论文检索

Minglai Yang, Ethan Huang, Liang Zhang, Mihai Surdeanu, William Yang Wang, Liangming Pan

We introduce Grade School Math with Distracting Context (GSM-DC), a synthetic benchmark to evaluate Large Language Models’ (LLMs) reasoning robustness against systematically controlled irrelevant context (IC). GSM-DC constructs symbolic reasoning graphs with precise distractor injections, enabling rigorous, reproducible evaluation. Our experiments demonstrate that LLMs are significantly sensitive to IC, affecting both reasoning path selection and arithmetic accuracy. Additionally, training models with strong distractors improves performance in both in-distribution and out-of-distribution scenarios. We further propose a stepwise tree search guided by a process reward model, which notably enhances robustness in out-of-distribution conditions.

Jintian Zhang, Yuqi Zhu, Mengshu Sun, Yujie Luo, Shuofei Qiao, Lun Du, Da Zheng, Huajun Chen, Ningyu Zhang

Large language models (LLMs) have shown remarkable performance in complex reasoning tasks, but their efficiency is hindered by the substantial memory and computational costs associated with generating lengthy tokens. In this paper, we propose LightThinker, a novel method that enables LLMs to dynamically compress intermediate thoughts during reasoning. Inspired by human cognitive processes, LightThinker compresses verbose thought steps into compact representations and discards the original reasoning chains, thereby significantly reducing the number of tokens stored in the context window.This is achieved by training the model on when and how to perform compression through data construction, mapping hidden states to condensed gist tokens, and creating specialized attention masks. Additionally, we introduce the Dependency (Dep) metric to quantify the degree of compression by measuring the reliance on historical tokens during generation. Extensive experiments on four datasets and two models show that LightThinker reduces peak memory usage and inference time, while maintaining competitive accuracy. Our work provides a new direction for improving the efficiency of LLMs in complex reasoning tasks without sacrificing performance.

Yejin Yoon, Yuri Son, Namyoung So, Minseo Kim, Minsoo Cho, Chanhee Park, Seungshin Lee, Taeuk Kim

Conversational agents have traditionally been developed for either task-oriented dialogue (TOD) or open-ended chitchat, with limited progress in unifying the two. Yet, real-world conversations naturally involve fluid transitions between these modes. To address this gap, we introduce TACT (TOD-And-Chitchat Transition), a dataset designed for transition-aware dialogue modeling that incorporates structurally diverse and integrated mode flows. TACT supports both user- and agent-driven mode switches, enabling robust modeling of complex conversational dynamics.To evaluate an agent’s ability to initiate and recover from mode transitions, we propose two new metrics—Switch and Recovery.Models trained on TACT outperform baselines in both intent detection and mode transition handling. Moreover, applying Direct Preference Optimization (DPO) to TACT-trained models yields additionalgains, achieving 75.74% joint mode-intent accuracy and a 70.1% win rate against GPT-4o in human evaluation.These results demonstrate that pairing structurally diverse data with DPO enhances response quality and transition control, paving the way for more proactive and transition-aware conversational agents.

Jiyuan Liu, Jielin Song, Yunhe Pang, Zhiyu Shen, Yanghui Rao

Disagreement detection is a crucial task in natural language processing (NLP), particularly in analyzing online discussions and social media content. Large language models (LLMs) have demonstrated significant advancements across various NLP tasks. However, the performance of LLM in disagreement detection is limited by two issues: *conceptual gap* and *reasoning gap*. In this paper, we propose a novel two-stage framework, Concept Alignment and Reasoning Enhancement (CARE), to tackle the issues. The first stage, Concept Alignment, addresses the gap between expert and model by performing **sub-concept taxonomy extraction**, aligning the model’s comprehension with human experts. The second stage, Reasoning Enhancement, improves the model’s reasoning capabilities by introducing curriculum learning workflow, which includes **rationale to critique** and **counterfactual to detection** for reducing spurious association. Extensive experiments on disagreement detection task demonstrate the effectiveness of our framework, showing superior performance in zero-shot and supervised learning settings, both within and across domains.

Jiyuan Liu, Jiaxing Yan, Chunjiang Zhu, Xingyu Liu, Li Qing, Yanghui Rao

Topic modeling is a powerful unsupervised tool for knowledge discovery. However, existing work struggles with generating limited-quality topics that are uninformative and incoherent, which hindering interpretable insights from managing textual data. In this paper, we improve the original variational autoencoder framework by incorporating contextual and graph information to address the above issues. First, the encoder utilizes topic fusion techniques to combine contextual and bag-of-words information well, and meanwhile exploits the constraints of topic alignment and topic sharpening to generate informative topics. Second, we develop a simple word co-occurrence graph information fusion strategy that efficiently increases topic coherence. On three benchmark datasets, our new framework generates more coherent and diverse topics compared to various baselines, and achieves strong performance on both automatic and manual evaluations.

Brendan Murphy, Dillon Bowen, Shahrad Mohammadzadeh, Tom Tseng, Julius Broomfield, Adam Gleave, Kellin Pelrine

AI systems are rapidly advancing in capability, and frontier model developers broadly acknowledge the need for safeguards against serious misuse. However, this paper demonstrates that fine-tuning, whether via open weights or closed fine-tuning APIs, can produce helpful-only models with safeguards destroyed. In contrast to prior work which is blocked by modern moderation systems or achieved only partial removal of safeguards or degraded output quality, our jailbreak-tuning method teaches models to generate detailed, high-quality responses to arbitrary harmful requests. For example, OpenAI, Google, and Anthropic models will fully comply with requests for CBRN assistance, executing cyberattacks, and other criminal activity. We further show that backdoors can increase not only the stealth but also the severity of attacks. Stronger jailbreak prompts become even more effective in fine-tuning attacks, linking attacks and potentially defenses in the input and weight spaces. Not only are current models vulnerable, more recent ones also appear to be becoming even more vulnerable to these attacks, underscoring the urgent need for tamper-resistant safeguards. Until such safeguards are discovered, companies and policymakers should view the release of any fine-tunable model as simultaneously releasing its evil twin: equally capable as the original model, and usable for any malicious purpose within its capabilities.

Hongbo Zhang, Han Cui, Guangsheng Bao, Linyi Yang, Jun Wang, Yue Zhang

We introduce Direct Value Optimization (DVO), an innovative offline reinforcement learning framework for enhancing large language models in complex reasoning tasks. Unlike traditional methods relying on preference labels, DVO utilizes value signals at individual reasoning steps, optimizing models via a mean squared error loss. The key benefit of DVO lies in its fine-grained supervision, circumventing the need for labor-intensive human annotations. Target values within the DVO are estimated using either Monte Carlo Tree Search or an outcome value model. Our empirical analysis on 3 math reasoning, 4 commonsense reasoning, and 3 coding tasks shows that DVO consistently outperforms existing offline preference optimization techniques by a significant margin of 4% to 6%, and is competitive to online GRPO but with higher sample efficiency. These findings underscore the importance of value signals in advancing reasoning capabilities and highlight DVO as a superior methodology under scenarios lacking explicit human preference information.

Yunxiao Shi, Haoning Shang, Xing Zi, Wujiang Xu, Yue Feng, Min Xu

Narrative-driven recommendation queries are common in question-answering platforms, AI search engines, social forums, and some domain-specific vertical applications. Users typically submit free-form text requests for recommendations, e.g., “Any mind-bending thrillers like Shutter Island you’d recommend?” Such special queries have traditionally been addressed as generic QA task under the RAG paradigm. This work formally introduces narrative recommendation as a distinct task and contends that the RAG paradigm is inherently ill-suited for it, owing to information loss in LLMs when retrieving information from from multiple long and fragmented contexts, and limitations in ranking effectiveness. To overcome these limitations, we propose a novel retrieve-rank paradigm by theoretically demonstrating its superiority over RAG paradigm. Central to this new paradigm, we specially focus on the information retrieval stage and introduce Open-domain Candidate Generation (OCG)-Agent that generatively retrieves structurally adaptive and semantically aligned candidates, ensuring both extensive candidate coverage and high-quality information. We validate effectiveness of new paradigm and OCG-Agent’s retrieve mechanism under real-world datasets from Reddit and corporate education-consulting scenarios. Further extensive ablation studies confirming the rationality of each OCG-Agent component.

Chunyang Jiang, Chi-Min Chan, Yiyang Cai, Yulong Liu, Wei Xue, Yike Guo

Recently, the pretrain-finetune paradigm has become a cornerstone in various deep learning areas. While in general the pre-trained model would promote both effectiveness and efficiency of downstream tasks fine-tuning, studies have shown that not all knowledge acquired during pre-training is beneficial. Some of the knowledge may actually bring detrimental effects to the fine-tuning tasks, which is also known as negative transfer. To address this problem, graceful forgetting has emerged as a promising approach. The core principle of graceful forgetting is to enhance the learning plasticity of the target task by selectively discarding irrelevant knowledge. However, this approach remains underexplored in the context of generative language models, and it is often challenging to migrate existing forgetting algorithms to these models due to architecture incompatibility. To bridge this gap, in this paper we propose a novel framework, Learning With Forgetting (LWF), to achieve graceful forgetting in generative language models. With Fisher Information Matrix weighting the intended parameter updates, LWF computes forgetting confidence to evaluate self-generated knowledge regarding the forgetting task, and consequently, knowledge with high confidence is periodically unlearned during fine-tuning. Our experiments demonstrate that, although thoroughly uncovering the mechanisms of knowledge interaction remains challenging in pre-trained language models, applying graceful forgetting can contribute to enhanced fine-tuning performance.

Dingkun Zhang, Shuhan Qi, Xinyu Xiao, Kehai Chen, Xuan Wang

Recent advances in Multimodal Large Language Models (MLLMs) have enhanced their versatility as they integrate a growing number of modalities. Considering the heavy cost of training MLLMs, it is efficient to reuse the existing ones and extend them to more modalities through Modality-incremental Continual Learning (MCL). The exploration of MCL is in its early stages. In this work, we dive into the causes of performance degradation in MCL. We uncover that it suffers not only from forgetting as in traditional continual learning, but also from misalignment between the modality-agnostic and modality-specific components. To this end, we propose an elegantly simple MCL paradigm called “MErge then ReAlign” (MERA) to address both forgetting and misalignment. MERA avoids introducing heavy model budgets or modifying model architectures, hence is easy to deploy and highly reusable in the MLLM community. Extensive experiments demonstrate the impressive performance of MERA, holding an average of 99.84% Backward Relative Gain when extending to four modalities, achieving nearly lossless MCL performance. Our findings underscore the misalignment issue in MCL. More broadly, our work showcases how to adjust different components of MLLMs during continual learning.

Canmiao Zhou, Han Huang

Symbolic regression is a powerful technique for discovering mathematical expressions that best fit observed data. While neural symbolic regression methods based on large-scale pre-trained models perform well on simple tasks, the reliance on fixed parametric knowledge typically limits their generalization to complex and diverse data distributions. To address this challenge, we propose a syntax-aware retrieval-augmented mechanism that leverages the syntactic structure of symbolic expressions to perform context-aware retrieval from a pre-constructed token datastore during inference. This mechanism enables the model to incorporate highly relevant non-parametric prior information to assist in expression generation. Additionally, we design an entropy-based confidence network that dynamically adjusts the fusion strength between neural and retrieved components by estimating predictive uncertainty. Extensive experiments on multiple symbolic regression benchmarks demonstrate that the proposed method significantly outperforms representative baselines, validating the effectiveness of retrieval augmentation in enhancing the generalization performance of neural symbolic regression models.

Ding-Chu Zhang, Yida Zhao, Jialong Wu, Liwen Zhang, Baixuan Li, Wenbiao Yin, Yong Jiang, Yu-Feng Li, Kewei Tu, Pengjun Xie 等

The rapid advancement of large language models (LLMs) has transformed the landscape of agentic information seeking capabilities through the integration of tools such as search engines and web browsers. However, current mainstream approaches for enabling LLM web search proficiency face significant challenges: supervised fine-tuning struggles with data production in open-search domains, while RL converges quickly, limiting their data utilization efficiency. To address these issues, we propose EvolveSearch, a novel iterative self-evolution framework that combines SFT and RL to enhance agentic web search capabilities without any external human-annotated reasoning data. Extensive experiments on seven multi-hop question-answering (MHQA) benchmarks demonstrate that EvolveSearch consistently improves performance across iterations, ultimately achieving an average improvement of 4.7% over the current state-of-the-art across seven benchmarks, opening the door to self-evolution agentic capabilities in open web search domains.

Bangde Du, Ziyi Ye, Zhijing Wu, Monika A. Jankowska, Shuqi Zhu, Qingyao Ai, Yujia Zhou, Yiqun Liu

As Large Language Models (LLMs) demonstrate increasingly strong human-like capabilities, the need to align them with human values has become significant. Recent advanced techniques, such as prompt learning and reinforcement learning, are being employed to bring LLMs closer to aligning with human values. While these techniques address broad ethical and helpfulness concerns, they rarely consider simulating individualized human values. To bridge this gap, we propose SimVBG, a framework that simulates individual values based on individual backstories that reflect their past experience and demographic information. SimVBG transforms structured data on an individual to a backstory and utilizes a multi-module architecture inspired by the Cognitive–Affective Personality System to simulate individual value based on the backstories. We test SimVBG on a self-constructed benchmark derived from the World Values Survey and show that SimVBG improves top-1 accuracy by more than 10% over the retrieval-augmented generation method. Further analysis shows that performance increases as additional interaction user history becomes available, indicating that the model can refine its persona over time. Code, dataset, and complete experimental results are available at https://github.com/bangdedadi/SimVBG.

Mohammad Beigi, Ying Shen, Parshin Shojaee, Qifan Wang, Zichao Wang, Chandan K. Reddy, Ming Jin, Lifu Huang

Despite the remarkable capabilities of large language models, current training paradigms inadvertently foster sycophancy—alignment with user-provided information, regardless of factual accuracy. In this paper, we introduce SMART (Sycophancy Mitigation through Adaptive Reasoning Trajectories), reconceptualizing sycophancy as a reasoning optimization problem rather than an output alignment issue. SMART employs a two-stage approach: (1) Uncertainty-Aware Adaptive Monte Carlo Tree Search (UA-MCTS), which dynamically adjusts exploration based on state-level uncertainty; and (2) progress-based reinforcement learning that distills these improved reasoning patterns into model adaptation. Through extensive experiments, we show that SMART significantly outperforms existing baselines in effectively reducing sycophancy while maintaining performance on out-of-distribution inputs. These findings demonstrate the importance of optimizing internal reasoning processes for developing aligned truthful AI assistant.

Xisheng Xiao, Hanlin Zhao

Large language models (LLMs) have demonstrated strong performance in solving math problems, and there is growing research on evaluating their robustness. Unlike previous studies that create problem variants by adding perturbations to a single problem, this paper focuses on the interaction between problems. Specifically, we combine two original problems with a logical connection to get a new math problem, and measure the LLMs’ performance on it to evaluate its compositional generalization, which is an important and essential reasoning capability in human intelligence. The result of experiments that cover 14 different LLMs shows that even when the mathematical essence remains unchanged, a simple form of combination can significantly reduce the performance of LLMs, revealing the limitation of their generalization ability. Additionally, we propose an automated pipeline with 98.2% accuracy to assist in annotating datasets (1 manual, 2 synthetic). The extensive experiments conducted on these datasets further verify the conclusion and obtain some important findings. Finally, we analyze the impact of factors such as difficulty and length on LLMs’ performance, offering insights for future research.

Wei Yang, Jinwei Xiao, Hongming Zhang, Qingyang Zhang, Yanna Wang, Bo Xu

Recent advancements in Large Language Models (LLMs) have driven growing interest in LLM-based agents for complex planning tasks. To avoid costly agent training, many studies adopted memory mechanism that enhances LLM with offline experiences or online trajectory analysis. However, existing works focus on single-granularity memory derived from dynamic environmental interactions, which are inherently constrained by the quality of the collected experiences. This limitation, in turn, constrain the diversity of knowledge and the flexibility of planning. We propose Coarse-to-Fine Grounded Memory (CFGM), a novel framework that grounds coarse-to-fine memories with LLM, thereby fully leverage them for flexible adaptation to diverse scenarios. CFGM grounds environmental information into coarse-grained focus points to guide experience collection in training tasks, followed by grounding of actionable hybrid-grained tips from each experience. At inference, CFGM retrieves task-relevant experiences and tips to support planning. When facing environmental anomalies, the LLM grounds the current situation into fine-grained key information, enabling flexible self-QA reflection and plan correction. Extensive experiments on AlfWorld, Webshop and ScienceWorld demonstrate that CFGM significantly outperforms competitive baselines and comprehensively optimizes memory-enhanced LLM Agent system.

Weiyi Yang, Richong Zhang, Junfan Chen, Jiawei Sheng

Semi-supervised text classification (SSTC) aims to train text classification models with few labeled data and massive unlabeled data. Existing studies develop effective pseudo-labeling methods, but they can struggle with unlabeled data that have imbalanced classes mismatched with the labeled data, making the pseudo-labeling biased towards majority classes, resulting in catastrophic error propagation. We believe it is crucial to explicitly estimate the overall class distribution, and use it to calibrate pseudo-labeling to constrain majority classes. To this end, we formulate the pseudo-labeling as an optimal transport (OT) problem, which transports the unlabeled sample distribution to the class distribution. With a memory bank, we dynamically collect both the high-confidence pseudo-labeled data and true labeled data, thus deriving reliable (pseudo-) labels for class distribution estimation. Empirical results on 3 commonly used benchmarks demonstrate that our model is effective and outperforms previous state-of-the-art methods.

Nakyeong Yang, Minsung Kim, Seunghyun Yoon, Joongbo Shin, Kyomin Jung

Various studies have attempted to remove sensitive or private knowledge from a language model to prevent its unauthorized exposure. However, prior studies have overlooked the inherent complexity and interconnectedness of knowledge, which requires careful examination. To resolve this problem, we first define a new concept called superficial unlearning, which refers to the phenomenon where an unlearning method either fails to erase the interconnected knowledge it should remove or unintentionally erases irrelevant knowledge. Based on the definition, we introduce a novel benchmark, FaithUn, to analyze and evaluate the faithfulness of unlearning in real-world knowledge QA settings. Furthermore, we propose a novel unlearning method, KLUE, which updates only knowledge-related neurons to achieve faithful unlearning. KLUE leverages a regularized explainability method to localize contextual knowledge neurons, updating only these neurons using carefully selected unforgotten samples. Experimental results demonstrate that existing unlearning methods fail to ensure faithful unlearning, while our method shows significant effectiveness in real-world QA unlearning.

Senyu Li, Jiayi Wang, Felermino D. M. A. Ali, Colin Cherry, Daniel Deutsch, Eleftheria Briakou, Rui Sousa-Silva, Henrique Lopes Cardoso, Pontus Stenetorp, David Ifeoluwa Adelani

Evaluating machine translation (MT) quality for under-resourced African languages remains a significant challenge, as existing metrics often suffer from limited language coverage and poor performance in low-resource settings. While recent efforts, such as AfriCOMET, have addressed some of the issues, they are still constrained by small evaluation sets, a lack of publicly available training data tailored to African languages, and inconsistent performance in extremely low-resource scenarios. In this work, we introduce SSA-MTE, a large-scale human-annotated MT evaluation (MTE) dataset covering 13 African language pairs from the News domain, with over 63,000 sentence-level annotations from a diverse set of MT systems. Based on this data, we develop SSA-COMET and SSA-COMET-QE, improved reference-based and reference-free evaluation metrics. We also benchmark prompting-based approaches using state-of-the-art LLMs like GPT-4o and Claude. Our experimental results show that SSA-COMET models significantly outperform AfriCOMET and are competitive with the strongest LLM (Gemini 2.5 Pro) evaluated in our study, particularly on low-resource languages such as Twi, Luo, and Yoruba. All resources are released under open licenses to support future research.

Han Peng, Jinhao Jiang, Zican Dong, Wayne Xin Zhao, Lei Fang

Advancements in Large Language Models (LLMs) have extended their input context length, yet they still struggle with retrieval and reasoning in long-context inputs. Existing methods propose to utilize the prompt strategy and Retrieval-Augmented Generation (RAG) to alleviate this limitation. However, they still face challenges in balancing retrieval precision and recall, impacting their efficacy in answering questions. To address this, we introduce **CAFE**, a two-stage coarse-to-fine method to enhance multi-document question-answering capacities. By gradually eliminating the negative impacts of background and distracting documents, CAFE makes the responses more reliant on the evidence documents. Initially, a coarse-grained filtering method leverages retrieval heads to identify and rank relevant documents. Then, a fine-grained steering method guides attention to the most relevant content. Experiments across benchmarks show that CAFE outperforms baselines, achieving an average SubEM improvement of up to 22.1% and 13.7% over SFT and RAG methods, respectively, across three different models. Our code is available at https://github.com/RUCAIBox/CAFE.