedded agents is designed to handle each one of the different sub-steps. Our method utilizes high-level agents to control the index selection process and low-level agents to select and revise indexes. Through extensive experiments, we show that our proposed AMAZe not only achieves the SOTA performance compared to the heuristic methods, but also outperforms learning-based and prompt-based methods with higher efficiency and better zero-shot inference ability.

cs.DB↗

arXiv2025

From Post To Personality: Harnessing LLMs for MBTI Prediction in Social Media

Personality prediction from social media posts is a critical task that implies diverse applications in psychology and sociology. The Myers Briggs Type Indicator (MBTI), a popular personality inventory, has been traditionally predicted by machine learning (ML) and deep learning (DL) techniques. Recently, the success of Large Language Models (LLMs) has revealed their huge potential in understanding and inferring personality traits from social media content. However, directly exploiting LLMs for MBTI prediction faces two key challenges: the hallucination problem inherent in LLMs and the naturally imbalanced distribution of MBTI types in the population. In this paper, we propose PostToPersonality (PtoP), a novel LLM based framework for MBTI prediction from social media posts of individuals. Specifically, PtoP leverages Retrieval Augmented Generation with in context learning to mitigate hallucination in LLMs. Furthermore, we fine tune a pretrained LLM to improve model specification in MBTI understanding with synthetic minority oversampling, which balances the class imbalance by generating synthetic samples. Experiments conducted on a real world social media dataset demonstrate that PtoP achieves state of the art performance compared with 10 ML and DL baselines.

cs.CL↗

arXiv2025

VL-Cogito: Progressive Curriculum Reinforcement Learning for Advanced Multimodal Reasoning

Reinforcement learning has proven its effectiveness in enhancing the reasoning capabilities of large language models. Recent research efforts have progressively extended this paradigm to multimodal reasoning tasks. Due to the inherent complexity and diversity of multimodal tasks, especially in semantic content and problem formulations, existing models often exhibit unstable performance across various domains and difficulty levels. To address these limitations, we propose VL-Cogito, an advanced multimodal reasoning model trained via a novel multi-stage Progressive Curriculum Reinforcement Learning (PCuRL) framework. PCuRL systematically guides the model through tasks of gradually increasing difficulty, substantially improving its reasoning abilities across diverse multimodal contexts. The framework introduces two key innovations: (1) an online difficulty soft weighting mechanism, dynamically adjusting training difficulty across successive RL training stages; and (2) a dynamic length reward mechanism, which encourages the model to adaptively regulate its reasoning path length according to task complexity, thus balancing reasoning efficiency with correctness. Experimental evaluations demonstrate that VL-Cogito consistently matches or surpasses existing reasoning-oriented models across mainstream multimodal benchmarks spanning mathematics, science, logic, and general understanding, validating the effectiveness of our approach.

cs.CV↗

arXiv2025

Hierarchical Graph Information Bottleneck for Multi-Behavior Recommendation

In real-world recommendation scenarios, users typically engage with platforms through multiple types of behavioral interactions. Multi-behavior recommendation algorithms aim to leverage various auxiliary user behaviors to enhance prediction for target behaviors of primary interest (e.g., buy), thereby overcoming performance limitations caused by data sparsity in target behavior records. Current state-of-the-art approaches typically employ hierarchical design following either cascading (e.g., view$\rightarrow$cart$\rightarrow$buy) or parallel (unified$\rightarrow$behavior$\rightarrow$specific components) paradigms, to capture behavioral relationships. However, these methods still face two critical challenges: (1) severe distribution disparities across behaviors, and (2) negative transfer effects caused by noise in auxiliary behaviors. In this paper, we propose a novel model-agnostic Hierarchical Graph Information Bottleneck (HGIB) framework for multi-behavior recommendation to effectively address these challenges. Following information bottleneck principles, our framework optimizes the learning of compact yet sufficient representations that preserve essential information for target behavior prediction while eliminating task-irrelevant redundancies. To further mitigate interaction noise, we introduce a Graph Refinement Encoder (GRE) that dynamically prunes redundant edges through learnable edge dropout mechanisms. We conduct comprehensive experiments on three real-world public datasets, which demonstrate the superior effectiveness of our framework. Beyond these widely used datasets in the academic community, we further expand our evaluation on several real industrial scenarios and conduct an online A/B testing, showing again a significant improvement in multi-behavior recommendations. The source code of our proposed HGIB is available at https://github.com/zhy99426/HGIB.

cs.IR↗

arXiv2025

Lack of Bulge Alignment in Late-type Galaxies with Large-scale Filaments Plausibly Unveils a Radial Migration Formation Scenario

The formation sequence of bulges and disks in late-type galaxies (LTGs) remains debated. While some theories suggest bulges form before disks, others propose the reverse. To address this, we analyze a bulge+disk decomposition catalog from the Sloan Digital Sky Survey, examining the alignment between central bulge major axes and large-scale filaments. For LTGs near filament spines, no significant alignment is found. However, LTGs farther from filaments show a marginal ($\sim 1σ$) perpendicular alignment. Notably, central bulges and outer disks exhibit strong alignment in the sky plane, suggesting bulge formation may be driven by material migration from outer disks. Although further simulations are needed, our results provide a novel alignment-based perspective on bulge formation, advancing understanding of galactic structural evolution.

astro-ph.GA↗

arXiv2025

STAR-R1: Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across diverse tasks, yet they lag significantly behind humans in spatial reasoning. We investigate this gap through Transformation-Driven Visual Reasoning (TVR), a challenging task requiring identification of object transformations across images under varying viewpoints. While traditional Supervised Fine-Tuning (SFT) fails to generate coherent reasoning paths in cross-view settings, sparse-reward Reinforcement Learning (RL) suffers from inefficient exploration and slow convergence. To address these limitations, we propose STAR-R1, a novel framework that integrates a single-stage RL paradigm with a fine-grained reward mechanism tailored for TVR. Specifically, STAR-R1 rewards partial correctness while penalizing excessive enumeration and passive inaction, enabling efficient exploration and precise reasoning. Comprehensive evaluations demonstrate that STAR-R1 achieves state-of-the-art performance across all 11 metrics, outperforming SFT by 23% in cross-view scenarios. Further analysis reveals STAR-R1's anthropomorphic behavior and highlights its unique ability to compare all objects for improving spatial reasoning. Our work provides critical insights in advancing the research of MLLMs and reasoning models. The codes, model weights, and data will be publicly available at https://github.com/zongzhao23/STAR-R1.

cs.CV↗

arXiv2025

Adaptive Graph Integration for Cross-Domain Recommendation via Heterogeneous Graph Coordinators

In the digital era, users typically interact with diverse items across multiple domains (e.g., e-commerce, streaming platforms, and social networks), generating intricate heterogeneous interaction graphs. Leveraging multi-domain data can improve recommendation systems by enriching user insights and mitigating data sparsity in individual domains. However, integrating such multi-domain knowledge for cross-domain recommendation remains challenging due to inherent disparities in user behavior and item characteristics and the risk of negative transfer, where irrelevant or conflicting information from the source domains adversely impacts the target domain's performance. To tackle these challenges, we propose HAGO, a novel framework with \textbf{H}eterogeneous \textbf{A}daptive \textbf{G}raph co\textbf{O}rdinators, which dynamically integrates multi-domain graphs into a cohesive structure. HAGO adaptively adjusts the connections between coordinators and multi-domain graph nodes to enhance beneficial inter-domain interactions while alleviating negative transfer. Furthermore, we introduce a universal multi-domain graph pre-training strategy alongside HAGO to collaboratively learn high-quality node representations across domains. Being compatible with various graph-based models and pre-training techniques, HAGO demonstrates broad applicability and effectiveness. Extensive experiments show that our framework outperforms state-of-the-art methods in cross-domain recommendation scenarios, underscoring its potential for real-world applications. The source code is available at https://github.com/zhy99426/HAGO.

cs.IR↗

arXiv2025

Strong Correlation between Galactic HI-to-stellar Mass Ratio And Halo Spin Explored by HI-rich Galaxies

Using a semi-analytic approach, we estimate halo spins for a large sample of HI-rich galaxies from the Arecibo Legacy Fast Alfa Survey and examine the correlation between HI mass fractions and halo spins. Our analysis reveals a strong correlation between halo spin and the HI-to-stellar mass ratio in both low-mass and massive galaxy samples. This finding suggests a universal formation scenario: higher halo spin reduces angular momentum loss and gas condensation, leading to lower star formation rates and weaker feedback, which in turn helps retain gas within dark matter halos.

astro-ph.GA↗

arXiv2025

Analyzing LLMs' Knowledge Boundary Cognition Across Languages Through the Lens of Internal Representations

While understanding the knowledge boundaries of LLMs is crucial to prevent hallucination, research on the knowledge boundaries of LLMs has predominantly focused on English. In this work, we present the first study to analyze how LLMs recognize knowledge boundaries across different languages by probing their internal representations when processing known and unknown questions in multiple languages. Our empirical studies reveal three key findings: 1) LLMs' perceptions of knowledge boundaries are encoded in the middle to middle-upper layers across different languages. 2) Language differences in knowledge boundary perception follow a linear structure, which motivates our proposal of a training-free alignment method that effectively transfers knowledge boundary perception ability across languages, thereby helping reduce hallucination risk in low-resource languages; 3) Fine-tuning on bilingual question pair translation further enhances LLMs' recognition of knowledge boundaries across languages. Given the absence of standard testbeds for cross-lingual knowledge boundary analysis, we construct a multilingual evaluation suite comprising three representative types of knowledge boundary data. Our code and datasets are publicly available at https://github.com/DAMO-NLP-SG/LLM-Multilingual-Knowledge-Boundaries.

cs.CL↗

arXiv2025

The Internal Kinematics, Stellar Population, and Gas-phase Properties of The Pseudobulge in An Ultra-diffuse Galaxy: AGC721966

Leveraging spectroscopic data from the Sloan Digital Sky Survey, we conduct a comprehensive analysis of the central stellar velocity dispersion, stellar population properties, star formation history, and gas-phase chemical abundances in AGC721966, a unique ultra-diffuse galaxy (UDG) harboring a pseudobulge. Our findings reveal that the pseudobulge formed in the early universe but underwent a recent episode of rejuvenated star formation. The system exhibits a mass-weighted (light-weighted) stellar population age of $τ_{\star}\sim 7.4\pm2.5$ ($2.9\pm1.5$)~Gyr, a stellar metallicity of [M/H]$\sim -0.62\pm0.26$ ($-0.55\pm0.20$), an $α$-element enhancement of [$α$/Fe]$\sim 0.36\pm0.09$ ($0.37\pm0.07$), and a gas-phase oxygen abundance of \Oabund$\sim 8.15\pm0.03$. The central stellar velocity dispersion is measured as $σ_{\rm c}\sim 57.9\pm15.7$~km/s. These results provide robust evidence supporting the early halo-halo merging formation scenario proposed by \cite{Rong25}, while unequivocally ruling out the ``failed'' $L^{\star}$ formation model, at least for AGC721966. Furthermore, through systematic application of the baryonic Tully-Fisher relation, we establish that these pseudobulge-hosting UDGs are neither misidentified nuclear star cluster-bearing dwarf galaxies nor bulge-dominated massive galaxies, thereby affirming their distinct evolutionary pathway.

astro-ph.GA↗