论文检索

Zhiyi Duan, Hongyu Yuan, Rui Liu

Knowledge Tracing (KT) infers a student’s knowledge state from past interactions to predict future performance. Conventional Deep Learning (DL)-based KT models are typically tied to platform-specific identifiers and latent representations, making them hard to transfer and interpret. Large Language Model (LLM)-based methods can be either ungrounded under prompting or overly domain-dependent under fine-tuning. In addition, most existing KT methods are developed and evaluated under a same-distribution assumption. In real deployments, educational data often arise from heterogeneous platforms with substantial distribution shift, which often degrades generalization. To this end, we propose RAG-KT, a retrieval-augmented paradigm that frames cross-platform KT as reliable context constrained inference with LLMs. It builds a unified multi-source structured context with cross-source alignment via Question Group abstractions and retrieves complementary rich and reliable context for each prediction, enabling grounded prediction and interpretable diagnosis. Experiments on three public KT benchmarks demonstrate consistent gains in accuracy and robustness, including strong performance under cross-platform conditions.

Haiqi Yang, Zhiyuan Li, Yi Chang, Yuan Wu

The Retentive Network (RetNet) has recently emerged as a formidable successor to the Transformer architecture. Although the self-attention mechanism excels at capturing global dependencies, its inherent quadratic complexity imposes significant memory constraints and inhibits scalability during long-sequence modeling. To overcome these challenges, RetNet introduces an innovative retention mechanism that integrates the inductive bias of recurrent neural networks with the parallelizable training advantages of attention-based models. This unified representation allows RetNet to achieve constant-time inference and linear-time training without sacrificing representational capacity. Despite the growing body of research demonstrating the efficacy of RetNet across diverse fields such as natural language processing, computer vision, and time-series analysis, a systematic synthesis of the current literature is currently unavailable. This paper presents the first comprehensive survey of Retentive Networks through a detailed examination of its architectural foundations, core innovations, and specialized variants. Furthermore, we provide a multi-disciplinary analysis of its applications ranging from basic sequence tasks to complex cross-modal scenarios. Finally, we offer prospective insights and suggest strategic avenues for future inquiry to facilitate the continued evolution of RetNet in both academic research and large-scale industrial applications.

Hao Yang, Qinghua Zhao, Lei Li, Lingyi Meng, Mengda Yu

Chain-of-Thought (CoT) prompting significantly enhances model reasoning, yet its internal mechanisms remain poorly understood. We analyze CoT’s operational principles by reversely tracing information flow across decoding, projection, and activation phases. Our quantitative analysis suggests that CoT may serve as a decoding space pruner, leveraging answer templates to guide output generation, with higher template adherence strongly correlating with improved performance. Furthermore, we surprisingly find that CoT modulates neuron engagement in a task-dependent manner: reducing neuron activation in open-domain tasks, yet increasing it in closed-domain scenarios. These findings offer a novel mechanistic interpretability framework and critical insights for the NLP community, enabling targeted CoT interventions to design more efficient and robust prompts. We released our code and data at https://github.com/How-Young-X/cot

Zixu Wang, Bingbing Xu, Yige Yuan, Huawei Shen, Xueqi Cheng

Multi-Agent Systems (MAS) built on large language models typically solve complex tasks by coordinating multiple agents through workflows. Existing approaches generates workflows either at task level or query level, but their relative costs and benefits remain unclear. After rethinking and empirical analyses, we show that query-level workflow generation is not always necessary, since a small set of top-K best task-level workflows together already covers equivalent or even more queries. We further find that exhaustive execution-based task-level evaluation is both extremely token-costly and frequently unreliable. Inspired by the idea of self-evolution and generative reward modeling, we propose a low-cost task-level generation framework \textbf{SCALE}, which means \textbf{S}elf prediction of the optimizer with few shot \textbf{CAL}ibration for \textbf{E}valuation instead of full validation execution. Extensive experiments demonstrate that \textbf{SCALE} maintains competitive performance, with an average degradation of just 0.61% compared to existing approach across multiple datasets, while cutting overall token usage by up to 83%.

Yang Zerui, Weichuan Wang, Yanwei Xu, Linqi Song, Yudai Matsuda, Wei Han, Bo Bai

Existing NL2SQL systems face two critical limitations : (1) they rely on in-context learning with only correct examples, overlooking the rich signal in historical error–fix pairs that could guide more robust self-correction; and (2) test-time scaling (TTS) approaches often decompose questions arbitrarily, producing near-identical SQL candidates across runs and diminishing ensemble gains. Moreover, these methods suffer from a stark accuracy–efficiency trade-off: high performance demands excessive computation, while fast variants compromise quality. We present Memo-SQL, a training-free framework that addresses these issues through two simple ideas: structured decomposition and experience-aware self-correction. Instead of leaving decomposition to chance, we apply three clear strategies, entity-wise, hierarchical, and atomic sequential, to encourage diverse reasoning. For correction, we build a dynamic memory of both successful queries and historical error–fix pairs, and use retrieval-augmented prompting to bring relevant examples into context at inference time, no fine-tuning or external APIs required. On BIRD, Memo-SQL achieves 68.5% execution accuracy, setting a new state of the art among open, zero-fine-tuning methods, while using over 10× fewer resources than prior TTS approaches.

Shuaimin Li, Liyang Fan, Zeyang li, Zhuoyue Wan, Yufang Lin, Shiwen Ni, Feiteng Fang, Hamid Alinejad-Rokny, Yuanfeng Song, Kun Jing 等

Evaluating code large language models (Code LLMs) requires reliable detection of data leakage, where benchmark performance is artificially inflated by exposure to benchmark data during pre-training. Existing approaches either assume access to proprietary training corpora, rely on brittle heuristics such as timestamp filtering, or use external reference sets with manually tuned, non-generalizable thresholds. To address these limitations, we introduce SrDetection, a unified self-referential leakage detection framework for both gray-box (access to model logits) and black-box (access to model outputs) settings. SrDetection generates semantically equivalent variants of a benchmark sample and detects leakage by contrasting the model’s behavior on the original versus its variants, flagging cases where the original is disproportionately easier for the model. We further design a controlled leakage detection testbed and evaluate SrDetection in this environment. Across different models and training stages, SrDetection improves average F1 by 21.52 points in the gray-box setting and 14.46 points in the black-box setting over strong baselines, demonstrating robust, threshold-independent leakage detection. Finally, a gray-box study of 15 widely used Code LLMs on four popular benchmarks reveals benchmark-specific leakage patterns beyond prior overlap-based analyses[Source code and data are available at <https://github.com/SMinL/SrDetectionCode>].

Jiaming Zhou, Yujie Guo, Shiwan Zhao, Yao Lu, Jianye Wang, Haoqin Sun, Hui Wang, Yong Qin

Automatic speech recognition (ASR) for children remains challenging due to developmental variability and the scarcity of high-quality corpora, especially for Mandarin and its dialects. In this paper, we present ChildTalk, a large-scale Chinese child speech corpus designed to address this gap. It contains 112.5 hours of speech from 498 children (aged 2–8) and 500 caregivers, recorded as natural child–caregiver conversations. Unlike prior Mandarin child ASR corpora that mainly release isolated utterances, ChildTalk provides full-length dialogues with complete transcriptions, preserving turn-taking and discourse context. To our knowledge, it is the first publicly available Mandarin child speech corpus with full-length dialogues and systematic coverage of standard Mandarin, eight Mandarin dialect subgroups, and two additional dialects (Southern Min and Jin). We benchmark end-to-end models trained from scratch, large pre-trained ASR models fine-tuned on ChildTalk, omni-modal LLMs in a zero-shot setting, and commercial speech transcription APIs. Fine-tuning on ChildTalk consistently improves both in-domain and cross-domain performance. These results indicate that ChildTalk provides a challenging, broad-coverage testbed for Chinese child ASR, dialect robustness, and dialogue-level modeling. The dataset will be made freely available for all academic purposes.

Hongyi Lan, Jiaqi Song, Zhengjia Zhong, Hui Li, Hong Liu, Xianming Lin, Rongrong Ji

Large Language Models (LLMs) demonstrate strong generation and reasoning abilities, but they still face challenges in long-term memory retention and multi-turn conversational consistency. Existing memory-augmented methods often incorporate full dialog histories without filtering, resulting in information redundancy and inference latency. Inspired by the episodic memory mechanism in human cognition, we abstract conversational context into Episodic Memory Units (EMUs). We then propose a comprehensive framework, Episodic Memory Agent (EMA), along with a filtering decision module called MemDecider. Specifically, EMA organizes and retrieves EMUs to support response generation, while MemDecider filters information to reduce noise and improve overall performance. Experiments on two widely-used benchmarks show that EMA maintains competitive performance, and integrating MemDecider into other methods reduces their token consumption by an average of 11.48% while effectively improving the overall performance. Code is available at https://github.com/Hongyi4221/EMA.

Yaoyuan Zhang, Zonghao Ying, Aishan Liu, Jian Yang, Tianlin Li, Yaodong Yang, Xianglong Liu

Large language models (LLMs) exhibit growing safety and alignment risks, hindering their deployment in high-stakes decision-making scenarios. In this paper, we identify a previously underexplored risk: similar to humans, LLMs can exhibit egoistic decision-making, in which they pursue short-term self-benefits through improper means while disregarding collective welfare and ethical constraints. We term this phenomenon Strategic Egoism (SE). To systematically evaluate SE, we introduce SEBench, a benchmark comprising 880 decision-making scenarios across 11 domains involving explicit profit temptations, which measures egoistic behavior along 6 psychologically grounded dimensions (e.g., rule circumvention). Each scenario adopts a single-role decision-making setting with carefully designed choice options to elicit self-serving strategies. Extensive experiments on 9 proprietary LLMs reveal that SE behaviors are widespread, with an average occurrence rate of 67.96%, and frequently manifest as manipulative coercion. Notably, we find that models more susceptible to profit temptations also exhibit broader safety deficiencies, including higher toxicity, lower truthfulness, increased jailbreak vulnerability, and elevated Dark Triad–style trait scores. Drawing inspiration from psychological interventions, we further propose SEGuard, a lightweight mitigation that reinforces situational constraints and suppresses egoistic tactics.

TianZe Zhang, Sirui Sun, Yuhang Xie, Xin Zhang, Zhiqiang Wu, Guojie Song

Large Language Models (LLMs) have shown promise in simulating human behavior, yet existing agents often exhibit behavioral rigidity, a flaw frequently masked by the self-referential bias of current "LLM-as-a-judge" evaluations. By evaluating against empirical ground truth, we reveal a counter-intuitive phenomenon: increasing the intensity of prompt-driven reasoning does not enhance fidelity but rather exacerbates value polarization, collapsing population diversity. To address this, we propose the Context-Value-Action (CVA) architecture, grounded in the Stimulus-Organism-Response (S-O-R) model and Schwartz’s Theory of Basic Human Values. Unlike methods relying on self-verification, CVA decouples action generation from cognitive reasoning via a novel Value Verifier trained on authentic human data to explicitly model dynamic value activation. Experiments on CVABench, which comprises over 1.1 million real-world interaction traces, demonstrate that CVA significantly outperforms baselines. Our approach effectively mitigates polarization while offering superior behavioral fidelity and interpretability.

Jonggwon Park, Byungmu Yoon, Soobum Kim, Kyoyun Choi

Automated radiology report generation (RRG) holds potential to reduce the workload of radiologists, and recent advances in multimodal large language models (MLLMs) have enabled multimodal chest X-ray (CXR) report generation. However, existing MLLMs are computationally expensive, require large-scale training data, and may produce hallucinated content, limiting their practical deployment. To address these limitations, we propose RA-RRG, a retrieval-augmented RRG framework that combines multimodal retrieval with large language models (LLMs) to generate radiology reports while reducing hallucinations and computational demands. RA-RRG uses LLMs to extract clinically essential key phrases from radiology reports and retrieves relevant phrases given an input image. By conditioning LLMs on the retrieved phrases, RA-RRG effectively suppresses hallucinations while maintaining strong report generation performance. Experiments on the MIMIC-CXR and IU X-ray datasets show state-of-the-art results on CheXbert metrics and competitive RadGraph F1 scores compared to MLLMs. Furthermore, RA-RRG naturally generalizes to multi-view RRG by aggregating phrases retrieved from multiple images, highlighting its broad applicability to real-world clinical scenarios. Code is available at https://github.com/deepnoid-ai/RA-RRG.

Elias Schuhmacher, Andrianos Michail, Juri Opitz, Rico Sennrich, Simon Clematide

To be discoverable in an embedding-based search process, each part of a document should be reflected in its embedding representation. To quantify any potential reflection biases, we introduce a permutation-based evaluation framework. With this, we observe that state-of-the-art embedding models exhibit systematic positional and language biases when documents are longer and consist of multiple segments. Specifically, early segments and segments in higher-resource languages like English are over-represented, while later segments and segments in lower-resource languages are marginalized. In our further analysis, we find that the positional bias stems from front-loaded attention distributions in pooling-token embeddings, where early tokens receive more attention. To mitigate this issue, we introduce an inference-time attention calibration method that redistributes attention more evenly across document positions, increasing discoverabiltiy of later segments. Our evaluation framework and attention calibration is available at https://github.com/impresso/fair-sentence-transformers

Yihong Liu, Junyi Li, Hongyu Lu, Xin Zhao, Ji-Rong Wen

Combinatorial optimization has long been dominated by manually engineered heuristics, a paradigm requiring substantial expert intuition and implementation overhead. The advent of Large Language Models has disrupted this landscape, enabling the autonomous synthesis and optimization of algorithms. Recent approaches typically iterate on heuristic populations using LLMs as mutators; however, these strategies often suffer from limited exploration, leading to stagnation in local optima. To overcome this, we present the Experience-Driven Reflective Co-Evolution of Prompt and Heuristics (EvoPH) for autonomous algorithm design, a novel framework that couples an island migration model with elite selection to maintain population diversity. Uniquely, EvoPH co-evolves both the guiding prompts and the heuristics themselves, using a feedback loop driven by past experience to refine the search process. We demonstrate EvoPH’s efficacy on the Traveling Salesman and Bin Packing Problems. Our results show that EvoPH achieves superior accuracy compared to baselines, marking a significant step forward in LLM-aided algorithm design.

Ziwen Pan, Zihan Liang, Jad Kabbara, Ali Emami

Large language models (LLMs) tuned for safety often avoid acknowledging demographic differences, even when such acknowledgment is factually correct (e.g., ancestry-based disease incidence) or contextually justified (e.g., religious hiring preferences). This *identity-blindness* yields incorrect responses, unnecessary refusals, or generic "equal-treatment" defaults. We study this via difference-awareness classification: given a question involving demographic groups, the task is not to answer directly, but to classify whether a correct answer requires recognizing group differences (**YES**) or whether groups should be treated identically (**NO**). Crucially, fine-tuning for accuracy triggers *harm drift*: model-generated explanations become increasingly harmful as decision accuracy improves, whether by elaborating harmful content, introducing problematic assumptions, or failing to flag harms the baseline identified. To mitigate this, we introduce **DART** (**D**istill–**A**udit–**R**epair **T**raining), which distills label-conditioned reasoning from a teacher, audits outputs for harm drift cases relative to baseline, and repairs problematic cases via severity-weighted fine-tuning. On eight benchmarks, DART improves Llama-3-8B-Instruct accuracy from 39.0% to 68.8%, with largest gains on equal-treatment prompts (11.3% → 72.6%), while reducing harm drift cases by 72.6%. It also transfers to 280 open-ended real-world queries across medical, legal, policy, and educational domains, improving difference-appropriate responses from 39.8% to 77.5% while reducing refusals from 34.3% to 3.0%. Our results demonstrate that accuracy and safety need not conflict when explicit detection and repair mechanisms are in place.

Dongliang Chen, Xinlin Zhuang, Junjie Xu, Luojian Xie, Zehui Wang, Jiaxi Zhuang, Haolin Yang, Liang Dou, Xiao He, Xingjiao Wu 等

Multi-objective alignment for text-to-image generation is commonly implemented via static linear scalarization, but fixed weights often fail under heterogeneous rewards, leading to optimization imbalance where models overfit high-variance, high-responsiveness objectives (e.g., OCR) while under-optimizing perceptual goals. We identify two mechanistic causes: variance hijacking, where reward dispersion induces implicit reweighting that dominates the normalized training signal, and gradient conflicts, where competing objectives produce opposing update directions and trigger seesaw-like oscillations. We propose APEX (Adaptive Priority-based Efficient X-objective Alignment), which stabilizes heterogeneous rewards with Dual-Stage Adaptive Normalization and dynamically schedules objectives via \mathcal{P}^3 Adaptive Priorities that combine learning potential, conflict penalty, and progress need. On Stable Diffusion 3.5, APEX achieves improved Pareto trade-offs across four heterogeneous objectives, with balanced gains of +1.31 PickScore, +0.35 DeQA, and +0.53 Aesthetics while maintaining competitive OCR accuracy, mitigating the instability of multi-objective alignment.

Maike Züfle, Ondrej Klejch, Nicholas Sanders, Jan Niehues, Alexandra Birch, Tsz Kin Lam

Spoken conversational systems require more than accurate speech generation to have human-like conversations: to feel natural and engaging, they must produce conversational behaviour that adapts dynamically to the context. Current spoken conversational systems, however, rarely allow such customization, limiting their naturalness and usability. In this work, we present the first open, instruction-following full-duplex conversational speech model that can be trained efficiently under typical academic resource constraints. By keeping the audio encoder frozen and finetuning only the language model, our model requires just 2,000 hours of data, without relying on large-scale pretraining or multi-stage optimization. The model can follow explicit instructions to control speaker voice, conversation topic, conversational behaviour (e.g., backchanneling and interruptions), and dialogue initiation. We propose a single-stage training protocol and systematically analyze design choices. Both the model and training code is released to enable reproducible research on controllable full-duplex speech systems.

Jinyang Du, Ruihao Gong, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, Wxuefei, Yaoyuan Wang, Jinyang Guo 等

Training large language models (LLMs) at 4-bit precision offers substantial efficiency gains but remains challenging due to the limited dynamic range and coarse numerical resolution. Existing 4-bit training pipelines typically rely on max-scaling, which is ill-suited for heavy-tailed LLM tensor distributions and leads to severe under-utilization of the FP4 quantization grid in the low-magnitude region. This effect causes pronounced representation collapse and large rounding errors for the values that dominate LLM computation. In this work, we derive the theoretically optimal scaling for FP4 under heavy-tailed inputs, revealing why max-scaling is intrinsically suboptimal. Guided by this analysis, we propose Half-S, a simple and efficient scaling strategy that uses half-scaling as a hardware-friendly default and falls back to an MSE-based clipping threshold when needed, yielding a close approximation to the theoretical optimum under real LLM statistics. Extensive experiments on large-scale pretraining and downstream fine-tuning show that Half-S consistently narrows the gap to BF16 in both convergence and final model quality, while preserving the efficiency benefits of 4-bit computation. Under native FP4 support, Half-S is estimated to provide up to 1.8\times end-to-end training speedup. These results indicate that Half-S provides a simple and effective correction to max-scaling, substantially improving the stability and accuracy of 4-bit LLM training.

Wenpeng Xing, Zhonghao Qi, Yupeng Qin, Yilin Li, Caini Chang, Jiahui Yu, Changting Lin, Zhenzhen Xie, Meng Han

While Large Language Models (LLMs) have achieved remarkable performance, they remain vulnerable to jailbreak. The integration of Large Language Models (LLMs) with external tools via protocols such as the Model Context Protocol (MCP) introduces critical security vulnerabilities, including prompt injection, data exfiltration, and other threats. To counter these challenges, we propose MCP-Guard, a robust, layered defense architecture designed for LLM–tool interactions. MCP-Guard employs a three-stage detection pipeline that balances efficiency with accuracy: it progresses from lightweight static scanning for overt threats and a deep neural detector for semantic attacks, to our fine-tuned E5-based model achieves 96.01% accuracy in identifying adversarial prompts. Finally, an LLM arbitrator synthesizes these signals to deliver the final decision. To enable rigorous training and evaluation, we introduce MCP-AttackBench, a comprehensive benchmark comprising 70,448 samples augmented by GPT-4. This benchmark simulates diverse real-world attack vectors that circumvent conventional defenses in the MCP paradigm, thereby laying a solid foundation for future research on securing LLM-tool ecosystems.

Feihu Jin, Ying Tan

Fine-tuning large language models (LLMs) using standard first-order (FO) optimization oftendrives training toward sharp, poorly generalizing minima. Conversely, zeroth-order (ZO) methods offer stronger exploratory behaviorwithout relying on explicit gradients, yet suffer from slow convergence. More critically, our analysis reveals that in generative tasks, the vast output and search space significantly amplify estimation variance, rendering ZO methods both noisy and inefficient. To address these challenges, we propose Hi-ZFO (Hierarchical Zeroth- and First-Order optimization), a hybrid framework designed to synergize the precision of FO gradients with the exploratory capability of ZO estimation. Hi-ZFO adaptively partitions the model through layer-wise importance profiling, applying precise FO updates to critical layers while leveraging ZO optimization for less sensitive ones. Notably, ZO in Hi-ZFO is not merely a memory-saving surrogate; it is intentionally introduced as a source of "beneficial stochasticity" to help the model escape the local minima where pure FO optimization tends to stagnate. Validated across diverse generative, mathematical, and code reasoning tasks, Hi-ZFO consistently achieves superior performance while significantly reducing the training time. These results demonstrate the effectiveness of hierarchical hybrid optimization for LLM fine-tuning.

Kun Huang, Rui Qiu, Xiaoming Li, Salah Uddin

Recent advances in Large Vision–language Models (VLMs) suggest their potential for multimodal misinformation detection. However, existing multimodal misinformation detectors often fail to effectively integrate them, relying instead on passive aggregation of multimodal features and social signals. Such correlation-driven paradigms are vulnerable to spurious associations and multimodal noise, and lack explicit verification mechanisms. In this paper, we propose Logic-Guided Adaptive Reasoning (LoGAR), a verification-oriented framework that integrates VLMs into multimodal misinformation detection through explicit rationale-guided reasoning. LoGAR leverages a VLM to generate an explicit verification rationale, which serves as a global semantic anchor to condition the entire reasoning process. Concretely, the rationale functions as an active query to guide multimodal feature fusion and as a conditioning signal to modulate message passing over heterogeneous social graphs, enabling hypothesis-aware evidence aggregation. Furthermore, LoGAR introduces an instance-aware adaptive depth mechanism that dynamically determines the required reasoning depth. Experimental results on multiple multimodal misinformation benchmarks demonstrate that LoGAR consistently outperforms state-of-the-art methods while significantly reducing computational cost.