论文检索

Jin-woo Lee, Junhwa Choi, Bongkyu Hwang, Jinho Choo, Bogun Kim, Jeongseon Yi, Joonseok Lee, DongYoung Jung, Jaeseon Park, Kyoungwon Park 等

We revisit continual pre-training for large language models and argue that progress now depends less on scaling parameters than on scaling the right structure. We introduce SCALE, a width upscaling architecture that inserts lightweight expansions into linear modules while freezing all pre-trained parameters, preserving residual and attention topologies and increasing capacity without perturbing the base model’s original functionality. SCALE follows two principles: Persistent Preservation, which maintains the base model’s behavior via preservation-oriented initialization and freezing of the pre-trained weights, and Collaborative Adaptation, which trains only selected expansion components to acquire new knowledge with minimal interference. We instantiate these ideas as SCALE-Preserve (preservation-first), SCALE-Adapt (adaptation-first), and SCALE-Route, an optional routing extension that performs token-level routing between preservation and adaptation heads. On a controlled synthetic biography benchmark, SCALE reduces the severe forgetting seen in depth expansion while still learning new knowledge. In continual pre-training on a Korean corpus, SCALE variants forget less on English evaluations and achieve competitive gains on Korean benchmarks, yielding the best overall stability-plasticity trade-off. We further analyze when preservation holds provably and why combining preservation and adaptation stabilizes optimization relative to standard continual learning.

Mainak Singha, Tanisha Gupta, Ankit Jha, Muhammad Haris Khan, Sayantani Ghosh, Biplab Banerjee

Pretrained biomedical vision–language models (VLMs) such as BioMedCLIP perform well on average but often degrade on challenging modalities where inter-class margins are small and acquisition-specific variations are pronounced, especially under few-shot supervision and when modality priors differ from pretraining corpora substantially. We propose BioVLM, a prompt-learning framework that improves cross-domain generalization without extensive backbone fine-tuning. BioVLM learns a diverse prompt bank and introduces dynamic prompt selection: for each input, it selects the most discriminative prompts via a low-entropy criterion on the predictive distribution, effectively coupling sparse few-shot evidence with rich LLM semantic priors. To strengthen this coupling, we distill high-confidence LLM-derived attributes and enforce robust knowledge transfer through strong/weak augmentation consistency. At test time, BioVLM adapts by choosing modality-appropriate prompts, enabling transfer to unseen categories and domains, while keeping training lightweight and inference efficient. On 11 MedMNIST+ 2D datasets, BioVLM achieves new state of the art across three distinct generalization settings. Codes are available at https://github.com/mainaksingha01/BioVLM.

Sadat Shahriar, Zheng Qi, Nikolaos Pappas, Srikanth Doss, Kishaloy Halder, Monica Sunkara, Manuel Mager, Yassine Benajiba

Aligning Large Language Models (LLM) to address subjectivity and nuanced preference levels requires adequate flexibility and control, which can be a resource-intensive and time-consuming procedure. Existing training-time alignment methods require full re-training when a change is needed and inference-time ones typically require access to the reward model at each inference step. We introduce **MEAV**, an inference-time model-editing-based LLM alignment method that learns encoded representations of preference dimensions, called *Alignment Vectors* (AV). These representations enable dynamic adjusting of the model behavior during inference through simple linear operations. Here, we focus on three gradual response levels across three specialized domains: medical, legal, and financial, exemplifying its practical potential. This new alignment paradigm introduces adjustable preference knobs during inference, allowing users to tailor their LLM outputs while reducing the inference cost by half compared to the prompt engineering approach. Additionally, we find that AVs are transferable across different fine-tuning stages of the same model, demonstrating their flexibility. AVs also facilitate multidomain, diverse preference alignment, making the process 12x faster than the retraining approach.

Chenye Zhao, Cornelia Caragea

Stance detection aims to ascertain whether an author’s text is in favor, against, or neutral toward specific targets like public policies or social issues. While pretrained language models (PLMs) have greatly enhanced stance detection, they remain vulnerable to adversarial attacks—manipulations that maintain textual semantics but lead to incorrect predictions. Such vulnerabilities remain underexplored for stance detection. In this study, we introduce StanceAttack, an innovative adversarial attack method leveraging ChatGPT to create adversarial examples that can mislead well-trained stance detection models. We conduct experiments to evaluate our attack model by attacking state-of-the-art PLMs on two benchmark datasets. Results demonstrate that StanceAttack outperforms traditional adversarial methods with higher success rates and fewer retries. Human evaluations confirm that our adversarial examples preserve the original semantic meanings and naturalness. We share our code and data in https://github.com/chenyez/StanceAttack.

Jiazhou Liang, Yifan Simon Liu, David Guo, Yilun Jiang, Minqi Sun, Scott Sanner

The growing ubiquity of Extended Reality (XR) is driving Conversational Recommendation Systems (CRS) toward visually immersive experiences. We formalize this paradigm as Immersive CRS (ICRS), where recommended items are highlighted directly in the user’s scene-based visual environment and augmented with in-situ labels. While item recommendation has been widely studied, the problem of how to select and evaluate which information to present as immersive labels remains an open problem. To this end, we introduce a principled categorization of information needs into explicit intent satisfaction and proactive information needs and use these to define novel evaluation metrics for item label selection. We benchmark IR-, LLM-, and VLM-based methods across three datasets and ICRS scenarios: fashion, movie recommendation, and retail shopping. Our evaluation reveals three important limitations of existing methods: (1) they fail to leverage scenario-specific information modalities (e.g., visual cues for fashion, metadata for retail), (2) they present redundant information that is visually inferable, and (3) they poorly anticipate users’ proactive information needs from explicit dialogue alone. In summary, this work provides both a novel evaluation paradigm for in-situ item labeling in ICRS and highlights key challenges for future work.

Manan Suri, Xiangci Li, Mehdi Shojaie, Songyang Han, Chao-Chun Hsu, Shweta Garg, Aniket Anand Deshmukh, Varun Kumar

Current AI-powered code assistance tools often struggle with ambiguous problem statements that lack sufficient task context and requirements specification. Recent analysis of software engineering agents reveals that failures on such ambiguous requests are highly correlated with longer trajectories involving either over-exploration or repeated attempts at applying the same fix without proper evolution or testing, leading to suboptimal outcomes across software development tasks. We introduce CodeScout, a breakthrough contextual query refinement approach that systematically converts ambiguous user requests into comprehensive, actionable problem statements through lightweight pre-exploration of the target codebase. Our key innovation is demonstrating that structured analysis before task execution can supplement existing agentic capabilities without requiring any modifications to their underlying scaffolds. CodeScout performs targeted context scoping, conducts multi-perspective analysis examining potential fixes and exploration opportunities, then synthesizes these insights into enhanced problem statements with reproduction steps, expected behaviors, and targeted exploration hints. This pre-exploration directly addresses the identified failure patterns by reducing non-converging agent trajectories while clarifying user intent in natural language space. We evaluate CodeScout using state-of-the-art agentic scaffolds and language models on SWEBench-Verified, demonstrating a 20% improvement in resolution rates with up to 27 additional issues resolved compared to the default baseline method. Our results suggest that systematic query refinement through contextual analysis represents a promising direction for enhancing AI code assistance capabilities.

Hosein Hasani, Mohammadali Banayeeanzade, Ali Nafisi, Sadegh Mohammadian, Fatemeh Askari, Mobin Bagherian, Amirmohammad Izadi, Mahdieh Soleymani Baghshah

Large language models (LLMs), despite strong performance on complex mathematical problems, exhibit systematic limitations in counting tasks. This issue arises from the architectural limits of transformers, where counting is performed across layers, leading to degraded precision for larger counting problems due to depth constraints. To address this limitation, we propose a simple test-time strategy inspired by System-2 cognitive processes that decomposes large counting tasks into smaller, independent sub-problems that the model can reliably solve. We evaluate this approach using observational and causal mediation analyses to understand the underlying mechanism of this System-2-like strategy. Our mechanistic analysis identifies key components: latent counts are computed and stored in the final item representations of each part, transferred to intermediate steps via dedicated attention heads, and aggregated in the final stage to produce the total count. Experimental results demonstrate that this strategy enables LLMs to surpass architectural limitations and achieve higher accuracy on large-scale counting tasks. This work provides mechanistic insight into System-2 counting in LLMs and presents a generalizable approach for improving and understanding their reasoning behavior.

Ran Zhang, Steffen Eger, Arda Tezcan, Wei Zhao, Simone Paolo Ponzetto, Lieve Macken

Large language models (LLMs) are increasingly used for creative tasks such as literary translation. Yet translational creativity remains underexplored and is rarely evaluated at scale, while source-text comprehension is typically studied in isolation, despite the fact that, in professional translation, comprehension and creativity are tightly intertwined. We address these gaps with a paired-task framework applied to literary excerpts from 11 books. Task 1 assesses source-text comprehension, and Task 2 evaluates translational creativity through Units of Creative Potential (UCPs), such as metaphors and wordplay. Using a scalable evaluation setup that combines expert human annotations with UCP-based automatic scoring, we benchmark 23 models and four creativity-oriented prompts. Our findings show that strong comprehension does not translate into human-level creativity: models often produce literal or contextually inappropriate renderings, with particularly large gaps for the more distant English–Chinese language pair. Creativity-oriented prompts yield only modest gains, and only one model, Mistral-Large, comes close to human-level creativity (0.167 vs. 0.246). Across all model–prompt combinations, only three exceed a creativity score of 0.1, while the rest remain at or near zero.

Xinyi Zhao, Haoqi Hu, Ziyu Wang, Jinfeng Xiao

Large language models (LLMs) are increasingly used as judges to evaluate, rank, and supervise other models, yet their reliability in judging LLMs’ reasoning process under long-context settings remains underexplored. Existing benchmarks either overly rely on human annotators, who may miss subtle flaws in lengthy reasoning chains, or focus solely on final responses while ignoring the underlying context and reasoning process. We introduce Long-Reason Bench (LRBench), a large-scale benchmark for evaluating LLM-based judges. LRBench comprises over 100K annotated instances spanning medical, legal, and academic-review scenarios, with fine-grained labels indicating violations of six core principles: Logical Correctness, Factual Consistency, Bias and Fairness, Groundedness, Helpfulness, and Harmlessness. Experimental results reveal that state-of-the-art LLM judges struggle to identify nuanced reasoning errors in long contexts. To improve judge reliability, we further present Judge-R1, which combines reinforcement learning with multi-turn search to enable grounded and principle-aware evaluation. Across domains and principles, Judge-R1 consistently outperforms single-turn baselines, enabling scalable and trustworthy evaluation of LLM reasoning. Our dataset and code are available at https://github.com/Xinyi-0724/Judge-R1.

Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha

LLM agents with tool-calling capabilities often fail when user instructions are ambiguous or incomplete, leading to incorrect invocations and task failures. Existing approaches operate in unstructured language spaces, generating clarifying questions through prompting strategies that lack principled criteria for determining which questions to ask and when to stop. We introduce a principled formulation of structured uncertainty that operates directly over tool parameters and their domains, cleanly separating specification uncertainty (what the user wants) from model uncertainty (what the LLM predicts). Our formulation uses Expected Value of Perfect Information (EVPI) to quantify the disambiguation value of each potential question, balanced against aspect-based cost modeling that prevents redundant questioning. We demonstrate the versatility of this formulation through two applications. First, SAGE-Agent uses structured uncertainty for inference-time question selection, achieving 7–39% higher coverage on ambiguous tasks while reducing clarification questions by 1.5–2.7 x compared to strong prompting and uncertainty-based baselines. Second, we show that structured uncertainty provides effective training signals: uncertainty-guided reward modeling boosts When2Call accuracy from 36.5% to 65.2% (3B model) and 36.7% to 62.9% (7B model) through uncertainty-weighted GRPO training, demonstrating more sample-efficient reinforcement learning for tool-calling agents. To enable evaluation, we present ClarifyBench, the first multi-turn dynamic tool-calling disambiguation benchmark. Our results establish structured uncertainty as a principled framework that improves both inference-time interaction efficiency and training-time sample efficiency in tool-augmented agents.

Antonio Lopardo, Avyukth Harish, Catherine Arnett, Akshat Gupta

Weight tying, i.e. sharing parameters between input and output embedding matrices, is common practice in language model design, yet its impact on the learned embedding space remains poorly understood. In this paper, we show that tied embedding matrices align more closely with output (unembedding) matrices than with input embeddings of comparable untied models, indicating that the shared matrix is shaped primarily for output prediction rather than input representation. This unembedding bias arises because output gradients dominate early in training. Using tuned lens analysis, we show this negatively affects early-layer computations, which contribute less effectively to the residual stream. Scaling input gradients during training reduces this bias, providing causal evidence for the role of gradient imbalance. This is mechanistic evidence that weight tying optimizes the embedding matrix for output prediction, compromising its role in input representation. These results help explain why weight tying can harm performance at scale and have implications for training smaller LLMs, where the embedding matrix contributes substantially to total parameter count.

Leixin Zhang, Cagri Coltekin

Humans often hold different perspectives on the same issues. In many NLP tasks, annotation disagreement can reflect valid subjective perspectives. Modeling annotator perspectives and understanding their relationship with other human factors, such as socio-demographic attributes, have received increasing attention. Prior work typically focuses on single demographic factors or limited combinations. However, in real-world settings, annotator perspectives are shaped by complex social contexts, and finer-grained socio-demographic attributes can better explain human perspectives. In this work, we propose Socio-Contrastive Learning, a method that jointly models annotator perspectives while learning socio-demographic representations. Our method provides an effective approach for the fusion of socio-demographic features and textual representations to predict annotator perspectives, outperforming standard concatenation-based methods. The learned representations further enable analysis and visualization of how demographic factors relate to variation in annotator perspectives. Our code is available at GitHub: https://github.com/Leixin-Zhang/Socio_Contrastive_Learning

Samuel Gideon Balter, Ethan Jerzak, Connor Thomas Jerzak

Multimodal LLMs can accurately perceive numerical content across modalities yet fail to perform exact multi-digit multiplication when the identical underlying arithmetic problem is presented as numerals, number words, images, or in audio form. Because existing benchmarks often lack systematically paired instances across modalities, it remains difficult to compare genuine arithmetic limits within and across model families. We therefore introduce a controlled multimodal multiplication benchmark that factorially varies digit length, digit sparsity, representation (e.g., numerals vs. number words), and modality (text, rendered images, and audio), with paired instances from a reproducible generator. We also define arithmetic load, C, as the product of the total and non-zero digit number as a compact, mechanistically motivated proxy for operation count. Across evaluations, accuracy falls sharply as C grows, often nearing zero by C > 100. Indeed, C remains predictive of performance across modalities and models, with R^2 > 0.5, nearing the value from more complex measures of arithmetic load that count the number of intermediate arithmetic steps. A separate perception-versus-computation decomposition shows that multimodal degradation is primarily computational rather than perceptual: on matched perception checks, models are near-perfect (>99\%) across modalities even when multiplication accuracy drops substantially. Beyond measuring when models fail, we ask which procedures they are predisposed to follow. We introduce a style-controlled forced-completion loss probe that scores heuristic-specific reasoning prefixes—including columnar multiplication, distributive decomposition, and rounding/compensation. Here, distributive decomposition is favored in both text and vision modalities; heuristic-specific LoRA adapters produce near-orthogonal updates yet degrade accuracy, indicating the base model maintains a well-tuned internal router.

Jianpeng Zhou, Qisheng Hu, Jiahai Wang, Wenya Wang

Steering large language model (LLM) reasoning via high-level reasoning actions offers a promising approach to improve robustness and interpretability. However, existing action-based paradigms, ranging from training-free prompting to static plan retrieval or prediction, often fail to consistently outperform standard generation because their planners tend to degenerate into repetitive loops or fixed patterns. We propose PG-HAP (Policy-Guided High-Level Action Planning), a lightweight stepwise planner–executor framework that learns to select reasoning actions dynamically while keeping the executor LLM fully frozen. The planner is trained with reinforcement learning to optimize answer correctness. To prevent degeneration, we introduce two targeted mechanisms: (i) an Action-Dependency Logit Mask that enforces valid transitions to avoid redundancy, and (ii) an Action Diversity Reward that discourages mode collapse by promoting varied action sequences. Across mathematical and commonsense reasoning benchmarks, PG-HAP improves accuracy over strong baselines while producing less redundant, more adaptive trajectories. This demonstrates that learning high-level planning alone can substantially strengthen reasoning without expensive end-to-end model tuning.

Jinjia Feng, Wenda Wang, Zhewei Wei

Large Language Models (LLMs) have shown great potential in molecular understanding by aligning molecular representations with text. However, existing approaches remain limited to static motif recognition without comprehending the generative principles—the connection rules governing how motifs assemble into valid topological structures. To address this challenge, we introduce **MotifAgent**, a multi-agent reinforcement learning framework inspired by emergent collective intelligence. We formulate molecular assembly as a collaborative problem where each motif is represented by an agent sharing a common LLM backbone, learning connection rules through explicit inter-motif negotiation rather than implicit sequence memorization. Key innovations include: (1) dynamic inter-agent negotiation for modeling motif connections; (2) Set-based Behavioral Cloning for learning multiple topologically equivalent assembly paths; (3) topology-aware reward shaping with MAPPO to maintain chemical validity while optimizing target properties. Extensive experiments demonstrate that MotifAgent achieves state-of-the-art performance across molecular property prediction, description generation, and reaction prediction tasks, with our generalist model surpassing specialized expert models.

Michela Lorandi, Anya Belz, Simon Mille, Craig Thomson

Systematic reviews are a cornerstone of modern science, synthesising evidence from published research to provide the highest level of research evidence in a field. The process includes categorising studies on a number of different dimensions which is laborious and time consuming. Automatic approaches are beginning to be explored but the complexity of the task means we are currently far from a satisfactory solution. In this paper, we test different annotation scheme-agnostic methods for automatic NLP paper categorisation for systematic reviews, and test them on two tasks: (i) annotating NLP papers for categories of reported controlled-text generation methods, and (ii) annotating NLP papers for categories of reported human evaluations. We find that reasoning-enhanced fine-tuning combined with DAPO reinforcement learning rewarding both correctness and output format substantially improves the performance of LLMs (by up to +53.8 points), even when they have been pre-trained to perform reasoning, and cuts time required for annotation by around 80% in a human-in-the-loop setting.

Young Min Cho, Daniele Bonadiman, Divya Bhargavi, Tamer Alkhouli, Salvatore Romeo, Dongwei Jiang, Khushbu Pahwa, Yubin Ge, Etsuko Ishii, Monica Sunkara 等

Training large foundation models for agentic tasks is increasingly impractical due to the high computational costs, long iteration cycles, and rapid obsolescence as new models are continuously released. Instead of post-training massive models for every new task or domain, we propose Supplement Generation Training (SGT), a more efficient and sustainable strategy. SGT trains a smaller LLM to generate useful supplemental text that, when appended to the original input, helps the larger LLM solve the task more effectively. These lightweight models can dynamically adapt supplements to task requirements, improving performance without modifying the underlying large models. This approach decouples task-specific optimization from large foundation models and enables more flexible, cost-effective deployment of LLM-powered agents in real-world applications.

Atharv Sonwane, Eng-Shen Tu, Wei-Chung Lu, Claas Beger, Carter Larsen, Debjit Dhar, Simon Alford, Rachel Chen, Ronit Pattanayak, Tuan Anh Dang 等

LLM-powered coding agents are redefining how real-world software is developed. To drive the research towards better coding agents, we require challenging benchmarks that can rigorously evaluate the ability of such agents to perform various software engineering tasks. However, popular coding benchmarks such as HumanEval and SWE-Bench focus on narrowly scoped tasks such as competition programming and patch generation. In reality, software engineers have to handle a broader set of tasks for real-world software development. To address this gap, we propose OmniCode, a novel software engineering benchmark that contains a broader and more diverse set of task categories beyond code or patch generation. Overall, OmniCode contains 1794 tasks spanning three programming languages – Python, Java, and C++ – and four key categories: bug fixing, test generation, code review fixing, and style fixing. In contrast to prior software engineering benchmarks, the tasks in OmniCode are (1) manually validated to eliminate ill-defined problems, and (2) synthetically crafted or recently curated to avoid data leakage issues, presenting a new framework for synthetically generating diverse software tasks from limited real-world data. We evaluate OmniCode with popular agent frameworks such as SWE-Agent and show that while they may perform well on bug fixing for Python, they fall short on tasks such as Test Generation and in languages such as C++ and Java. For instance, SWE-Agent achieves a maximum of 25.0% with DeepSeek-V3.1 on C++ Test Generation. OmniCode aims to serve as a robust benchmark and spur the development of agents that can perform well across different aspects of software development.

Pretam Ray, Pratik Prabhanjan Brahma, Zicheng Liu, Emad Barsoum

Evolutionary agentic systems intensify the trade-off between computational efficiency and reasoning capability by repeatedly invoking large language models (LLMs) during inference. This setting raises a central question: how can an agent dynamically select an LLM that is sufficiently capable for the current generation step while remaining computationally efficient? While model cascades offer a practical mechanism for balancing this trade-off, existing routing strategies typically rely on static heuristics or external controllers and do not explicitly account for model uncertainty. We introduce AdaptEvolve: Adaptive LLM Selection for Multi-LLM Evolutionary Refinement within an evolutionary sequential refinement framework that leverages intrinsic generation confidence to estimate real-time solvability. Empirical results show that confidence-driven selection yields a favorable Pareto frontier, reducing total inference cost by an average of 37.9% across benchmarks while retaining 97.5% of the upper-bound accuracy of static large-model baselines.

Midhat Urooj, Ayan Banerjee, Sandeep Gupta

Deep learning models deployed in clinical settings face two major challenges: domain generalization (DG) and long-tailed (LT) recognition. DG requires learning domain-invariant features to ensure robustness across heterogeneous acquisition protocols and patient populations. However, we identify a fundamental trade-off: objectives that enforce domain invariance often suppress class-discriminative signals essential for long-tailed recognition.To address this, we propose the Agentic Causal Disentanglement (CANDICE) Framework, a modular architecture that integrates explicit clinical expertise from sonographers, radiologists, and specialists as a form of causal intervention. The framework combines clinical reasoning, causal representation learning, and automated pipeline construction to disentangle domain-invariant and class-discriminative features. By incorporating domain-specific causal knowledge, it effectively decouples the objectives of DG and LT learning. We evaluate CANDICE on 10 diverse medical imaging datasets spanning four modalities. The framework achieves an average performance improvement of 10.3% across both multi-domain and in-domain long-tailed tasks, demonstrating its effectiveness in handling distribution shifts while preserving minority class performance.