arXiv subjects

Yue Deng

Publications and source records attributed to Yue Deng.

At least 55 records · Page 3Linked to original sources

arXiv2025

You Only Look One Step: Accelerating Backpropagation in Diffusion Sampling with Gradient Shortcuts

Diffusion models (DMs) have recently demonstrated remarkable success in modeling large-scale data distributions. However, many downstream tasks require guiding the generated content based on specific differentiable metrics, typically necessitating backpropagation during the generation process. This approach is computationally expensive, as generating with DMs often demands tens to hundreds of recursive network calls, resulting in high memory usage and significant time consumption. In this paper, we propose a more efficient alternative that approaches the problem from the perspective of parallel denoising. We show that full backpropagation throughout the entire generation process is unnecessary. The downstream metrics can be optimized by retaining the computational graph of only one step during generation, thus providing a shortcut for gradient propagation. The resulting method, which we call Shortcut Diffusion Optimization (SDO), is generic, high-performance, and computationally lightweight, capable of optimizing all parameter types in diffusion sampling. We demonstrate the effectiveness of SDO on several real-world tasks, including controlling generation by optimizing latent and aligning the DMs by fine-tuning network parameters. Compared to full backpropagation, our approach reduces computational costs by $\sim 90\%$ while maintaining superior performance. Code is available at https://github.com/deng-ai-lab/SDO.

cs.LG↗

arXiv2025

Image Restoration via Multi-domain Learning

Due to adverse atmospheric and imaging conditions, natural images suffer from various degradation phenomena. Consequently, image restoration has emerged as a key solution and garnered substantial attention. Although recent Transformer architectures have demonstrated impressive success across various restoration tasks, their considerable model complexity poses significant challenges for both training and real-time deployment. Furthermore, instead of investigating the commonalities among different degradations, most existing restoration methods focus on modifying Transformer under limited restoration priors. In this work, we first review various degradation phenomena under multi-domain perspective, identifying common priors. Then, we introduce a novel restoration framework, which integrates multi-domain learning into Transformer. Specifically, in Token Mixer, we propose a Spatial-Wavelet-Fourier multi-domain structure that facilitates local-region-global multi-receptive field modeling to replace vanilla self-attention. Additionally, in Feed-Forward Network, we incorporate multi-scale learning to fuse multi-domain features at different resolutions. Comprehensive experimental results across ten restoration tasks, such as dehazing, desnowing, motion deblurring, defocus deblurring, rain streak/raindrop removal, cloud removal, shadow removal, underwater enhancement and low-light enhancement, demonstrate that our proposed model outperforms state-of-the-art methods and achieves a favorable trade-off among restoration performance, parameter size, computational cost and inference latency. The code is available at: https://github.com/deng-ai-lab/SWFormer.

eess.IV↗

arXiv2025

Aquatic-GS: A Hybrid 3D Representation for Underwater Scenes

Representing underwater 3D scenes is a valuable yet complex task, as attenuation and scattering effects during underwater imaging significantly couple the information of the objects and the water. This coupling presents a significant challenge for existing methods in effectively representing both the objects and the water medium simultaneously. To address this challenge, we propose Aquatic-GS, a hybrid 3D representation approach for underwater scenes that effectively represents both the objects and the water medium. Specifically, we construct a Neural Water Field (NWF) to implicitly model the water parameters, while extending the latest 3D Gaussian Splatting (3DGS) to model the objects explicitly. Both components are integrated through a physics-based underwater image formation model to represent complex underwater scenes. Moreover, to construct more precise scene geometry and details, we design a Depth-Guided Optimization (DGO) mechanism that uses a pseudo-depth map as auxiliary guidance. After optimization, Aquatic-GS enables the rendering of novel underwater viewpoints and supports restoring the true appearance of underwater scenes, as if the water medium were absent. Extensive experiments on both simulated and real-world datasets demonstrate that Aquatic-GS surpasses state-of-the-art underwater 3D representation methods, achieving better rendering quality and real-time rendering performance with a 410x increase in speed. Furthermore, regarding underwater image restoration, Aquatic-GS outperforms representative dewatering methods in color correction, detail recovery, and stability. Our models, code, and datasets can be accessed at https://aquaticgs.github.io.

cs.CV↗

arXiv2025

SMAC-R1: The Emergence of Intelligence in Decision-Making Tasks

StarCraft Multi-Agent Challenge (SMAC) has been one of the most commonly used experimental environments in multi-agent reinforcement learning (MARL), where the specific task is to control a set number of allied units to defeat enemy forces. Traditional MARL algorithms often require interacting with the environment for millions of steps to train a parametric model, of which the resulting policies are typically non-interpretable with weak transferability. In this paper, we introduce SMAC-R1 which is based on the Qwen2.5-7B-Base LLM distilled from DeepSeek-Coder-v2.5-236B. Similar to online reinforcement learning after behavior cloning in offline learning process, in our pipeline, agents leverage the DeepSeek LLM to generate decision tree code by providing task descriptions, and the agents are further self-reflected using feedback from the rewards provided by the environment. Based on that, we augment the generated scripts to fine-tune a small LLM, Qwen2.5-7B-Base, to distill the decision-making ability via Supervised Fine-Tuning (SFT) and enhance the script generation ability by the Group Relative Policy Optimization (GRPO) algorithm. We conduct experiments in the original 23 SMAC tasks and 10 newly-designed tasks to demonstrate that our method can produce high-quality, interpretable decision trees with minimal environmental exploration. Moreover, these scripts exhibit strong transferability, successfully applying to homogeneous SMAC environments without modification. We believe this approach offers a new direction for solving decision-making tasks and domain-specific LLM training pipelines in the future.

cs.AI↗

arXiv2025

Babel: Open Multilingual Large Language Models Serving Over 90% of Global Speakers

Large language models (LLMs) have revolutionized natural language processing (NLP), yet open-source multilingual LLMs remain scarce, with existing models often limited in language coverage. Such models typically prioritize well-resourced languages, while widely spoken but under-resourced languages are often overlooked. To address this disparity, we introduce $\texttt{Babel}$, an open multilingual LLM that covers the top 25 languages by number of speakers, supports over 90% of the global population, and includes many languages neglected by other open multilingual LLMs. Unlike traditional continue pretraining approaches, Babel expands its parameter count through a layer extension technique that elevates Babel's performance ceiling. We introduce two variants: $\texttt{Babel-9B}$, designed for efficient inference and fine-tuning, and $\texttt{Babel-83B}$, which sets a new standard for open multilingual LLMs. Extensive evaluations on multilingual tasks demonstrate its superior performance compared to open LLMs of comparable size. In addition, using open-source supervised fine-tuning datasets, Babel achieves remarkable performance, with Babel-9B-Chat leading among 10B-sized LLMs and Babel-83B-Chat setting a new standard for multilingual tasks, reaching the same level of commercial models.

cs.CL↗

arXiv2025

"Auntie, Please Don't Fall for Those Smooth Talkers": How Chinese Younger Family Members Safeguard Seniors from Online Fraud

Online fraud substantially harms individuals and seniors are disproportionately targeted. While family is crucial for seniors, little research has empirically examined how they protect seniors against fraud. To address this gap, we employed an inductive thematic analysis of 124 posts and 16,872 comments on RedNote (Xiaohongshu), exploring the family support ecosystem for senior-targeted online fraud in China. We develop a taxonomy of senior-targeted online fraud from a familial perspective, revealing younger members often spot frauds hard for seniors to detect, such as unusual charges. Younger family members fulfill multiple safeguarding roles, including preventative measures, fraud identification, fraud persuasion, loss recovery, and education. They also encounter numerous challenges, such as seniors' refusal of help and considerable mental and financial stress. Drawing on these, we develop a conceptual framework to characterize family support in senior-targeted fraud, and outline implications for researchers and practitioners to consider the broader stakeholder ecosystem and cultural aspects.

cs.HC↗

arXiv2024

SMAC-Hard: Enabling Mixed Opponent Strategy Script and Self-play on SMAC

The availability of challenging simulation environments is pivotal for advancing the field of Multi-Agent Reinforcement Learning (MARL). In cooperative MARL settings, the StarCraft Multi-Agent Challenge (SMAC) has gained prominence as a benchmark for algorithms following centralized training with decentralized execution paradigm. However, with continual advancements in SMAC, many algorithms now exhibit near-optimal performance, complicating the evaluation of their true effectiveness. To alleviate this problem, in this work, we highlight a critical issue: the default opponent policy in these environments lacks sufficient diversity, leading MARL algorithms to overfit and exploit unintended vulnerabilities rather than learning robust strategies. To overcome these limitations, we propose SMAC-HARD, a novel benchmark designed to enhance training robustness and evaluation comprehensiveness. SMAC-HARD supports customizable opponent strategies, randomization of adversarial policies, and interfaces for MARL self-play, enabling agents to generalize to varying opponent behaviors and improve model stability. Furthermore, we introduce a black-box testing framework wherein agents are trained without exposure to the edited opponent scripts but are tested against these scripts to evaluate the policy coverage and adaptability of MARL algorithms. We conduct extensive evaluations of widely used and state-of-the-art algorithms on SMAC-HARD, revealing the substantial challenges posed by edited and mixed strategy opponents. Additionally, the black-box strategy tests illustrate the difficulty of transferring learned policies to unseen adversaries. We envision SMAC-HARD as a critical step toward benchmarking the next generation of MARL algorithms, fostering progress in self-play methods for multi-agent systems. Our code is available at https://github.com/devindeng94/smac-hard.

cs.AI↗

arXiv2024

RL-LLM-DT: An Automatic Decision Tree Generation Method Based on RL Evaluation and LLM Enhancement

Traditionally, AI development for two-player zero-sum games has relied on two primary techniques: decision trees and reinforcement learning (RL). A common approach involves using a fixed decision tree as one player's strategy while training an RL agent as the opponent to identify vulnerabilities in the decision tree, thereby improving its strategic strength iteratively. However, this process often requires significant human intervention to refine the decision tree after identifying its weaknesses, resulting in inefficiencies and hindering full automation of the strategy enhancement process. Fortunately, the advent of Large Language Models (LLMs) offers a transformative opportunity to automate the process. We propose RL-LLM-DT, an automatic decision tree generation method based on RL Evaluation and LLM Enhancement. Given an initial decision tree, the method involves two important iterative steps. Response Policy Search: RL is used to discover counter-strategies targeting the decision tree. Policy Improvement: LLMs analyze failure scenarios and generate improved decision tree code. In our method, RL focuses on finding the decision tree's flaws while LLM is prompted to generate an improved version of the decision tree. The iterative refinement process terminates when RL can't find any flaw of the tree or LLM fails to improve the tree. To evaluate the effectiveness of this integrated approach, we conducted experiments in a curling game. After iterative refinements, our curling AI based on the decision tree ranks first on the Jidi platform among 34 curling AIs in total, which demonstrates that LLMs can significantly enhance the robustness and adaptability of decision trees, representing a substantial advancement in the field of Game AI. Our code is available at https://github.com/Linjunjie99/RL-LLM-DT.

cs.AI↗

arXiv2024

Jamming is a first-order transition with quenched disorder in amorphous materials sheared by cyclic quasistatic deformations

Jamming is an athermal transition between flowing and rigid states in amorphous systems such as granular matter, colloidal suspensions, complex fluids and cells. The jamming transition seems to display mixed aspects of a first-order transition, evidenced by a discontinuity in the coordination number, and a second-order transition, indicated by power-law scalings and diverging lengths. Here we demonstrate that jamming is a first-order transition with quenched disorder in cyclically sheared systems with quasistatic deformations, in two and three dimensions. Based on scaling analyses, we show that fluctuations of the jamming density in finite-sized systems have important consequences on the finite-size effects of various quantities, resulting in a square relationship between disconnected and connected susceptibilities, a key signature of the first-order transition with quenched disorder. This study puts the jamming transition into the category of a broad class of transitions in disordered systems where sample-to-sample fluctuations dominate over thermal fluctuations, suggesting that the nature and behavior of the jamming transition might be better understood within the developed theoretical framework of the athermally driven random-field Ising model.

cond-mat.soft↗

arXiv2024

Physics-aligned Schrödinger bridge

The reconstruction of physical fields from sparse measurements is pivotal in both scientific research and engineering applications. Traditional methods are increasingly supplemented by deep learning models due to their efficacy in extracting features from data. However, except for the low accuracy on complex physical systems, these models often fail to comply with essential physical constraints, such as governing equations and boundary conditions. To overcome this limitation, we introduce a novel data-driven field reconstruction framework, termed the Physics-aligned Schrödinger Bridge (PalSB). This framework leverages a diffusion Schrödinger bridge mechanism that is specifically tailored to align with physical constraints. The PalSB approach incorporates a dual-stage training process designed to address both local reconstruction mapping and global physical principles. Additionally, a boundary-aware sampling technique is implemented to ensure adherence to physical boundary conditions. We demonstrate the effectiveness of PalSB through its application to three complex nonlinear systems: cylinder flow from Particle Image Velocimetry experiments, two-dimensional turbulence, and a reaction-diffusion system. The results reveal that PalSB not only achieves higher accuracy but also exhibits enhanced compliance with physical constraints compared to existing methods. This highlights PalSB's capability to generate high-quality representations of intricate physical interactions, showcasing its potential for advancing field reconstruction techniques.

physics.flu-dyn↗

arXiv2024

SeaLLMs 3: Open Foundation and Chat Multilingual Large Language Models for Southeast Asian Languages

Large Language Models (LLMs) have shown remarkable abilities across various tasks, yet their development has predominantly centered on high-resource languages like English and Chinese, leaving low-resource languages underserved. To address this disparity, we present SeaLLMs 3, the latest iteration of the SeaLLMs model family, tailored for Southeast Asian languages. This region, characterized by its rich linguistic diversity, has lacked adequate language technology support. SeaLLMs 3 aims to bridge this gap by covering a comprehensive range of languages spoken in this region, including English, Chinese, Indonesian, Vietnamese, Thai, Tagalog, Malay, Burmese, Khmer, Lao, Tamil, and Javanese. Leveraging efficient language enhancement techniques and a specially constructed instruction tuning dataset, SeaLLMs 3 significantly reduces training costs while maintaining high performance and versatility. Our model excels in tasks such as world knowledge, mathematical reasoning, translation, and instruction following, achieving state-of-the-art performance among similarly sized models. Additionally, we prioritized safety and reliability by addressing both general and culture-specific considerations and incorporated mechanisms to reduce hallucinations. This work underscores the importance of inclusive AI, showing that advanced LLM capabilities can benefit underserved linguistic and cultural communities.

cs.CL↗

arXiv2024

SeaLLMs -- Large Language Models for Southeast Asia

Despite the remarkable achievements of large language models (LLMs) in various tasks, there remains a linguistic bias that favors high-resource languages, such as English, often at the expense of low-resource and regional languages. To address this imbalance, we introduce SeaLLMs, an innovative series of language models that specifically focuses on Southeast Asian (SEA) languages. SeaLLMs are built upon the Llama-2 model and further advanced through continued pre-training with an extended vocabulary, specialized instruction and alignment tuning to better capture the intricacies of regional languages. This allows them to respect and reflect local cultural norms, customs, stylistic preferences, and legal considerations. Our comprehensive evaluation demonstrates that SeaLLM-13b models exhibit superior performance across a wide spectrum of linguistic tasks and assistant-style instruction-following capabilities relative to comparable open-source models. Moreover, they outperform ChatGPT-3.5 in non-Latin languages, such as Thai, Khmer, Lao, and Burmese, by large margins while remaining lightweight and cost-effective to operate.

cs.CL↗

arXiv2024

Modelling infectious disease transmission dynamics in conference environments: An individual-based approach

The global public health landscape is perpetually challenged by the looming threat of infectious diseases. Central to addressing this concern is the imperative to prevent and manage disease transmission during pandemics, particularly in unique settings. This study addresses the transmission dynamics of infectious diseases within conference venues, presenting a computational model designed to simulate transmission processes within a condensed timeframe (one day), beginning with sporadic cases. Our model intricately captures the activities of individual attendees within the conference venue, encompassing meetings, rest intervals, and meal breaks. While meetings entail proximity seating, rest and lunch periods allow attendees to interact with diverse individuals. Moreover, the restroom environment poses an additional avenue for potential infection transmission. Employing an individual-based model, we meticulously replicated the transmission dynamics of infectious diseases, with a specific emphasis on close-contact interactions between infected and susceptible individuals. Through comprehensive analysis of model simulations, we elucidated the intricacies of disease transmission dynamics within conference settings and assessed the efficacy of control strategies to curb disease dissemination. Ultimately, our study proffers a numerical framework for assessing the risk of infectious disease transmission during short-duration conferences, furnishing conference organizers with valuable insights to inform the implementation of targeted prevention and control measures.

q-bio.PE↗

arXiv2024

Persuasion or Insulting? Unpacking Discursive Strategies of Gender Debate in Everyday Feminism in China

Speaking out for women's daily needs on social media has become a crucial form of everyday feminism in China. Gender debate naturally intertwines with such feminist advocacy, where users in opposite stances discuss gender-related issues through intense discourse. The complexities of gender debate necessitate a systematic understanding of discursive strategies for achieving effective gender communication that balances civility and constructiveness. To address this problem, we adopted a mixed-methods study to navigate discursive strategies in gender debate, focusing on 38,636 posts and 187,539 comments from two representative cases in China. Through open coding, we identified a comprehensive taxonomy of linguistic strategies in gender debate, capturing five overarching themes including derogation, gender distinction, intensification, mitigation, and cognizance guidance. Further, we applied regression analysis to unveil these strategies' correlations with user participation and response, illustrating the tension between debating tactics and public engagement. We discuss design implications to facilitate feminist advocacy on social media.

cs.HC↗

arXiv2024

DPoser: Diffusion Model as Robust 3D Human Pose Prior

This work targets to construct a robust human pose prior. However, it remains a persistent challenge due to biomechanical constraints and diverse human movements. Traditional priors like VAEs and NDFs often exhibit shortcomings in realism and generalization, notably with unseen noisy poses. To address these issues, we introduce DPoser, a robust and versatile human pose prior built upon diffusion models. DPoser regards various pose-centric tasks as inverse problems and employs variational diffusion sampling for efficient solving. Accordingly, designed with optimization frameworks, DPoser seamlessly benefits human mesh recovery, pose generation, pose completion, and motion denoising tasks. Furthermore, due to the disparity between the articulated poses and structured images, we propose truncated timestep scheduling to enhance the effectiveness of DPoser. Our approach demonstrates considerable enhancements over common uniform scheduling used in image domains, boasting improvements of 5.4%, 17.2%, and 3.8% across human mesh recovery, pose completion, and motion denoising, respectively. Comprehensive experiments demonstrate the superiority of DPoser over existing state-of-the-art pose priors across multiple tasks.

cs.CV↗

arXiv2024

Multilingual Jailbreak Challenges in Large Language Models

While large language models (LLMs) exhibit remarkable capabilities across a wide range of tasks, they pose potential safety concerns, such as the ``jailbreak'' problem, wherein malicious instructions can manipulate LLMs to exhibit undesirable behavior. Although several preventive measures have been developed to mitigate the potential risks associated with LLMs, they have primarily focused on English. In this study, we reveal the presence of multilingual jailbreak challenges within LLMs and consider two potential risky scenarios: unintentional and intentional. The unintentional scenario involves users querying LLMs using non-English prompts and inadvertently bypassing the safety mechanisms, while the intentional scenario concerns malicious users combining malicious instructions with multilingual prompts to deliberately attack LLMs. The experimental results reveal that in the unintentional scenario, the rate of unsafe content increases as the availability of languages decreases. Specifically, low-resource languages exhibit about three times the likelihood of encountering harmful content compared to high-resource languages, with both ChatGPT and GPT-4. In the intentional scenario, multilingual prompts can exacerbate the negative impact of malicious instructions, with astonishingly high rates of unsafe output: 80.92\% for ChatGPT and 40.71\% for GPT-4. To handle such a challenge in the multilingual context, we propose a novel \textsc{Self-Defense} framework that automatically generates multilingual training data for safety fine-tuning. Experimental results show that ChatGPT fine-tuned with such data can achieve a substantial reduction in unsafe content generation. Data is available at \url{https://github.com/DAMO-NLP-SG/multilingual-safety-for-LLMs}.

cs.CL↗

arXiv2024

Understanding Emotional Disclosure via Diary-keeping in Quarantine on Social Media

Quarantine is a widely-adopted measure during health crises caused by highly-contagious diseases like COVID-19, yet it poses critical challenges to public mental health. Given this context, emotional disclosure on social media in the form of keeping a diary emerges as a popular way for individuals to express emotions and record their mental health status. However, the exploration of emotional disclosure via diary-keeping on social media during quarantine is underexplored, understanding which could be beneficial to facilitate emotional connections and enlighten health intervention measures. Focusing on this particular form of self-disclosure, this work proposes a quantitative approach to figure out the prevalence and changing patterns of emotional disclosure during quarantine, and the possible factors contributing to the negative emotions. We collected 58, 796 posts with the "Quarantine Diary" keyword on Weibo, a popular social media website in China. Through text classification, we capture diverse emotion categories that characterize public emotion disclosure during quarantine, such as annoyed, anxious, boring, happy, hopeful and appreciative. Based on temporal analysis, we uncover the changing patterns of emotional disclosure from long-term perspectives and period-based perspectives (e.g., the gradual decline of all negative emotions and the upsurge of the annoyed emotion near the end of quarantine). Leveraging topic modeling, we also encapsulate the possible influencing factors of negative emotions, such as freedom restriction and solitude, and uncertainty of infection and supply. We reflect on how our findings could deepen the understanding of mental health on social media and further provide practical and design implications to mitigate mental health issues during quarantine.

cs.HC↗

arXiv2023

Engage Wider Audience or Facilitate Quality Answers? a Mixed-methods Analysis of Questioning Strategies for Research Sensemaking on a Community Q&A Site

Discussing research-sensemaking questions on Community Question and Answering (CQA) platforms has been an increasingly common practice for the public to participate in science communication. Nonetheless, how users strategically craft research-sensemaking questions to engage public participation and facilitate knowledge construction is a significant yet less understood problem. To fill this gap, we collected 837 science-related questions and 157,684 answers from Zhihu, and conducted a mixed-methods study to explore user-developed strategies in proposing research-sensemaking questions, and their potential effects on public engagement and knowledge construction. Through open coding, we captured a comprehensive taxonomy of question-crafting strategies, such as eyecatching narratives with counter-intuitive claims and rigorous descriptions with data use. Regression analysis indicated that these strategies correlated with user engagement and answer construction in different ways (e.g., emotional questions attracted more views and answers), yet there existed a general divergence between wide participation and quality knowledge establishment, when most questioning strategies could not ensure both. Based on log analysis, we further found that collaborative editing afforded unique values in refining research-sensemaking questions regarding accuracy, rigor, comprehensiveness and attractiveness. We propose design implications to facilitate accessible, accurate and engaging science communication on CQA platforms.

cs.HC↗