Searcharxiv⌕ Search

arXiv subjects

Dawei Yin

Publications and source records attributed to Dawei Yin.

At least 73 records · Page 4Linked to original sources
arXiv2026

FlexSpec: Frozen Drafts Meet Evolving Targets in Edge-Cloud Collaborative LLM Speculative Decoding

Deploying large language models (LLMs) in mobile and edge computing environments is constrained by limited on-device resources, scarce wireless bandwidth, and frequent model evolution. Although edge-cloud collaborative inference with speculative decoding (SD) can reduce end-to-end latency by executing a lightweight draft model at the edge and verifying it with a cloud-side target model, existing frameworks fundamentally rely on tight coupling between the two models. Consequently, repeated model synchronization introduces excessive communication overhead, increasing end-to-end latency, and ultimately limiting the scalability of SD in edge environments. To address these limitations, we propose FlexSpec, a communication-efficient collaborative inference framework tailored for evolving edge-cloud systems. The core design of FlexSpec is a shared-backbone architecture that allows a single and static edge-side draft model to remain compatible with a large family of evolving cloud-side target models. By decoupling edge deployment from cloud-side model updates, FlexSpec eliminates the need for edge-side retraining or repeated model downloads, substantially reducing communication and maintenance costs. Furthermore, to accommodate time-varying wireless conditions and heterogeneous device constraints, we develop a channel-aware adaptive speculation mechanism that dynamically adjusts the speculative draft length based on real-time channel state information and device energy budgets. Extensive experiments demonstrate that FlexSpec achieves superior performance compared to conventional SD approaches in terms of inference efficiency.

cs.DC↗
arXiv2025

Thinking Forward and Backward: Multi-Objective Reinforcement Learning for Retrieval-Augmented Reasoning

Retrieval-augmented generation (RAG) has proven to be effective in mitigating hallucinations in large language models, yet its effectiveness remains limited in complex, multi-step reasoning scenarios. Recent efforts have incorporated search-based interactions into RAG, enabling iterative reasoning with real-time retrieval. Most approaches rely on outcome-based supervision, offering no explicit guidance for intermediate steps. This often leads to reward hacking and degraded response quality. We propose Bi-RAR, a novel retrieval-augmented reasoning framework that evaluates each intermediate step jointly in both forward and backward directions. To assess the information completeness of each step, we introduce a bidirectional information distance grounded in Kolmogorov complexity, approximated via language model generation probabilities. This quantification measures both how far the current reasoning is from the answer and how well it addresses the question. To optimize reasoning under these bidirectional signals, we adopt a multi-objective reinforcement learning framework with a cascading reward structure that emphasizes early trajectory alignment. Empirical results on seven question answering benchmarks demonstrate that Bi-RAR surpasses previous methods and enables efficient interaction and reasoning with the search engine during training and inference.

cs.CL↗
arXiv2025

Grounding Long-Context Reasoning with Contextual Normalization for Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has become an essential approach for extending the reasoning and knowledge capacity of large language models (LLMs). While prior research has primarily focused on retrieval quality and prompting strategies, the influence of how the retrieved documents are framed, i.e., context format, remains underexplored. We show that seemingly superficial choices, such as delimiters or structural markers in key-value extraction, can induce substantial shifts in accuracy and stability, even when semantic content is identical. To systematically investigate this effect, we design controlled experiments that vary context density, delimiter styles, and positional placement, revealing the underlying factors that govern performance differences. Building on these insights, we introduce Contextual Normalization, a lightweight strategy that adaptively standardizes context representations before generation. Extensive experiments on both controlled and real-world RAG benchmarks across diverse settings demonstrate that the proposed strategy consistently improves robustness to order variation and strengthens long-context utilization. These findings underscore that reliable RAG depends not only on retrieving the right content, but also on how that content is presented, offering both new empirical evidence and a practical technique for better long-context reasoning.

cs.CL↗
arXiv2025

COOPER: A Unified Model for Cooperative Perception and Reasoning in Spatial Intelligence

Visual Spatial Reasoning is crucial for enabling Multimodal Large Language Models (MLLMs) to understand object properties and spatial relationships, yet current models still struggle with 3D-aware reasoning. Existing approaches typically enhance either perception, by augmenting RGB inputs with auxiliary modalities such as depth and segmentation, or reasoning, by training on spatial VQA datasets and applying reinforcement learning, and thus treat these two aspects in isolation. In this work, we investigate whether a unified MLLM can develop an intrinsic ability to enhance spatial perception and, through adaptive interleaved reasoning, achieve stronger spatial intelligence. We propose \textbf{COOPER}, a unified MLLM that leverages depth and segmentation as auxiliary modalities and is trained in two stages to acquire auxiliary modality generation and adaptive, interleaved reasoning capabilities. COOPER achieves an average \textbf{6.91\%} improvement in spatial reasoning while maintaining general performance. Moreover, even a variant trained only for auxiliary modality generation attains a \textbf{7.92\%} gain on distance and size estimation, suggesting that learning to generate auxiliary modalities helps internalize spatial knowledge and strengthen spatial understanding.

cs.CV↗
arXiv2025

Beyond ReAct: A Planner-Centric Framework for Complex Tool-Augmented LLM Reasoning

Existing tool-augmented large language models (LLMs) encounter significant challenges when processing complex queries. Current frameworks such as ReAct are prone to local optimization traps due to their reliance on incremental decision-making processes. To address these limitations, we propose a novel Planner-centric Plan-Execute paradigm that fundamentally resolves local optimization bottlenecks through architectural innovation. Central to our approach is a novel Planner model that performs global Directed Acyclic Graph (DAG) planning for complex queries, enabling optimized execution beyond conventional tool coordination. We also introduce ComplexTool-Plan, a large-scale benchmark dataset featuring complex queries that demand sophisticated multi-tool composition and coordination capabilities. Additionally, we develop a two-stage training methodology that integrates Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO), systematically enhancing the Planner's tool selection accuracy and global planning awareness through structured DAG-based planning. When integrated with a capable executor, our framework achieves state-of-the-art performance on the StableToolBench benchmark for complex user queries, demonstrating superior end-to-end execution capabilities and robust handling of intricate multi-tool workflows.

cs.AI↗
arXiv2025

On the Performance of Low-complexity Decoders of LDPC Codes

Efficient decoding is crucial to high-throughput and power-sensitive wireless communication scenarios. A theoretical analysis of the performance-complexity tradeoff toward low-complexity decoding is required for a better understanding of the fundamental limits in the above-mentioned scenarios. This study aims to explore the performance of LDPC codes under belief propagation (BP) decoding with complexity constraints. In other words, for a small number of iterations, we present a closed-form lower bound on the bit error rate (BER) of LDPC codes as a function of complexity. Specifically, for the regular LDPC code ensembles, the dominant term in the order of the lower bound we provide matches that of the upper bound given by Lentmaier. Furthermore, for irregular LDPC code ensembles, in addition to adopting the approach used for regular codes, we also propose a method to iteratively obtain the lower bound on the average BER.

cs.IT↗
arXiv2025

Embedding in Recommender Systems: A Survey

Recommender systems have become an essential component of many online platforms, providing personalized recommendations to users. A crucial aspect is embedding techniques that convert the high-dimensional discrete features, such as user and item IDs, into low-dimensional continuous vectors, which can enhance the recommendation performance. Embedding techniques have revolutionized the capture of complex entity relationships, generating significant research interest. This survey presents a comprehensive analysis of recent advances in recommender system embedding techniques. We examine centralized embedding approaches across matrix, sequential, and graph structures. In matrix-based scenarios, collaborative filtering generates embeddings that effectively model user-item preferences, particularly in sparse data environments. For sequential data, we explore various approaches including recurrent neural networks and self-supervised methods such as contrastive and generative learning. In graph-structured contexts, we analyze techniques like node2vec that leverage network relationships, along with applicable self-supervised methods. Our survey addresses critical scalability challenges in embedding methods and explores innovative directions in recommender systems. We introduce emerging approaches, including AutoML, hashing techniques, and quantization methods, to enhance performance while reducing computational complexity. Additionally, we examine the promising role of Large Language Models (LLMs) in embedding enhancement. Through detailed discussion of various architectures and methodologies, this survey aims to provide a thorough overview of state-of-the-art embedding techniques in recommender systems, while highlighting key challenges and future research directions.

cs.IR↗
arXiv2025

An Efficient Framework for Whole-Page Reranking via Single-Modal Supervision

The whole-page reranking plays a critical role in shaping the user experience of search engines, which integrates retrieval results from multiple modalities, such as documents, images, videos, and LLM outputs. Existing methods mainly rely on large-scale human-annotated data, which is costly to obtain and time-consuming. This is because whole-page annotation is far more complex than single-modal: it requires assessing the entire result page while accounting for cross-modal relevance differences. Thus, how to improve whole-page reranking performance while reducing annotation costs is still a key challenge in optimizing search engine result pages(SERP). In this paper, we propose SMAR, a novel whole-page reranking framework that leverages strong Single-modal rankers to guide Modal-wise relevance Alignment for effective Reranking, using only limited whole-page annotation to outperform fully-annotated reranking models. Specifically, high-quality single-modal rankers are first trained on data specific to their respective modalities. Then, for each query, we select a subset of their outputs to construct candidate pages and perform human annotation at the page level. Finally, we train the whole-page reranker using these limited annotations and enforcing consistency with single-modal preferences to maintain ranking quality within each modality. Experiments on the Qilin and Baidu datasets demonstrate that SMAR reduces annotation costs by about 70-90\% while achieving significant ranking improvements compared to baselines. Further offline and online A/B testing on Baidu APPs also shows notable gains in standard ranking metrics as well as user experience indicators, fully validating the effectiveness and practical value of our approach in real-world search scenarios.

cs.IR↗
arXiv2025

VeriCite: Towards Reliable Citations in Retrieval-Augmented Generation via Rigorous Verification

Retrieval-Augmented Generation (RAG) has emerged as a crucial approach for enhancing the responses of large language models (LLMs) with external knowledge sources. Despite the impressive performance in complex question-answering tasks, RAG still struggles with hallucinations. Attributing RAG-generated content through in-line citations has demonstrated potential in reducing hallucinations and facilitating human verification. Existing citation generation methods primarily rely on either fine-tuning the generator or employing post-processing approaches for citation matching. However, the former approach demands substantial annotated data and computational resources, while the latter often encounters difficulties in managing multiple citations and frequently produces suboptimal results. In this paper, we introduce a novel framework, called VeriCite, designed to rigorously validate supporting evidence and enhance answer attribution. Specifically, VeriCite breaks down into a three-stage generation: 1) The initial answer generation first generates a response based on all available contexts and has its claims verified through the NLI model; 2) the supporting evidence selection assesses the utility of each document and extracts useful supporting evidences; 3) the final answer refinement integrates the initial response and collected evidences to produce the final, refined answer.We conduct experiments across five open-source LLMs and four datasets, demonstrating that VeriCite can significantly improve citation quality while maintaining the correctness of the answers.

cs.IR↗
arXiv2025

Utility-Focused LLM Annotation for Retrieval and Retrieval-Augmented Generation

This paper explores the use of large language models (LLMs) for annotating document utility in training retrieval and retrieval-augmented generation (RAG) systems, aiming to reduce dependence on costly human annotations. We address the gap between retrieval relevance and generative utility by employing LLMs to annotate document utility. To effectively utilize multiple positive samples per query, we introduce a novel loss that maximizes their summed marginal likelihood. Using the Qwen-2.5-32B model, we annotate utility on the MS MARCO dataset and conduct retrieval experiments on MS MARCO and BEIR, as well as RAG experiments on MS MARCO QA, NQ, and HotpotQA. Our results show that LLM-generated annotations enhance out-of-domain retrieval performance and improve RAG outcomes compared to models trained solely on human annotations or downstream QA metrics. Furthermore, combining LLM annotations with just 20% of human labels achieves performance comparable to using full human annotations. Our study offers a comprehensive approach to utilizing LLM annotations for initializing QA systems on new corpora.

cs.IR↗
arXiv2025

Distilling a Small Utility-Based Passage Selector to Enhance Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating retrieved information. Standard retrieval process prioritized relevance, focusing on topical alignment between queries and passages. In contrast, in RAG, the emphasis has shifted to utility, which considers the usefulness of passages for generating accurate answers. Despite empirical evidence showing the benefits of utility-based retrieval in RAG, the high computational cost of using LLMs for utility judgments limits the number of passages evaluated. This restriction is problematic for complex queries requiring extensive information. To address this, we propose a method to distill the utility judgment capabilities of LLMs into smaller, more efficient models. Our approach focuses on utility-based selection rather than ranking, enabling dynamic passage selection tailored to specific queries without the need for fixed thresholds. We train student models to learn pseudo-answer generation and utility judgments from teacher LLMs, using a sliding window method that dynamically selects useful passages. Our experiments demonstrate that utility-based selection provides a flexible and cost-effective solution for RAG, significantly reducing computational costs while improving answer quality. We present the distillation results using Qwen3-32B as the teacher model for both relevance ranking and utility-based selection, distilled into RankQwen1.7B and UtilityQwen1.7B. Our findings indicate that for complex questions, utility-based selection is more effective than relevance ranking in enhancing answer generation performance. We will release the relevance ranking and utility-based selection annotations for the MS MARCO dataset, supporting further research in this area.

cs.IR↗
arXiv2025

Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning

Retrieval-augmented generation (RAG) is widely utilized to incorporate external knowledge into large language models, thereby enhancing factuality and reducing hallucinations in question-answering (QA) tasks. A standard RAG pipeline consists of several components, such as query rewriting, document retrieval, document filtering, and answer generation. However, these components are typically optimized separately through supervised fine-tuning, which can lead to misalignments between the objectives of individual components and the overarching aim of generating accurate answers. Although recent efforts have explored using reinforcement learning (RL) to optimize specific RAG components, these approaches often focus on simple pipelines with only two components or do not adequately address the complex interdependencies and collaborative interactions among the modules. To overcome these limitations, we propose treating the complex RAG pipeline with multiple components as a multi-agent cooperative task, in which each component can be regarded as an RL agent. Specifically, we present MMOA-RAG, Multi-Module joint Optimization Algorithm for RAG, which employs multi-agent reinforcement learning to harmonize all agents' goals toward a unified reward, such as the F1 score of the final answer. Experiments conducted on various QA benchmarks demonstrate that MMOA-RAG effectively boost the overall performance of the pipeline and outperforms existing baselines. Furthermore, comprehensive ablation studies validate the contributions of individual components and demonstrate MMOA-RAG can be adapted to different RAG pipelines and benchmarks.

cs.CL↗
arXiv2025

GRAF: Multi-turn Jailbreaking via Global Refinement and Active Fabrication

Large Language Models (LLMs) have demonstrated remarkable performance across diverse tasks. Nevertheless, they still pose notable safety risks due to potential misuse for malicious purposes. Jailbreaking, which seeks to induce models to generate harmful content through single-turn or multi-turn attacks, plays a crucial role in uncovering underlying security vulnerabilities. However, prior methods, including sophisticated multi-turn approaches, often struggle to adapt to the evolving dynamics of dialogue as interactions progress. To address this challenge, we propose \ours (JailBreaking via \textbf{G}lobally \textbf{R}efining and \textbf{A}daptively \textbf{F}abricating), a novel multi-turn jailbreaking method that globally refines the attack trajectory at each interaction. In addition, we actively fabricate model responses to suppress safety-related warnings, thereby increasing the likelihood of eliciting harmful outputs in subsequent queries. Extensive experiments across six state-of-the-art LLMs demonstrate the superior effectiveness of our approach compared to existing single-turn and multi-turn jailbreaking methods. Our code will be released at https://github.com/Ytang520/Multi-Turn_jailbreaking_Global-Refinment_and_Active-Fabrication.

cs.CL↗
arXiv2025

Mitigating Hallucinations in Large Vision-Language Models via Entity-Centric Multimodal Preference Optimization

Large Visual Language Models (LVLMs) have demonstrated impressive capabilities across multiple tasks. However, their trustworthiness is often challenged by hallucinations, which can be attributed to the modality misalignment and the inherent hallucinations of their underlying Large Language Models (LLMs) backbone. Existing preference alignment methods focus on aligning model responses with human preferences while neglecting image-text modality alignment, resulting in over-reliance on LLMs and hallucinations. In this paper, we propose Entity-centric Multimodal Preference Optimization (EMPO), which achieves enhanced modality alignment compared to existing human preference alignment methods. Besides, to overcome the scarcity of high-quality multimodal preference data, we utilize open-source instruction datasets to automatically construct high-quality preference data across three aspects: image, instruction, and response. Experiments on two human preference datasets and five multimodal hallucination benchmarks demonstrate the effectiveness of EMPO, e.g., reducing hallucination rates by 85.9\% on Object-HalBench and 49.8\% on MM-HalBench.

cs.CV↗
arXiv2025

Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking

Large Reasoning Models (LRMs) have recently demonstrated impressive performances across diverse domains. However, how the safety of Large Language Models (LLMs) benefits from enhanced reasoning capabilities against jailbreak queries remains unexplored. To bridge this gap, in this paper, we propose Reasoning-to-Defend (R2D), a novel training paradigm that integrates a safety-aware reasoning mechanism into LLMs' generation process. This enables self-evaluation at each step of the reasoning process, forming safety pivot tokens as indicators of the safety status of responses. Furthermore, in order to improve the accuracy of predicting pivot tokens, we propose Contrastive Pivot Optimization (CPO), which enhances the model's perception of the safety status of given dialogues. LLMs dynamically adjust their response strategies during reasoning, significantly enhancing their safety capabilities defending jailbreak attacks. Extensive experiments demonstrate that R2D effectively mitigates various attacks and improves overall safety, while maintaining the original performances. This highlights the substantial potential of safety-aware reasoning in improving robustness of LRMs and LLMs against various jailbreaks.

cs.CL↗
arXiv2025

Staying in the Sweet Spot: Responsive Reasoning Evolution via Capability-Adaptive Hint Scaffolding

Reinforcement learning with verifiable rewards (RLVR) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs). However, existing RLVR methods often suffer from exploration inefficiency due to mismatches between the training data's difficulty and the model's capability. LLMs fail to discover viable reasoning paths when problems are overly difficult, while learning little new capability when problems are too simple. In this work, we formalize the impact of problem difficulty by quantifying the relationship between loss descent speed and rollout accuracy. Building on this analysis, we propose SEELE, a novel supervision-aided RLVR framework that dynamically adjusts problem difficulty to stay within the high-efficiency region. SEELE augments each training sample by appending a hint (part of a full solution) after the original problem. Unlike previous hint-based approaches, SEELE deliberately and adaptively adjusts the hint length for each problem to achieve an optimal difficulty. To determine the optimal hint length, SEELE employs a multi-round rollout sampling strategy. In each round, it fits an item response theory model to the accuracy-hint pairs collected in preceding rounds to predict the required hint length for the next round. This instance-level, real-time difficulty adjustment aligns problem difficulty with the evolving model capability, thereby improving exploration efficiency. Experimental results show that SEELE outperforms Group Relative Policy Optimization (GRPO) and Supervised Fine-tuning (SFT) by +11.8 and +10.5 points, respectively, and surpasses the best previous supervision-aided approach by +3.6 points on average across six math reasoning benchmarks.

cs.LG↗
arXiv2025

CoRanking: Collaborative Ranking with Small and Large Ranking Agents

Large Language Models (LLMs) have demonstrated superior listwise ranking performance. However, their superior performance often relies on large-scale parameters (\eg, GPT-4) and a repetitive sliding window process, which introduces significant efficiency challenges. In this paper, we propose \textbf{CoRanking}, a novel collaborative ranking framework that combines small and large ranking models for efficient and effective ranking. CoRanking first employs a small-size reranker to pre-rank all the candidate passages, bringing relevant ones to the top part of the list (\eg, top-20). Then, the LLM listwise reranker is applied to only rerank these top-ranked passages instead of the whole list, substantially enhancing overall ranking efficiency. Although more efficient, previous studies have revealed that the LLM listwise reranker have significant positional biases on the order of input passages. Directly feed the top-ranked passages from small reranker may result in the sub-optimal performance of LLM listwise reranker. To alleviate this problem, we introduce a passage order adjuster trained via reinforcement learning, which reorders the top passages from the small reranker to align with the LLM's preferences of passage order. Extensive experiments on three IR benchmarks demonstrate that CoRanking significantly improves efficiency (reducing ranking latency by about 70\%) while achieving even better effectiveness compared to using only the LLM listwise reranker.

cs.CL↗
arXiv2025

Igniting Creative Writing in Small Language Models: LLM-as-a-Judge versus Multi-Agent Refined Rewards

Large Language Models (LLMs) have demonstrated remarkable creative writing capabilities, yet their substantial computational demands hinder widespread use. Enhancing Small Language Models (SLMs) offers a promising alternative, but current methods like Supervised Fine-Tuning (SFT) struggle with novelty, and Reinforcement Learning from Human Feedback (RLHF) is costly. This paper explores two distinct AI-driven reward strategies within a Reinforcement Learning from AI Feedback (RLAIF) framework to ignite the creative writing of a 7B-parameter SLM, specifically for generating Chinese greetings. The first strategy employs a RM trained on high-quality preference data curated by a novel multi-agent rejection sampling framework designed for creative tasks. The second, more novel strategy utilizes a principle-guided LLM-as-a-Judge, whose reward function is optimized via an adversarial training scheme with a reflection mechanism, to directly provide reward signals. Comprehensive experiments reveal that while both approaches significantly enhance creative output over baselines, the principle-guided LLM-as-a-Judge demonstrably yields superior generation quality. Furthermore, it offers notable advantages in training efficiency and reduced dependency on human-annotated data, presenting a more scalable and effective path towards creative SLMs. Our automated evaluation methods also exhibit strong alignment with human judgments. Our code and data are publicly available at https://github.com/weixiaolong94-hub/Igniting-Creative-Writing-in-Small-Language-Models.

cs.CL↗