arXiv subjects

Yang Song

Publications and source records attributed to Yang Song.

At least 73 records · Page 4Linked to original sources

arXiv2026

Elo-Evolve: A Co-evolutionary Framework for Language Model Alignment

Current alignment methods for Large Language Models (LLMs) rely on compressing vast amounts of human preference data into static, absolute reward functions, leading to data scarcity, noise sensitivity, and training instability. We introduce Elo-Evolve, a co-evolutionary framework that redefines alignment as dynamic multi-agent competition within an adaptive opponent pool. Our approach makes two key innovations: (1) eliminating Bradley-Terry model dependencies by learning directly from binary win/loss outcomes in pairwise competitions, and (2) implementing Elo-orchestrated opponent selection that provides automatic curriculum learning through temperature-controlled sampling. We ground our approach in PAC learning theory, demonstrating that pairwise comparison achieves superior sample complexity and empirically validate a 4.5x noise reduction compared to absolute scoring approaches. Experimentally, we train a Qwen2.5-7B model using our framework with opponents including Qwen2.5-14B, Qwen2.5-32B, and Qwen3-8B models. Results demonstrate a clear performance hierarchy: point-based methods < static pairwise training < Elo-Evolve across Alpaca Eval 2.0 and MT-Bench, validating the progressive benefits of pairwise comparison and dynamic opponent selection for LLM alignment.

cs.CL↗

arXiv2026

SWE-Master: Unleashing the Potential of Software Engineering Agents via Post-Training

In this technical report, we present SWE-Master, an open-source and fully reproducible post-training framework for building effective software engineering agents. SWE-Master systematically explores the complete agent development pipeline, including teacher-trajectory synthesis and data curation, long-horizon SFT, RL with real execution feedback, and inference framework design. Starting from an open-source base model with limited initial SWE capability, SWE-Master demonstrates how systematical optimization method can elicit strong long-horizon SWE task solving abilities. We evaluate SWE-Master on SWE-bench Verified, a standard benchmark for realistic software engineering tasks. Under identical experimental settings, our approach achieves a resolve rate of 61.4\% with Qwen2.5-Coder-32B, substantially outperforming existing open-source baselines. By further incorporating test-time scaling~(TTS) with LLM-based environment feedback, SWE-Master reaches 70.8\% at TTS@8, demonstrating a strong performance potential. SWE-Master provides a practical and transparent foundation for advancing reproducible research on software engineering agents. The code is available at https://github.com/RUCAIBox/SWE-Master.

cs.SE↗

arXiv2026

FOCA: Frequency-Oriented Cross-Domain Forgery Detection, Localization and Explanation via Multi-Modal Large Language Model

Advances in image tampering techniques, particularly generative models, pose significant challenges to media verification, digital forensics, and public trust. Existing image forgery detection and localization (IFDL) methods suffer from two key limitations: over-reliance on semantic content while neglecting textural cues, and limited interpretability of subtle low-level tampering traces. To address these issues, we propose FOCA, a multimodal large language model-based framework that integrates discriminative features from both the RGB spatial and frequency domains via a cross-attention fusion module. This design enables accurate forgery detection and localization while providing explicit, human-interpretable cross-domain explanations. We further introduce FSE-Set, a large-scale dataset with diverse authentic and tampered images, pixel-level masks, and dual-domain annotations. Extensive experiments show that FOCA outperforms state-of-the-art methods in detection performance and interpretability across both spatial and frequency domains.

cs.CV↗

arXiv2026

Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts

We present Nanbeige4.1-3B, a unified generalist language model that simultaneously achieves strong agentic behavior, code generation, and general reasoning with only 3B parameters. To the best of our knowledge, it is the first open-source small language model (SLM) to achieve such versatility in a single model. To improve reasoning and preference alignment, we combine point-wise and pair-wise reward modeling, ensuring high-quality, human-aligned responses. For code generation, we design complexity-aware rewards in Reinforcement Learning, optimizing both correctness and efficiency. In deep search, we perform complex data synthesis and incorporate turn-level supervision during training. This enables stable long-horizon tool interactions, allowing Nanbeige4.1-3B to reliably execute up to 600 tool-call turns for complex problem-solving. Extensive experimental results show that Nanbeige4.1-3B significantly outperforms prior models of similar scale, such as Nanbeige4-3B-2511 and Qwen3-4B, even achieving superior performance compared to much larger models, such as Qwen3-30B-A3B. Our results demonstrate that small models can achieve both broad competence and strong specialization simultaneously, redefining the potential of 3B parameter models.

cs.AI↗

arXiv2026

PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning

Language Reasoning Models (LRMs) achieve strong performance by scaling test-time computation but often suffer from ``overthinking'', producing excessively long reasoning traces that increase latency and memory usage. Existing LRMs typically enforce conciseness with uniform length penalties, which over-compress crucial early deduction steps at the sequence level and indiscriminately penalize all queries at the group level. To solve these limitations, we propose \textbf{\model}, a dual-level framework for prefix-protected and difficulty-aware compression under hierarchical supervision. At the sequence level, prefix-protected optimization employs decaying mixed rollouts to maintain valid reasoning paths while promoting conciseness. At the group level, difficulty-aware penalty dynamically scales length constraints based on query complexity, maintaining exploration for harder questions while curbing redundancy on easier ones. Extensive experiments on DeepSeek-R1-Distill-Qwen (1.5B/7B) demonstrate that \model achieves a substantial reduction in token usage (up to \textbf{55.7\%}) while simultaneously improving accuracy (up to \textbf{4.1\%}) on math benchmarks, with generalization ability to code, science, and general domains.

cs.CL↗

arXiv2026

Visual Prompt-Agnostic Evolution

Visual Prompt Tuning (VPT) adapts a frozen Vision Transformer (ViT) to downstream tasks by inserting a small number of learnable prompt tokens into the token sequence at each layer. However, we observe that existing VPT variants often suffer from unstable training dynamics, characterized by gradient oscillations. A layer-wise analysis reveals that shallow-layer prompts tend to stagnate early, while deeper-layer prompts exhibit high-variance oscillations, leading to cross-layer mismatch. These issues slow convergence and degrade final performance. To address these challenges, we propose Prompt-Agnostic Evolution ($\mathtt{PAE}$), which strengthens vision prompt tuning by explicitly modeling prompt dynamics. From a frequency-domain perspective, we initialize prompts in a task-aware direction by uncovering and propagating frequency shortcut patterns that the backbone inherently exploits for recognition. To ensure coherent evolution across layers, we employ a shared Koopman operator that imposes a global linear transformation instead of uncoordinated, layer-specific updates. Finally, inspired by Lyapunov stability theory, we introduce a regularizer that constrains error amplification during evolution. Extensive experiments show that $\mathtt{PAE}$ accelerates convergence with an average $1.41\times$ speedup and improves accuracy by 1-3% on 25 datasets across multiple downstream tasks. Beyond performance, $\mathtt{PAE}$ is prompt-agnostic and lightweight, and it integrates seamlessly with diverse VPT variants without backbone modification or inference-time changes.

cs.CV↗

arXiv2026

SWE-World: Building Software Engineering Agents in Docker-Free Environments

Recent advances in large language models (LLMs) have enabled software engineering agents to tackle complex code modification tasks. Most existing approaches rely on execution feedback from containerized environments, which require dependency-complete setup and physical execution of programs and tests. While effective, this paradigm is resource-intensive and difficult to maintain, substantially complicating agent training and limiting scalability. We propose SWE-World, a Docker-free framework that replaces physical execution environments with a learned surrogate for training and evaluating software engineering agents. SWE-World leverages LLM-based models trained on real agent-environment interaction data to predict intermediate execution outcomes and final test feedback, enabling agents to learn without interacting with physical containerized environments. This design preserves the standard agent-environment interaction loop while eliminating the need for costly environment construction and maintenance during agent optimization and evaluation. Furthermore, because SWE-World can simulate the final evaluation outcomes of candidate trajectories without real submission, it enables selecting the best solution among multiple test-time attempts, thereby facilitating effective test-time scaling (TTS) in software engineering tasks. Experiments on SWE-bench Verified demonstrate that SWE-World raises Qwen2.5-Coder-32B from 6.2\% to 52.0\% via Docker-free SFT, 55.0\% with Docker-free RL, and 68.2\% with further TTS. The code is available at https://github.com/RUCAIBox/SWE-World

cs.SE↗

arXiv2026

HPE: Hallucinated Positive Entanglement for Backdoor Attacks in Federated Self-Supervised Learning

Federated self-supervised learning (FSSL) enables collaborative training of self-supervised representation models without sharing raw unlabeled data. While it serves as a crucial paradigm for privacy-preserving learning, its security remains vulnerable to backdoor attacks, where malicious clients manipulate local training to inject targeted backdoors. Existing FSSL attack methods, however, often suffer from low utilization of poisoned samples, limited transferability, and weak persistence. To address these limitations, we propose a new backdoor attack method for FSSL, namely Hallucinated Positive Entanglement (HPE). HPE first employs hallucination-based augmentation using synthetic positive samples to enhance the encoder's embedding of backdoor features. It then introduces feature entanglement to enforce tight binding between triggers and backdoor samples in the representation space. Finally, selective parameter poisoning and proximity-aware updates constrain the poisoned model within the vicinity of the global model, enhancing its stability and persistence. Experimental results on several FSSL scenarios and datasets show that HPE significantly outperforms existing backdoor attack methods in performance and exhibits strong robustness under various defense mechanisms.

cs.CR↗

arXiv2026

LLM Latent Reasoning as Chain of Superposition

Latent reasoning offers a computation-efficient alternative to Chain-of-Thought but often suffers from performance degradation due to distributional misalignment and ambiguous chain definitions. Ideally, latent reasoning should function as a superposition of multiple reasoning paths. To realize this, we introduce Latent-SFT, a unified framework addressing challenges at three levels: token, chain, and learning. First, we define the Latent-Vocab to constrain hidden states within the pre-trained vocab-space. Second, we construct the Latent-Chain via Induction-Supervision Masking to ensure semantic compactness and sufficiency. Third, we employ Latent-Optim with stochastic Gumbel-Softmax to guide the model toward generalizable solutions. Empirical results demonstrate that Latent-SFT consistently outperforms explicit SFT across six mathematical benchmarks (e.g., GSM8k, AIME24) while achieving a 2.7x to 5.5x reduction in reasoning length. Analysis confirms that our method effectively captures a superposition of diverse reasoning trajectories rather than merely compressing a single path.

cs.CL↗

arXiv2026

FAIRT2V: Training-Free Debiasing for Text-to-Video Diffusion Models

Text-to-video (T2V) diffusion models have achieved rapid progress, yet their demographic biases, particularly gender bias, remain largely unexplored. We present FairT2V, a training-free debiasing framework for text-to-video generation that mitigates encoder-induced bias without finetuning. We first analyze demographic bias in T2V models and show that it primarily originates from pretrained text encoders, which encode implicit gender associations even for neutral prompts. We quantify this effect with a gender-leaning score that correlates with bias in generated videos. Based on this insight, FairT2V mitigates demographic bias by neutralizing prompt embeddings via anchor-based spherical geodesic transformations while preserving semantics. To maintain temporal coherence, we apply debiasing only during early identity-forming steps through a dynamic denoising schedule. We further propose a video-level fairness evaluation protocol combining VideoLLM-based reasoning with human verification. Experiments on the modern T2V model Open-Sora show that FairT2V substantially reduces demographic bias across occupations with minimal impact on video quality.

cs.CV↗

arXiv2026

When Rules Fall Short: Agent-Driven Discovery of Emerging Content Issues in Short Video Platforms

Trends on short-video platforms evolve at a rapid pace, with new content issues emerging every day that fall outside the coverage of existing annotation policies. However, traditional human-driven discovery of emerging issues is too slow, which leads to delayed updates of annotation policies and poses a major challenge for effective content governance. In this work, we propose an automatic issue discovery method based on multimodal LLM agents. Our approach automatically recalls short videos containing potential new issues and applies a two-stage clustering strategy to group them, with each cluster corresponding to a newly discovered issue. The agent then generates updated annotation policies from these clusters, thereby extending coverage to these emerging issues. Our agent has been deployed in the real system. Both offline and online experiments demonstrate that this agent-based method significantly improves the effectiveness of emerging-issue discovery (with an F1 score improvement of over 20%) and enhances the performance of subsequent issue governance (reducing the view count of problematic videos by approximately 15%). More importantly, compared to manual issue discovery, it greatly reduces time costs and substantially accelerates the iteration of annotation policies.

cs.CV↗

arXiv2026

Reinforcement Learning for Follow-the-Leader Robotic Endoscopic Navigation via Synthetic Data

Autonomous navigation is crucial for both medical and industrial endoscopic robots, enabling safe and efficient exploration of narrow tubular environments without continuous human intervention, where avoiding contact with the inner walls has been a longstanding challenge for prior approaches. We present a follow-the-leader endoscopic robot based on a flexible continuum structure designed to minimize contact between the endoscope body and intestinal walls, thereby reducing patient discomfort. To achieve this objective, we propose a vision-based deep reinforcement learning framework guided by monocular depth estimation. A realistic intestinal simulation environment was constructed in \textit{NVIDIA Omniverse} to train and evaluate autonomous navigation strategies. Furthermore, thousands of synthetic intraluminal images were generated using NVIDIA Replicator to fine-tune the Depth Anything model, enabling dense three-dimensional perception of the intestinal environment with a single monocular camera. Subsequently, we introduce a geometry-aware reward and penalty mechanism to enable accurate lumen tracking. Compared with the original Depth Anything model, our method improves $δ_{1}$ depth accuracy by 39.2% and reduces the navigation J-index by 0.67 relative to the second-best method, demonstrating the robustness and effectiveness of the proposed approach.

cs.RO↗

arXiv2026

A Universal and Robust Framework for Multiple Gas Recognition Based-on Spherical Normalization-Coupled Mahalanobis Algorithm

Electronic nose (E-nose) systems face two interconnected challenges in open-set gas recognition: feature distribution shift caused by signal drift and decision boundary failure induced by unknown gas interference. Existing methods predominantly rely on Euclidean distance or conventional classifiers, failing to account for anisotropic feature distributions and dynamic signal intensity variations. To address these issues, this study proposes the Spherical Normalization coupled Mahalanobis (SNM) module, a universal post-processing module for open-set gas recognition. First, it achieves geometric decoupling through cascaded batch and L2 normalization, projecting features onto a unit hypersphere to eliminate signal intensity fluctuations. Second, it utilizes Mahalanobis distance to construct adaptive ellipsoidal decision boundaries that conform to the anisotropic feature geometry. The architecture-agnostic SNM-Module seamlessly integrates with mainstream backbones including Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Transformer. Experiments on the public Vergara dataset demonstrate that the Transformer+SNM configuration achieves near-theoretical-limit performance in discriminating among multiple target gases, with an AUROC of 0.9977 and an unknown gas detection rate of 99.57% at 5% false positive rate, significantly outperforming state-of-the-art methods with a 3.0% AUROC improvement and 91.0% standard deviation reduction compared to Class Anchor Clustering (CAC). The module maintains exceptional robustness across five sensor positions, with standard deviations below 0.0028. This work effectively addresses the critical challenge of simultaneously achieving high accuracy and high stability in open-set gas recognition, providing solid support for industrial E-nose deployment.

cs.LG↗

arXiv2025

Principles2Plan: LLM-Guided System for Operationalising Ethical Principles into Plans

Ethical awareness is critical for robots operating in human environments, yet existing automated planning tools provide little support. Manually specifying ethical rules is labour-intensive and highly context-specific. We present Principles2Plan, an interactive research prototype demonstrating how a human and a Large Language Model (LLM) can collaborate to produce context-sensitive ethical rules and guide automated planning. A domain expert provides the planning domain, problem details, and relevant high-level principles such as beneficence and privacy. The system generates operationalisable ethical rules consistent with these principles, which the user can review, prioritise, and supply to a planner to produce ethically-informed plans. To our knowledge, no prior system supports users in generating principle-grounded rules for classical planning contexts. Principles2Plan showcases the potential of human-LLM collaboration for making ethical automated planning more practical and feasible.

cs.AI↗

arXiv2025

Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models

We present Nanbeige4-3B, a family of small-scale but high-performing language models. Pretrained on 23T high-quality tokens and finetuned on over 30 million diverse instructions, we extend the boundary of the scaling law for small language models. In pre-training, we design a Fine-Grained Warmup-Stable-Decay (FG-WSD) training scheduler, which progressively refines data mixtures across stages to boost model performance. In post-training, to improve the quality of the SFT data, we design a joint mechanism that integrates deliberative generation refinement and chain-of-thought reconstruction, yielding substantial gains on complex tasks. Following SFT, we employ our flagship reasoning model to distill Nanbeige4-3B through our proposed Dual Preference Distillation (DPD) method, which leads to further performance gains. Finally, a multi-stage reinforcement learning phase was applied, leveraging verifiable rewards and preference modeling to strengthen abilities on both reasoning and human alignment. Extensive evaluations show that Nanbeige4-3B not only significantly outperforms models of comparable parameter scale but also rivals much larger models across a wide range of benchmarks. The model checkpoints are available at https://huggingface.co/Nanbeige.

cs.CL↗

arXiv2025

ToolMind Technical Report: A Large-Scale, Reasoning-Enhanced Tool-Use Dataset

Large Language Model (LLM) agents have developed rapidly in recent years to solve complex real-world problems using external tools. However, the scarcity of high-quality trajectories still hinders the development of stronger LLM agents. Most existing works on multi-turn dialogue synthesis validate correctness only at the trajectory level, which may overlook turn-level errors that can propagate during training and degrade model performance. To address these limitations, we introduce ToolMind, a large-scale, high-quality tool-agentic dataset with 160k synthetic data instances generated using over 20k tools and 200k augmented open-source data instances. Our data synthesis pipeline first constructs a function graph based on parameter correlations and then uses a multi-agent framework to simulate realistic user-assistant-tool interactions. Beyond trajectory-level validation, we employ fine-grained turn-level filtering to remove erroneous or suboptimal steps, ensuring that only high-quality reasoning traces are retained. This approach mitigates error amplification during training while preserving self-corrective reasoning signals essential for robust tool-use learning. Models fine-tuned on ToolMind show significant improvements over baselines on several benchmarks.

cs.AI↗

arXiv2025

Cross-Stain Contrastive Learning for Paired Immunohistochemistry and Histopathology Slide Representation Learning

Universal, transferable whole-slide image (WSI) representations are central to computational pathology. Incorporating multiple markers (e.g., immunohistochemistry, IHC) alongside H&E enriches H&E-based features with diverse, biologically meaningful information. However, progress is limited by the scarcity of well-aligned multi-stain datasets. Inter-stain misalignment shifts corresponding tissue across slides, hindering consistent patch-level features and degrading slide-level embeddings. To address this, we curated a slide-level aligned, five-stain dataset (H&E, HER2, KI67, ER, PGR) to enable paired H&E-IHC learning and robust cross-stain representation. Leveraging this dataset, we propose Cross-Stain Contrastive Learning (CSCL), a two-stage pretraining framework with a lightweight adapter trained using patch-wise contrastive alignment to improve the compatibility of H&E features with corresponding IHC-derived contextual cues, and slide-level representation learning with Multiple Instance Learning (MIL), which uses a cross-stain attention fusion module to integrate stain-specific patch features and a cross-stain global alignment module to enforce consistency among slide-level embeddings across different stains. Experiments on cancer subtype classification, IHC biomarker status classification, and survival prediction show consistent gains, yielding high-quality, transferable H&E slide-level representations. The code and data are available at https://github.com/lily-zyz/CSCL.

cs.CV↗

arXiv2025

CoSineVerifier: Tool-Augmented Answer Verification for Computation-Oriented Scientific Questions

Answer verification methods are widely employed in language model training pipelines spanning data curation, evaluation, and reinforcement learning with verifiable rewards (RLVR). While prior work focus on developing unified verifiers applicable across multiple reasoning scenarios, significant challenges remain in computation-oriented scientific domains, such as algebraic equivalence checking and physical constant substitution. In this paper, we introduce \model, a tool-augmented verifier that leverages external executors to perform precise computations and symbolic simplifications. \model enables robust verification that goes beyond simple semantic matching. We propose a novel two-stage pipeline, which begin with cold-start fine-tuning and followed by multi-turn reinforcement learning with tool integration. Extensive experiments conducted on STEM subjects, general QA, and long-form reasoning tasks demonstrates strong generalization of \model. The results shows that the \model achieves state-of-the-art performance on VerifyBench-Hard and SCI-Bench. And we also employ our \model in RLVR as a reward model, the results show that it consistently outperforms both rubric-based and model-based verifiers on AIME'24 and AIME'25, demonstrating strong potential to enhance reasoning capabilities of LLM. Our model is released at \hyperlink{https://huggingface.co/Nanbeige/CoSineVerifier-Tool-4B}{https://huggingface.co/Nanbeige/CoSineVerifier-Tool-4B}.

cs.LG↗