论文检索

Yuebin Xu, Zhiyi Chen, Zeyi Wen

Tuning inference hyperparameters, such as temperature and maximum output tokens, on downstream tasks can enhance inference performance. However, directly applying hyperparameter optimization to these hyperparameters is token-expensive. Multi-fidelity optimization improves HPO efficiency with low-fidelity evaluations, but its static scheduling strategies ignore token consumption, leading to high costs. To address these limitations, we propose a token-efficient multi-fidelity optimization method, which enhances inference performance and minimizes token usage. Our method is empowered by (i) a token-based fidelity definition with explicit token cost modeling on configurations; (ii) a novel Token-Aware Expected Improvement acquisition function that selects configurations based on performance gain per token; and (iii) a dynamic fidelity scheduling mechanism that adapts to real-time budget status. We evaluate our method on LLaMA-2 and LLaMA-3 series across MMLU, Humaneval, MedQA, and OpenBookQA. Our method improves over the HELM leaderboard by 7.1%, 24.3%, 21.9%, and 4.6%, respectively. Compared to existing multi-fidelity HPO baselines, our method reduces token consumption by over 80% while maintaining or surpassing performance, demonstrating the state-of-the-art token efficiency for inference-time optimization.

Yunuo Liu, Dawei Zhu, Zena Al-Khalili, Dai Cheng, Yanjun Chen, Dietrich Klakow, Wei Zhang, Xiaoyu Shen

We present PricingLogic, the first benchmarkthat probes whether Large Language Mod-els (LLMs) can reliably automate tourism-booking prices when multiple, overlapping farerules apply. Travel agencies are eager to of-fload this error-prone task to AI systems; how-ever, deploying LLMs without verified reliabil-ity could result in significant financial lossesand erode customer trust. PricingLogic com-prises 300 natural-language questions based onbooking requests derived from 42 real-worldpricing policies, spanning two levels of diffi-culty: (i) basic customer-type pricing and (ii)bundled-tour calculations involving interactingdiscounts. Evaluations of a line of LLMs re-veal a steep performance drop on the harder tier,exposing systematic failures in rule interpreta-tion and arithmetic reasoning. These resultshighlight that, despite their general capabilities,today’s LLMs remain unreliable for revenue-critical applications without further safeguardsor domain adaptation. Our code and dataset areavaliable in https://github.com/EIT-NLP/PricingLogic.

Johannes Moll, Louisa Fay, Asfandyar Azhar, Sophie Ostmeier, Sergios Gatidis, Tim Lueth, Curtis P. Langlotz, Jean-Benoit Delbrouck

Radiology reports are critical for clinical decision-making but often lack a standardized format, limiting both human interpretability and machine learning (ML) applications. While large language models (LLMs) have shown strong capabilities in reformatting clinical text, their high computational requirements, lack of transparency, and data privacy concerns hinder practical deployment. To address these challenges, we explore lightweight encoder-decoder models (<300M parameters)—specifically T5 and BERT2BERT—for structuring radiology reports from the MIMIC-CXR and CheXpert Plus datasets. We benchmark these models against eight open-source LLMs (1B–70B parameters), adapted using prefix prompting, in-context learning (ICL), and low-rank adaptation (LoRA) finetuning. Our best-performing lightweight model outperforms all LLMs adapted using prompt-based techniques on a human-annotated test set. While some LoRA-finetuned LLMs achieve modest gains over the lightweight model on the Findings section (BLEU 6.4%, ROUGE-L 4.8%, BERTScore 3.6%, F1-RadGraph 1.1%, GREEN 3.6%, and F1-SRR-BERT 4.3%), these improvements come at the cost of substantially greater computational resources. For example, LLaMA-3-70B incurred more than 400 times the inference time, cost, and carbon emissions compared to the lightweight model. These results underscore the potential of lightweight, task-specific models as sustainable and privacy-preserving solutions for structuring clinical text in resource-constrained healthcare settings.

Yuanjun Feng, Vivek Choudhary, Yash Raj Shrestha

Large language models (LLMs) are increasingly used for social-science simulations, yet most evaluations target task optimality rather than the variability and adaptation characteristic of human decision-making. We propose a process-oriented evaluation framework with progressive interventions (Intrinsicality, Instruction, and Imitation), and apply it to two classic economics tasks: the second-price auction and the newsvendor inventory problem.By default, LLMs adopt stable, conservative strategies that diverge from observed human behavior. Giving LLMs risk-framed instructions makes them behave more like humans. However, this also causes complex irregularities. Incorporating human decision trajectories via in-context learning further narrows distributional gaps, indicating that models can absorb human patterns. However, across all interventions, LLMs underexpress round-to-round variability relative to humans, revealing a persistent alignment gap in behavioral fidelity. Future evaluations of LLM-based social simulations should prioritize process-level realism.

Songze Li, Zhiqiang Liu, Zhengke Gui, Huajun Chen, Wen Zhang

Large Language Models (LLMs) exhibit strong reasoning capabilities in complex tasks. However, they still struggle with hallucinations and factual errors in knowledge-intensive scenarios like knowledge graph question answering (KGQA). We attribute this to the semantic gap between structured knowledge graphs (KGs) and unstructured queries, caused by inherent differences in their focuses and structures. Existing methods usually employ resource-intensive, non-scalable workflows reasoning on vanilla KGs, but overlook this gap. To address this challenge, we propose a flexible framework, Enrich-on-Graph (EoG), which leverages LLMs’ prior knowledge to enrich KGs, bridge the semantic gap between graphs and queries. EoG enables efficient evidence extraction from KGs for precise and robust reasoning, while ensuring low computational costs, scalability, and adaptability across different methods. Furthermore, we propose three graph quality evaluation metrics to analyze query-graph alignment in KGQA task, supported by theoretical validation of our optimization objectives. Extensive experiments on two KGQA benchmark datasets indicate that EoG can effectively generate high-quality KGs and achieve the state-of-the-art performance.

Yi Sun, Han Wang, Jiaqiang Li, Jiacheng Liu, Xiangyu Li, Hao Wen, Yizhen Yuan, Huiwen Zheng, Yan Liang, Yuanchun Li 等

Recent work has demonstrated the remarkable potential of Large Language Models (LLMs) in test-time scaling. By making models think before answering, they are able to achieve much higher accuracy with extra inference computation.However, in many real-world scenarios, models are used under time constraints, where an answer should be given within a certain output length. It is unclear whether and how the reasoning ability of different LLMs remain effective under strict constraints.We take a first look at this problem by conducting an in-depth empirical study. Specifically, we test 30 LLMs on common reasoning datasets under a wide range of output length budgets, and we analyze the correlation between the inference accuracy and various properties including model type, model size, prompt style, etc. We also consider the mappings between token budgets and actual on-device latency budgets.The results have demonstrated several interesting findings regarding the budget-aware LLM reasoning ability that differ from the unconstrained situation, e.g. the optimal choices of either model size or prompt style change under different budgets. These findings offer timely evaluation to this area and practical guidance for users to deploy LLMs under real-world latency constraints.

Jielong Tang, Yang Yang, Jianxing Yu, Zhen-Xing Wang, Haoyuan Liang, Liang Yao, Jian Yin

Grounded Multimodal Named Entity Recognition (GMNER) is a new information extraction task. It requires models to extract named entities and ground them to real-world visual objects. Previous methods, relying on domain-specific fine-tuning, struggle with unseen multimodal entities due to limited knowledge and generalization. Recently, multimodal large language models (MLLMs) have demonstrated strong open-set abilities. However, their performance is hindered by the lack of in-domain knowledge due to costly training for GMNER datasets. To address these limitations, we propose **UnCo**, a two-stage Uncertainty-driven Collaborative framework that leverages the complementary strengths of small fine-tuned models and MLLMs. Specifically, **in stage one**, we equip the small model with a unified uncertainty estimation (UE) for multimodal entities. This enables the small model to express \textit{"I do not know"} when recognizing unseen entities beyond its capabilities. Predictions with high uncertainty are then filtered and delegated to the MLLM. **In stage two**, an Uncertainty-aware Hierarchical Correction mechanism guides the MLLM to refine uncertain predictions using its open-domain knowledge. Ultimately, UnCo effectively retains the in-domain knowledge of small models while utilizing the capabilities of MLLMs to handle unseen samples. Extensive experiments demonstrate UnCo’s effectiveness on two GMNER benchmarks.

Anum Afzal, Florian Matthes, Alexander Fabbri

Large Language Models (LLMs) often don’t perform as expected under Domain Shift or after Instruct-tuning. A reliable indicator of LLM performance in these settings could assist in decision-making. We present a method that uses the known performance in high-resource domains and fine-tuning settings to predict performance in low-resource domains or base models, respectively. In our paper, we formulate the task of performance prediction, construct a dataset for it, and train regression models to predict the said change in performance. Our proposed methodology is lightweight and, in practice, can help researchers & practitioners decide if resources should be allocated for data labeling and LLM Instruct-tuning.

Kun Li, Lai Man Po, Hongzheng Yang, Xuyuan Xu, Kangcheng Liu, Yuzhi Zhao

Multimodal Large Language Models (MLLMs) are increasingly applied in Personalized Image Aesthetic Assessment (PIAA) as a scalable alternative to expert evaluations. However, their predictions may reflect subtle biases influenced by demographic factors such as gender, age, and education. In this work, we propose AesBiasBench, a benchmark designed to evaluate MLLMs along two complementary dimensions: (1) stereotype bias, quantified by measuring variations in aesthetic evaluations across demographic groups; and (2) alignment between model outputs and genuine human aesthetic preferences. Our benchmark covers three subtasks (Aesthetic Perception, Assessment, Empathy) and introduces structured metrics (IFD, NRD, AAS) to assess both bias and alignment. We evaluate 19 MLLMs, including proprietary models (e.g., GPT-4o, Claude-3.5-Sonnet) and open-source models (e.g., InternVL-2.5, Qwen2.5-VL). Results indicate that smaller models exhibit stronger stereotype biases, whereas larger models align more closely with human preferences. Incorporating identity information often exacerbates bias, particularly in emotional judgments. These findings underscore the importance of identity-aware evaluation frameworks in subjective vision-language tasks.

Minh-Phuc Truong, Hai An Vu, Tu Vu, Nguyen Thi Ngoc Diep, Linh Ngo Van, Thien Huu Nguyen, Trung Le

Knowledge distillation (KD) is crucial for compressing large text embedding models, but faces challenges when teacher and student models use different tokenizers (Cross-Tokenizer KD - CTKD). Vocabulary mismatches impede the transfer of relational knowledge encoded in deep representations, such as hidden states and attention matrices, which are vital for producing high-quality embeddings. Existing CTKD methods often focus on direct output alignment, neglecting this crucial structural information. We propose a novel framework tailored for CTKD embedding model distillation. We first map tokens one-to-one via Minimum Edit Distance (MinED). Then, we distill intra-model relational knowledge by aligning attention matrix patterns using Centered Kernel Alignment, focusing on the top-m most important tokens of the directly mapped tokens. Simultaneously, we align final hidden states via Optimal Transport with Importance-Scored Mass Assignment, which emphasizes semantically important token representations, based on importance scores derived from attention weights. We evaluate distillation from state-of-the-art embedding models (e.g., LLM2Vec, BGE) to a Bert-base-uncased model on embedding-reliant tasks such as text classification, sentence pair classification, and semantic textual similarity. Our proposed framework significantly outperforms existing CTKD baselines. By preserving attention structure and prioritizing key representations, our approach yields smaller, high-fidelity embedding models despite tokenizer differences.

Yuxin Gou, Xiaoning Dong, Qin Li, Shishen Gu, Richang Hong, Wenbo Hu

Multimodal large language models (MLLMs) demonstrate impressive capabilities by integrating visual and textual information. However, the incorporation of visual modalities also introduces new and complex safety risks, rendering even the most advanced models vulnerable to sophisticated jailbreak attacks. This paper first analyzes the impact of inserting safety reasoning prompt on various aspects of the model. We find that this external method can help the model resist jailbreak attacks to some extent, but the model still fails to distinguish specific semantic scenarios, resulting in a significantly increased refusal rate for benign queries. Inspired by this, we propose a novel training framework, SURE (Safety Understanding and Reasoning Enhancement for Multimodal Large Language Models), designed to help models internalize chain-of-thought-based safety decision-making capabilities. Extensive experiments demonstrate that SURE significantly improves model safety while effectively avoiding over-defense, achieving a good balance between safety and generality. Finally, we create a large-scale multimodal safety reasoning dataset, MLLM-SCoT-Plus, to facilitate research on safety alignment in multimodal models.Our code and the dataset are publicly available at https://github.com/hfutml/SURE.

Heng-Da Xu, Xian-Ling Mao, Fanshu Sun, Tian-Yi Che, Cheng-Xin Xin, Heyan Huang

High-quality datasets are essential for building effective task-oriented dialogue (TOD) systems. The existing TOD datasets often present overly simplified interactions, where users incrementally express straightforward requests that can be managed with basic slot-value style dialogue states, such as “hotel-area = east.” However, this approach does not reflect real-life scenarios in which users may express complex constraints and preferences. To address this gap, in this paper, we propose SQLWOZ, a novel TOD dataset designed to capture complex, real-world user requirements. The user requirements in SQLWOZ include the four categories: 1) multiple values for a slot, 2) excluded values within a slot, 3) preferred or prioritized values, and 4) conditional values based on other conditions. We utilize SQL statements as a formalized and expressive representation of dialogue states within SQLWOZ. To evaluate the dataset, we adapt large language models as dialogue agents and conduct extensive experiments on the SQL-based dialogue state tracking, dialogue response generation and end-to-end TOD tasks. The experimental results demonstrate the complexity and quality of SQLWOZ, establishing it as a new benchmark for advancing TOD research.

Weiqiu You, Anton Xue, Shreya Havaldar, Delip Rao, Helen Jin, Chris Callison-Burch, Eric Wong

In reasoning chains generated by large language models (LLMs), initial errors often propagate and undermine the reliability of the final conclusion. Current LLM-based error detection methods often fail to detect propagated errors because earlier errors can corrupt judgments of downstream reasoning. To better detect such errors, we introduce Autoregressive Reasoning Entailment Stability (ARES), a probabilistic framework that evaluates each reasoning step based solely on previously-verified premises. This inductive method yields a nuanced score for each step and provides certified statistical guarantees of its soundness, rather than a brittle binary label. ARES achieves state-of-the-art performance across four benchmarks (72.1% Macro-F1, +8.2 points) and demonstrates superior robustness on very long synthetic reasoning chains, where it excels at detecting propagated errors (90.3% F1, +27.6 points).

Kyeongman Park, Nakyeong Yang, Kyomin Jung

Large Language Models (LLMs) often generate repetitive and monotonous outputs, especially in tasks like story generation, due to limited creative diversity when given the same input prompt. To address this challenge, we propose a novel decoding strategy, ***Avoidance Decoding***, that modifies token logits by penalizing similarity to previously generated outputs, thereby encouraging more diverse multi-branch stories. This penalty adaptively balances two similarity measures: (1) Concept-level Similarity Penalty, which is prioritized in early stages to diversify initial story concepts, and (2) Narrative-level Similarity Penalty, which is increasingly emphasized later to ensure natural yet diverse plot development. Notably, our method achieves up to **2.6** times higher output diversity and reduces repetition by an average of 30% compared to strong baselines, while effectively mitigating text degeneration. Furthermore, we reveal that our method activates a broader range of neurons, demonstrating that it leverages the model’s intrinsic creative capacity.

Aziguli Wulamu, Kaiyuan Gong, Lyu Zhengyu, Yu Han, Zhihong Zhu, Bowen Xing

Relation Triplet Extraction (RTE) is a fundamental while challenge task in knowledge acquisition, which identifies and extracts all triplets from unstructured text. Despite the recent advancements, the deep integration of the entity-, relation- and triplet-specific information remains a challenge. In this paper, we propose a Graph-based Mixture-of-Experts mutual learning framework for RTE, namely RTE-GMoE, to address this limitation. As a model-agnostic framework, RTE-GMoE distinguishes itself by including and modeling the mutual interactions among three vital task-specific experts: entity expert, RTE expert, and relation expert. RTE expert corresponds to the main RTE task and can be implemented by any model and the other two correspond to the two auxiliary tasks: entity recognition and relation extraction. We construct an expert graph and achieve comprehensive and adaptive graph-based MoE interactions with a novel mutual learning mechanism. In our framework, these experts perform knowledge extractions collaboratively via dynamic information exchange and knowledge sharing. We conduct extensive experiments on four state-of-the-art backbones and evaluate them on several widely-used benchmarks. The results demonstrate that our framework brings consistent and promising improvements on all backbones and benchmarks. Component study and model analysis further verify the effectiveness and advantages of our method.

Lena Sophia Bolliger, Lena Ann Jäger

The type of a text profoundly shapes reading behavior, yet little is known about how different text types interact with word-level features and the properties of machine-generated texts and how these interactions influence how readers process language. In this study, we investigate how different text types affect eye movements during reading, how neural decoding strategies used to generate texts interact with text type, and how text types modulate the influence of word-level psycholinguistic features such as surprisal, word length, and lexical frequency. Leveraging EMTeC (Bolliger et al., 2025), the first eye-tracking corpus of LLM-generated texts across six text types and multiple decoding algorithms, we show that text type strongly modulates cognitive effort during reading, that psycholinguistic effects induced by word-level features vary systematically across genres, and that decoding strategies interact with text types to shape reading behavior. These findings offer insights into genre-specific cognitive processing and have implications for the human-centric design of AI-generated texts. Our code is publicly available at https://github.com/DiLi-Lab/Genre-Matters.

Wei Liu, Siya Qi, Xinyu Wang, Chen Qian, Yali Du, Yulan He

Recent advances, such as DeepSeek R1-Zero, highlight the effectiveness of incentive training, a reinforcement learning paradigm that computes rewards solely based on the final answer part of a language model’s output, thereby encouraging the generation of intermediate reasoning steps. However, these methods fundamentally rely on external verifiers, which limits their applicability to domains like mathematics and coding, where such verifiers are readily available. Although reward models can serve as verifiers, they require high-quality annotated data and are costly to train.In this work, we propose \textbf{NOVER}, \textbf{\textit{NO-VER}}ifier Reinforcement Learning, a general reinforcement learning framework that requires only standard supervised fine-tuning data with no need for an external verifier. NOVER enables incentive training across a wide range of text-to-text tasks and outperforms the model of the same size distilled from large reasoning models such as DeepSeek R1 671B by 7.7%. Moreover, the flexibility of NOVER enables new possibilities for optimizing large language models, such as inverse incentive training.

Wenrui Cai, Chengyu Wang, Junbing Yan, Jun Huang, Xiangzhong Fang

The reasoning capabilities of large language reasoning models (LRMs), such as OpenAI’s o1 and DeepSeek-R1, have seen substantial advancements through deep thinking. However, these enhancements come with significant resource demands, underscoring the need for training effective small reasoning models. A critical challenge is that small models possess different reasoning capacities and cognitive trajectories compared with their larger counterparts. Hence, directly distilling chain-of-thought (CoT) results from large LRMs to smaller ones can sometimes be ineffective and often requires a substantial amount of annotated data. In this paper, we first introduce a novel Critique-Rethink-Verify (CRV) system, designed for training smaller yet powerful LRMs. Our CRV system consists of multiple LLM agents, each specializing in unique abilities: (i) critiquing the CoT qualities according to the cognitive capabilities of smaller models, (ii) rethinking and refining these CoTs based on the critiques, and (iii) verifying the correctness of the refined results. Based on the CRV system, we further propose the Cognitive Preference Optimization (CogPO) algorithm to continuously enhance the reasoning abilities of smaller models by aligning their reasoning processes with their cognitive capacities. Comprehensive evaluations on challenging reasoning benchmarks demonstrate the efficacy of our CRV+CogPO framework, which outperforms other methods by a large margin.

Wangyi Jiang, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

Large Reasoning Models (LRMs) show strong System-2-style reasoning, but at the cost of significant computational overhead. In contrast, efficient System-1-style Large Language Models (LLMs) often struggle on complex tasks. We identify a critical asymmetry between these two paradigms: LRMs can implicitly self-distill their own reasoning, solving hard problems with near System-1-style efficiency while retaining superior performance. LLMs, however, lack such deep internal modes and collapse when forced to rely on their own reasoning rather than imitating external traces. This asymmetry explains why direct distillation from strong LRMs to weaker LLMs often fails: student models struggle to learn from LRMs’ overly complex explicit reasoning and gain little from their overly compact implicit solutions. To address this, we introduce a two-stage curriculum distillation framework, which first builds a robust internal problem-solving student model and then teaches the student model to externalize this latent knowledge as explicit reasoning. On challenging mathematical benchmarks, our method significantly outperforms single-stage baselines, creating compact models with strong reasoning ability.

Shuzheng Si, Haozhe Zhao, Gang Chen, Yunshui Li, Kangyang Luo, Chuancheng Lv, Kaikai An, Fanchao Qi, Baobao Chang, Maosong Sun

Aligning large language models to handle instructions with extremely long contexts has yet to be fully investigated. Previous studies have attempted to scale up the available data volume by synthesizing long instruction-following samples, as constructing such a dataset tends to be challenging for annotators. However, a lack of a well-defined strategy for ensuring data quality may introduce low-quality samples and restrict the model’s performance. Thus, we propose GATEAU, a novel framework to address the unique challenge of long context alignment by identifying the influential samples enriched with long-range dependency relations. Specifically, GATEAU measures the long-range dependencies from two essential aspects: the difficulty of generating target responses due to the long-range dependencies, and the difficulty of understanding long inputs due to such dependencies. Comprehensive experiments indicate that GATEAU effectively identifies influential samples and the model trained on these selected samples exhibits better instruction-following and long-context understanding capabilities.