论文检索

Anushka Sivakumar, Andrew Zhang, Zaber Ibn Abdul Hakim, Chris Thomas

This work introduces SteerVLM, a lightweight steering module designed to guide Vision-Language Models (VLMs) towards outputs that better adhere to desired instructions. Our approach learns from the latent embeddings of paired prompts encoding target and converse behaviors to dynamically adjust activations connecting the language modality with image context. This allows for fine-grained, inference-time control over complex output semantics without modifying model weights while preserving performance on off-target tasks. Our steering module requires learning parameters equal to 0.14% of the original VLM’s size. Our steering module gains model control through dimension-wise activation modulation and adaptive steering across layers without requiring pre-extracted static vectors or manual tuning of intervention points. Furthermore, we introduce VNIA (Visual Narrative Intent Alignment), a multimodal dataset specifically created to facilitate the development and evaluation of VLM steering techniques. Our method outperforms existing intervention techniques on steering and hallucination mitigation benchmarks for VLMs and proposes a robust solution for multimodal model control through activation engineering.

Sahiti Yerramilli, Nilay Pande, Rynaa Grover, Jayant Sravan Tamarapalli

This paper introduces GeoChain, a large-scale benchmark for evaluating step-by-step geographic reasoning in multimodal large language models (MLLMs). Leveraging 1.46 million Mapillary street-level images, GeoChain pairs each image with a 21-step chain-of-thought (CoT) question sequence (over 30 million Q&A pairs). These sequences guide models from coarse attributes to fine-grained localization across four reasoning categories - visual, spatial, cultural, and precise geolocation - annotated by difficulty. Images are also enriched with semantic segmentation (150 classes) and a visual locatability score. Our benchmarking of frontier MLLMs on a diverse 2,088-image subset reveals consistent challenges: models frequently exhibit weaknesses in visual grounding, display erratic reasoning, and struggle to achieve accurate localization, especially as the reasoning complexity escalates. GeoChain offers a robust diagnostic methodology, critical for fostering significant advancements in complex geographic reasoning within MLLMs.

Yoel Ashkenazi, Etzion Harari, Regev Yehezkel Imra, Naphtali Abudarham, Dekel Cohen, Yoram Louzoun

Knowledge discovery from large-scale, heterogeneous textual corpora presents a significant challenge. Document clustering offers a practical solution by organizing unstructured texts into coherent groups based on content and thematic similarity. However, clustering does not inherently ensure thematic consistency. Here, we propose a novel framework that constructs a similarity graph over document embeddings and applies iterative graph-based clustering algorithms to partition the corpus into initial clusters. To overcome the limitations of conventional methods in producing semantically consistent clusters, we incorporate iterative feedback from a large language model (LLM) to guide the refinement process. The LLM is used to assess cluster quality and adjust edge weights within the graph, promoting better intra-cluster cohesion and inter-cluster separation. The LLM guidance is based on a set of success Rate metrics that we developed to measure the semantic coherence of clusters. Experimental results on multiple benchmark datasets demonstrate that the iterative process and additional user-supplied a priori edges improve the summaries’ consistency and fluency, highlighting the importance of known connections among the documents. The removal of very rare or very frequent sentences has a mixed effect on the quality scores.Our full code is available here: https://github.com/D2CS-sub/D2CS

Gyunyeop Kim, Sangwoo Kang

The performance of MoE-based LLMs depends on the router’s ability to select suitable experts; however, the router is typically not explicitly supervised to acquire this routing ability. We propose Exploration-Driven Reinforcement Learning (ERL), which explicitly optimizes the router by exploration of alternative routing paths. For every input, ERL evaluates by (i) the original routing path and (ii) paths in which an \alpha-fraction of routing decisions is randomly perturbed, and treats their performance gap as an advantage signal in a reinforcement learning. Moreover, MoE-ERL_{wPL} mitigates the risk of performance collapse caused by routing reinforcement learning–induced expert over-specialization by intentionally enforcing overlap in experts’ knowledge. Without adding parameters or external reward models, our method improves summarization (SAMSum, XSUM), question answering (SQuAD), and language modeling (WikiText-2), and raises routing quality, delivering up to 8.9 × higher MRR than baselines over 100 perturbed routing paths. Code is available at our github.

WonJin Yoon, Ian Bulovic, Timothy A. Miller

Large language models perform surprisingly well on many zero-shot classification tasks, but are difficult to fairly compare to supervised classifiers due to the lack of a modifiable decision boundary. In this work, we propose and evaluate a method that transforms binary classification tasks into pairwise comparisons between instances within a dataset, using LLMs to produce relative rankings of those instances. Repeated pairwise comparisons can be used to score instances using the Elo rating system (used in chess and other competitions), inducing a confidence ordering over instances in a dataset. We evaluate scheduling algorithms for their ability to minimize comparisons, and show that our proposed algorithm leads to improved classification performance, while also providing more information than traditional zero-shot classification.

Md. Atabuzzaman, Andrew Zhang, Chris Thomas

Large Vision-Language Models (LVLMs) have demonstrated impressive performance on vision-language reasoning tasks. However, their potential for zero-shot fine-grained image classification, a challenging task requiring precise differentiation between visually similar categories, remains underexplored. We present a novel method that transforms zero-shot fine-grained image classification into a visual question-answering framework, leveraging LVLMs’ comprehensive understanding capabilities rather than relying on direct class name generation. We enhance model performance through a novel attention intervention technique. We also address a key limitation in existing datasets by developing more comprehensive and precise class description benchmarks. We validate the effectiveness of our method through extensive experimentation across multiple fine-grained image classification benchmarks. Our proposed method consistently outperforms the current state-of-the-art (SOTA) approach, demonstrating both the effectiveness of our method and the broader potential of LVLMs for zero-shot fine-grained classification tasks. Code and Datasets: https://github.com/Atabuzzaman/Fine-grained-classification

Yixuan Liu, Abel Elekes, Jianglin Lu, Rodrigo Dorantes-Gilardi, Albert-Laszlo Barabasi

Large language models (LLMs) are reshaping how scientific knowledge is accessed and represented. This study evaluates the extent to which popular and frontier LLMs including GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro recognize scientists, benchmarking their outputs against OpenAlex and Wikipedia. Using a dataset focusing on 100,000 physicists from OpenAlex to evaluate LLM recognition, we uncover substantial disparities: LLMs exhibit selective and inconsistent recognition patterns. Recognition correlates strongly with scholarly impact such as citations, and remains uneven across gender and geography. Women researchers, and researchers from Africa, Asia, and Latin America are significantly underrecognized. We further examine the role of training data provenance, identifying Wikipedia as a potential sources that contributes to recognition gaps. Our findings highlight how LLMs can reflect, and potentially amplify existing disparities in science, underscoring the need for more transparent and inclusive knowledge systems.

Narmeen Fatimah Oozeer, Luke Marks, Fazl Barez, Amir Abdullah

Controlling multiple behavioral attributes in large language models (LLMs) at inference time is a challenging problem due to interference between attributes and the limitations of linear steering methods, which assume additive behavior in activation space and require per-attribute tuning. We introduce K-Steering, a unified and flexible approach that trains a single non-linear multi-label classifier on hidden activations and computes intervention directions via gradients at inference time. This avoids linearity assumptions, removes the need for storing and tuning separate attribute vectors, and allows dynamic composition of behaviors without retraining. To evaluate our method, we propose two new benchmarks, TONEBANK and DEBATEMIX, targeting compositional behavioral control. Empirical results across 3 model families, validated by both activation-based classifiers and LLM-based judges, demonstrate that K-Steering outperforms strong baselines in accurately steering multiple behaviors.

Tianchun Li, Tianci Liu, Xingchen Wang, Rongzhe Wei, Pan Li, Lu Su, Jing Gao

Large language models (LLMs) have achieved promising results in tabular data generation. However, inherent historical biases in tabular datasets often cause LLMs to exacerbate fairness issues, particularly when multiple advantaged and protected features are involved. In this work, we introduce a universal debiasing framework that minimizes group-level dependencies by simultaneously reducing the mutual information between advantaged and protected attributes. By leveraging the autoregressive structure and analytic sampling distributions of LLM-based tabular data generators, our approach efficiently computes mutual information, reducing the need for cumbersome numerical estimations. Building on this foundation, we propose two complementary methods: a direct preference optimization (DPO)-based strategy, namely UDF-DPO, that integrates seamlessly with existing models, and a targeted debiasing technique, namely UDF-MIX, that achieves debiasing without tuning the parameters of LLMs. Extensive experiments demonstrate that our framework effectively balances fairness and utility, offering a scalable and practical solution for debiasing in high-stakes applications.

Mohammad Saim, Phan Anh Duong, Cat Luong, Aniket Bhanderi, Tianyu Jiang

The embodiment of emotional reactions from body parts contains rich information about our affective experiences. We propose a framework that utilizes state-of-the-art large vision language models (LVLMs) to generate Embodied LVLM Emotion Narratives (ELENA). These are well-defined, multi-layered text outputs, primarily comprising descriptions that focus on the salient body parts involved in emotional reactions. We also employ attention maps and observe that contemporary models exhibit a persistent bias towards the facial region. Despite this limitation, we observe that our employed framework can effectively recognize embodied emotions in face-masked images, outperforming baselines without any fine-tuning. ELENA opens a new trajectory for embodied emotion analysis across the modality of vision and enriches modeling in an affect-aware setting.

Joshua Tint

This position paper argues that recent progress with diversity in NLP is disproportionately concentrated on a small number of areas surrounding fairness. We further argue that this is the result of a number of incentives, biases, and barriers which come together to disenfranchise marginalized researchers in non-fairness fields, or to move them into fairness-related fields. We substantiate our claims with an investigation into the demographics of NLP researchers by subfield, using our research to support a number of recommendations for ensuring that all areas within NLP can become more inclusive and equitable. In particular, we highlight the importance of breaking down feedback loops that reinforce disparities, and the need to address geographical and linguistic barriers that hinder participation in NLP research.

Vishakh Padmakumar, Joseph Chee Chang, Kyle Lo, Doug Downey, Aakanksha Naik

The increasing volume of academic literature makes it essential for researchers to organize, compare, and contrast collections of documents. Large language models (LLMs) can support this process by generating schemas defining shared aspects along which to compare papers. However, progress on schema generation has been slow due to: (i) ambiguity in reference-based evaluations, and (ii) lack of editing/refinement methods. Our work is the first to address both issues. First, we present an approach for augmenting unannotated table corpora with synthesized intents, and apply it to create a dataset for studying schema generation conditioned on a given information need, thus reducing ambiguity. With this dataset, we show how incorporating table intents significantly improves baseline performance in reconstructing reference schemas. We start by comprehensively benchmarking several single-shot schema generation methods, including prompted LLM workflows and fine-tuned models, showing that smaller, open-weight models can be fine-tuned to be competitive with state-of-the-art prompted LLMs. Next, we propose several LLM-based schema refinement techniques and show that these can further improve schemas generated by these methods.

Lu Wang, Chiming Duan, Pu Zhao, Fangkai Yang, Yong Shi, Xuefeng Luo, Bingjing Xu, Weiwei Deng, Qingwei Lin, Dongmei Zhang

Measuring the relevance between user queries and advertisements is a critical task for advertisement (ad) recommendation systems, such as Microsoft Bing Ads and Google Ads. Traditionally, this requires expert data labeling, which is both costly and time-consuming. Recent advances have explored using Large Language Models (LLMs) for labeling, but these models often lack domain-specific knowledge. In-context learning (ICL), which involves providing a few demonstrations, is a common practice to enhance LLM performance on domain-specific tasks. However, retrieving high-quality demonstrations in a vast exploration space remains challenging. In this paper, we introduce ICL-Bandit, a practical and effective approach that leverages ICL to enhance the query-ad relevance labeling capabilities of LLMs. We develop a novel bandit learning method to identify and provide superior demonstrations for ICL, thereby improving labeling performance. Experimental results demonstrate that ICL-Bandit achieves state-of-the-art performance compared to existing methods. Additionally, ICL-Bandit has been deployed in Company X, that serves billions of users worldwide, confirming its robustness and effectiveness.

Joshua Tint

Out-of-distribution (OOD) detection is a key safeguard for large language models, especially when they’re deployed in real-world applications. However, existing OOD methods often struggle with prompts that are deliberately obfuscated, context-dependent, or superficially benign—making it hard to distinguish between harmless queries and adversarial or dangerous ones. These methods typically assess prompts in isolation, missing important semantic cues from the model’s response. We introduce PROOD, prompt-response OOD detection, a framework that jointly analyzes LLM prompts *and their corresponding outputs* to improve semantic understanding. PROOD supports zero-shot multiclass detection using synthetic data generation and it offers a tunable probabilistic classification output. We validate PROOD on three challenging benchmarks—TrustLLM, OR-Bench, and AdvBench—where consistently outperforms prior OOD techniques, improving F1 scores by up to 6.3 points, from 0.871 to 0.934. Our results show that incorporating model responses enables more accurate, context-aware OOD detection in complex and adversarial prompt environments.

Xiaoxue Han, Pengfei Hu, Chang Lu, Jun-En Ding, Feng Liu, Yue Ning

Deep learning models trained on extensive Electronic Health Records (EHR) data have achieved high accuracy in diagnosis prediction, offering the potential to assist clinicians in decision-making and treatment planning. However, these models lack two crucial features that clinicians highly value: interpretability and interactivity. The “black-box” nature of these models makes it difficult for clinicians to understand the reasoning behind predictions, limiting their ability to make informed decisions. Additionally, the absence of interactive mechanisms prevents clinicians from incorporating their own knowledge and experience into the decision-making process. To address these limitations, we propose II-KEA, a knowledge-enhanced agent-driven causal discovery framework that integrates personalized knowledge databases and agentic LLMs. II-KEA enhances interpretability through explicit reasoning and causal analysis, while also improving interactivity by allowing clinicians to inject their knowledge and experience through customized knowledge bases and prompts. II-KEA is evaluated on both MIMIC-III and MIMIC-IV, demonstrating superior performance along with enhanced interpretability and interactivity, as evidenced by its strong results from extensive case studies.

Chuan He, Zhuozhao Li, Song Guo, Xiaocheng Lu, Jinxiang Lai

Vision-Language Models (VLMs), such as CLIP, have exhibited significant advancements in recognizing visual concepts through natural language guidance. However, adapting these models to downstream tasks remains challenging. Existing adaptation methods either overlook the structural knowledge between the text and image modalities or create overly complex graphs containing redundant information for alignment, leading to suboptimal classification performance and increased computational overhead. This paper proposes a novel adapter-tuning methodology named Homogeneous Graph Adapter (HomoGraphAdapter), which transforms diverse textual and visual descriptions into a unified set of node representations and establishes edges between nodes for inter-modal and cross-modal semantic alignment. We leverage a straightforward homogeneous Graph Neural Network (GNN) to adapt positive and negative classifiers across text and image modalities. The classifiers comprehensively enhance the performance for few-shot classification and OOD generalization. Compared with the SOTA approach HeGraphAdapter, HomoGraphAdapter improves classification accuracy by an average of 1.51% for 1-shot and 0.74% for 16-shot on 11 datasets, while also reducing both precomputation time and training time.

Sophie Wu, Jan Philip Wahle, Saif M. Mohammad

This paper is the first investigation of the connection between emotion, embodiment, and everyday language in a large sample of natural language data. We created corpora of body part mentions (BPMs) in online English text (blog posts and tweets). This includes a subset featuring human annotations for the emotions of the person whose body part is mentioned in the text. We show that BPMs are common in personal narratives and tweets (~5% to 10% of posts include BPMs) and that their usage patterns vary markedly by time and location. Using word–emotion association lexicons and our annotated data, we show that text containing BPMs tends to be more emotionally charged, even when the BPM is not explicitly used to describe a physical reaction to the emotion in the text. Finally, we discover a strong and statistically significant correlation between body-related language and a variety of poorer health outcomes. In sum, we argue that investigating the role of body-part related words in language can open up valuable avenues of future research at the intersection of NLP, the affective sciences, and the study of human wellbeing.

Xingjian Tao, Yiwei Wang, Yujun Cai, Zhicheng Yang, Jing Tang

Multimodal large language models (MLLMs) have enabled GUI agents to interact with operating systems by grounding language into spatial actions. Despite their promising performance, these models frequently exhibit hallucinations—systematic localization errors that compromise reliability. We propose a fine-grained evaluation framework that categorizes model predictions into four distinct types, revealing nuanced failure modes beyond traditional accuracy metrics. To better quantify model uncertainty, we introduce the Peak Sharpness Score (PSS), a metric that evaluates the alignment between semantic continuity and logits distribution in coordinate prediction. Building on this insight, we further propose Context-Aware Cropping, a training-free technique that improves model performance by adaptively refining input context. Extensive experiments demonstrate that our framework and methods provide actionable insights and enhance the interpretability and robustness of GUI agent behavior.

Tianyi Zhang

Large language models (LLMs) encode rich internal representations of political ideology, but it remains unclear how these representations contribute to model decision-making, and how these latent dimensions interact with one another. In this work, we investigate whether ideological directions identified via linear probes—specifically, those predicting DW-NOMINATE scores from attention head activations—influence model behavior in downstream political tasks. We apply inference-time interventions to steer a decoder-only transformer along learned ideological directions, and evaluate their effect on three tasks: political bias detection, voting preference simulation, and bias neutralization via rewriting. Our results show that learned ideological representations generalize well to bias detection, but not as well to voting simulations, suggesting that political ideology is encoded in multiple, partially disentangled latent structures. We also observe asymmetries in how interventions affect liberal versus conservative outputs, raising concerns about pretraining-induced bias and post-training alignment effects. This work highlights the risks of using biased LLMs for politically sensitive tasks, and calls for deeper investigation into the interaction of social dimensions in model representations, as well as methods for steering them toward fairer, more transparent behavior.

Kazi Tasnim Zinat, Saad Mohammad Abrar, Shoumik Saha, Sharmila Duppala, Saimadhav Naga Sakhamuri, Zhicheng Liu

Vision-Language Models have shown both impressive capabilities and notable failures in data visualization understanding tasks, but we have limited understanding on how specific properties within a visualization type affect model performance. We present ProcVQA, a benchmark designed to analyze how VLM performance can be affected by structure type and structural density of visualizations depicting frequent patterns mined from sequence data. ProcVQA consists of mined process visualizations spanning three structure types (linear sequences, tree, graph) with varying levels of structural density (quantified using the number of nodes and edges), with expert-validated QA pairs on these visualizations. We evaluate 21 proprietary and open-source models on the dataset on two major tasks: visual data extraction (VDE) and visual question answering (VQA) (with four categories of questions). Our analysis reveals three key findings. First, models exhibit steep performance drops on multi-hop reasoning, with question type and structure type impacting the degradation. Second, structural density strongly affects VDE performance: hallucinations and extraction errors increase with edge density, even in frontier models. Third, extraction accuracy does not necessarily translate into strong reasoning ability. By isolating structural factors through controlled visualization generation, ProcVQA enables precise identification of VLM limitations. ProcVQA is available at: https://github.com/kzintas/ProcVQA.