paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance

Mika Okamoto, Ansel Kaplan Erol, Kutluhan Erol

Specifying a penalty can turn a legal obligation into a cost-benefit calculation that favors violation. We show that this enforcement information paradox occurs in AI agents. Most AI safety evaluations test whether models fail; we ask why, using compliance theory from law and economics as a diagnostic. We evaluate twelve instruction-tuned language models deployed as enterprise procurement chatbots. Each is given an environmental regulation in its system prompt covering large purchases, and a vendor list on which the certified suppliers cost nearly twice what the uncertified ones do. We test the agents against the predictions of deterrence, legitimacy, and expressive law, and find that each theory accounts for part of what we observe. Under identical conditions, compliance spans 46 percentage points across models, and models differ in which pressure breaks them: some treat the regulation as binding however it is worded, while others fail where theory predicts, under low penalties and non-command phrasing. Benchmark scores and developers' own descriptions of post-training do not predict where a model falls. Across all twelve, financial incentives, managerial demands, peer outcomes, and employee pressure each produce large compliance failures. These agents violate regulatory constraints to satisfy local user objectives in ways standard alignment benchmarks do not measure. Embedding the rule in the system prompt is not on its own enough to produce a compliant agent: model selection is itself a governance decision, and benchmark evaluation is not sufficient for compliance-sensitive deployments.

arXiv

LODESTAR: Robust Entropy-Based Answer Selection in Retrieval-Augmented Generation for Question Answering -- Directing Frozen-LLM Entropy with a Reinforcement-Learned Prompt Polarizer under Misleading Passages

Hung-Chun Hsu, Po-Jen Ko, Che-Cheng Wu, Li-Yang Chang, Chuan-Ju Wang

Predictive-distribution entropy is a strong answer-selection rule in retrieval-augmented generation (RAG) for question answering: across five QA benchmarks, selecting the answer a frozen respondent LLM produces with the lowest answer-token entropy lifts mean $F_1$ from 0.4769 to 0.5148 over the retriever's top-ranked passage, without gold answers. Yet this rule, which prior entropy-based selectors adopt, fails: a misleading passage makes the respondent confidently wrong, driving entropy down where the uncertainty signal looks most trustworthy. The failure comes from the passage the respondent reads, and the context it is read in is an input we can intervene on. We introduce LODESTAR: to our knowledge the first method to score a text intervention by the uncertainty it induces in a third-party frozen respondent, compared within one question. LODESTAR uses reinforcement learning (GRPO) to train, once and offline, a polarizer -- a short fixed natural-language string inserted into the respondent's prompt and never into its weights, directing entropy so that entropy-based answer selection stays robust to misleading passages; training labels are built from gold answers and two LLM judges, and inference reads neither. With every competing selector under the same frozen respondent and candidate pools on 5,008 questions, LODESTAR attains the highest mean $F_1$ of any inference-ready selector (0.5339), the highest macro exact match (0.4136), and the highest GPT-4o judge score of the frozen-respondent configurations judged (0.6435); its three-seed mean wins all 70 $F_1$ cells against fourteen published configurations and is paired-significant on $F_1$ against every one. The gain holds in-domain on NQ-Open and out-of-domain over SQuAD, TriviaQA, EntityQuestions and WebQuestions. Ablating the polarizer shows it is what makes the respondent read a misleading passage less often (26.0% vs 30.3%).

arXiv

SPyCE: Skill-Policy Co-evolution for Multimodal Agents

Ru Zhang, Weijie Qiu

Multimodal agents that think with images iteratively manipulate visual evidence and invoke tools across many steps. Existing reinforcement learning methods reduce trajectories to scalar rewards, forcing the policy to discover reusable tool-use patterns from scratch on every new task; memory-based alternatives retain past experience, yet they rely on test-time retrieval, without updating the policy to absorb reusable patterns from that experience. Our key insight is that multimodal reasoning trajectories should be distilled into reusable skills that co-evolve with the policy during training, rather than being consumed as rewards or retrieved from a static store. To this end, we propose SPyCE (Skill-Policy Co-evolution), a framework that distills trajectories into a hierarchical skill library and updates it throughout reinforcement learning. Execution skills capture local visual operations, while workflow skills encode high-level priors that orchestrate tool use. During training, the policy model conditions on retrieved skills to guide its rollouts, while the skill library evolves using valuable rollouts generated by the policy. This creates a closed loop in which improved policies yield better skills, and the evolving skill library, in turn, provides stronger priors for policy rollouts. Experiments across eight benchmarks demonstrate that SPyCE consistently outperforms both RL-based and memory-based baselines. Further analysis reveals that both the hierarchical skill design and the co-evolution mechanism are critical to our design. These results suggest joint skill-policy optimization as a promising paradigm for building capable multimodal agents.

arXiv

RepSelect: Robust LLM Unlearning via Representation Selectivity

Filip Sondej, Yushi Yang, Adam Mahdi

When LLM weights are open or fine-tuning is available through an API, suppressing hazardous knowledge and tendencies is not enough: removal has to be deep enough that an adversary cannot restore it. Existing unlearning is shallow by this standard: fine-tuning or a handful of in-context examples brings the behaviour back, and it often degrades general capabilities in the process. We identify a root cause: existing methods edit representations shared with the retain set and lying in the subspace that a fine-tuning attacker recovers, making unlearning simultaneously easy to undo and disruptive. Leveraging this, we propose RepSelect (Representation Selectivity), which isolates forget-set-specific representations by collapsing the top principal components of the weight gradients before each unlearning update, preserving general capabilities while limiting what fine-tuning can recover. Across five unlearning datasets spanning both knowledge (biohazard, cyber, facts about real individuals) and tendencies (abusive, sycophantic), and three model families covering dense and Mixture-of-Experts architectures, RepSelect yields a 4-40x larger drop in post-relearning answer probability than five widely used baselines (GradDiff, NPO, SimNPO, RMU, UNDIAL). It is also near-perfectly robust to few-shot prompting and holds under an adaptive attack designed to exploit its mechanism. Our results show that unlearning needs to be selective about which representations it edits.

arXiv

Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Models

Reza Khanmohammadi, Kundan Thind, Mohammad M. Ghassemi

A vision-language model can answer a question about a chest radiograph or a pathology slide fluently and confidently while barely using the image, relying instead on language priors. In medicine this is the failure that matters most: the answer looks trustworthy and is not, and the natural safeguard is a confidence score reliable enough to say when the model should abstain. We ask a deployment question rather than an accuracy one: how much medical imaging work a vision-language model can safely defer on its own, and which confidence signal makes that possible. We evaluate nine confidence estimators, spanning training-free logit baselines, prompt-based self-reports, and trained internal probes, across five open-weight LVLMs and three medical VQA datasets covering broad clinical imaging, radiology, and pathology, every probe trained only on natural images and applied to medicine without adaptation. Recast as bounded selective prediction, the comparison is cautionary. Standard metrics mislead: discrimination barely separates the estimators, and a fixed high-confidence cutoff separates them far less than it appears, because their scores sit on incomparable scales; no estimator is reliably best across domains or models. What can be safely deferred is set at two levels: base-model competence fixes a ceiling, and the confidence layer determines how much of it is reachable. At a 20% error tolerance the strongest estimator defers about a quarter of radiology cases under a distribution-free guarantee and a third under a held-out threshold, and little to none of pathology. The usable role is calibrated triage under clinical oversight, not autonomous deferral: a good estimator makes a competent model defer safely where it is competent, but none manufactures reliability where the base model lacks it. We release all outputs, correctness judgments, and confidence scores, with code.

arXiv

Self-Harness: Harnesses That Improve Themselves

Hangfan Zhang, Shao Zhang, Kangcong Li, Chen Zhang, Yang Chen, Yiqun Zhang, Lei Bai, Shuyue Hu

The performance of LLM-based agents is jointly shaped by their base models and the harnesses that mediate their interaction with the environment. Because different models exhibit distinct behaviors, effective harness design is inherently model-specific. Yet agent harnesses are still largely engineered by human experts, a paradigm that scales poorly as modern LLMs become increasingly diverse and rapidly evolving. In this paper, we introduce Self-Harness, a new paradigm in which an LLM-based agent improves its own operating harness, without relying on human engineers or stronger external agents. We operationalize Self-Harness as an iterative loop with three stages: Weakness Mining, which identifies model-specific failure patterns from execution traces; Harness Proposal, which generates diverse yet minimal harness modifications tied to these failures; and Proposal Validation, which accepts candidate edits only after regression testing. We instantiate Self-Harness across Terminal-Bench-2.0, SWE-bench Verified, and AppWorld using a minimal initial harness and three base models from diverse families: MiniMax M2.5, Qwen3.5-35B-A3B, and GLM-5. Across all nine model--benchmark combinations, every final harness improves both held-in and held-out pass rates, with overall relative gains of up to 132%. Qualitative analyses further show that the retained mechanisms address benchmark-specific bottlenecks in artifact handling and runtime control, software-patch verification, and application-state retrieval. These results suggest a path toward LLM-based agents that are not merely shaped by their harnesses, but can also participate in reshaping them.

arXiv

A Finite-Calibration Regime Map for LLM Judge Panels

Bin Zhu, Yi Xie, Yanghui Rao

Deploying an LLM judge panel spends human labels on fitting a calibrator, constructing candidate judge paths, and validating which candidate to deploy. We study when finite labels should support a low-dimensional stacker or reliability model, and when an unrestricted joint output table is worth its cell-count and unseen-pattern cost. We cast this as a finite-calibration regime map and instantiate it as Finite-Calibration Panel Selection (FCPS), a validation selector over judge path, deployed panel size, and aggregator family with support diagnostics. Across RewardBench, LLMBar, SummEval, and Arena100K with a seven-judge pool, scalar/reliability aggregation has lower MSE than unrestricted joint-table calibration in 16 of 20 real dataset--budget cells by point estimate, while paired 95% intervals exclude zero in 11 cells; richer backoff/shrinkage tables narrow some gaps while preserving the finite-support bottleneck. Controlled calibration-growth data show the opposite regime: when labels contain a six-way interaction, the selected table grows to the interaction-bearing prefix and its MSE falls from 0.224 to 0.061 once unseen mass vanishes. The practical deployment question is whether the next judge's information is estimable under the available human labels.

arXiv

Investigating the Interplay between Contextual and Parametric Chain-of-Thought Faithfulness under Optimization

Jingyi Sun, Qianli Wang, Pepa Atanasova, Nils Feldhus, Isabelle Augenstein

Chain-of-Thought (CoT) faithfulness, i.e., whether CoTs genuinely reflect large language models' (LLM) underlying behavior, is typically evaluated with metrics under two disjoint paradigms: contextual faithfulness, measured by perturbing the input or CoT trace, and parametric faithfulness, assessed by intervening on a model's parametric knowledge. Yet prior work compares them only descriptively. We fill this gap by proposing FaithMATE, a unified preference-alignment interface for optimizing models towards either faithfulness paradigm. It enables us to investigate the interplay between the two paradigms, examining whether and to what extent faithfulness gains generalize within and across paradigms. Across three models, two datasets, and six faithfulness metrics, we find that the two paradigms are positively coupled, yet asymmetric: optimizing towards parametric faithfulness yields consistent gains across both paradigms, whereas the contextual counterpart delivers more variable gains. Within the contextual paradigm, faithfulness gains on one metric do not consistently transfer to others, implying that existing contextual metrics capture disjoint facets of faithfulness and exposing inherent trade-offs. These findings imply that CoT faithfulness is not a monolithic objective and therefore requires multifaceted optimization and evaluation.

arXiv

EvoSelect: Data-Efficient LLM Evolution for Targeted Task Adaptation

Ting Wei Li, Sirui Chen, Jiaru Zou, Yingbing Huang, Tianxin Wei, Jingrui He, Hanghang Tong

Adapting large language models (LLMs) to a targeted task efficiently and effectively remains a fundamental challenge. Such adaptation often requires iteratively improving the model toward a targeted task, yet collecting high-quality human-labeled data to support this process is costly and difficult to scale. As a result, synthetic data generation has emerged as a flexible and scalable alternative. One straightforward approach is through an iterative generation-training loop, where candidate data are synthesized through an external generator, the model is updated using these data and the process is repeated over iterations. However, generated samples can be noisy, highly redundant, or even misaligned with the targeted task distribution. Training indiscriminately on such data can dilute useful learning signals and even degrade model performance. To address this, we introduce a refined paradigm, namely an iterative generation-selection-training loop, which incorporates a selection step prior to model updates. Building on this paradigm, we propose EvoSelect, a data-efficient framework to evolve LLM effectively. Given candidate samples produced by the data generator, EvoSelect selects training data by jointly modeling targeted task alignment and diversity. We estimate task relevance through optimal transport with proxy gradient representations, which quantifies how well candidate samples align with the targeted task distribution. To mitigate redundancy, we incorporate a diversification mechanism that promotes coverage of complementary training samples. By interleaving alignment and diversification, EvoSelect enables progressive LLM evolution toward targeted tasks. Extensive experiments on various benchmarks demonstrate that with either weak or strong data generators, EvoSelect consistently improves adaptation efficacy over existing data selection methods.

arXiv

Remask, Don't Replace: Token-to-Mask Refinement in Diffusion Language Models

Lin Yao

Diffusion language models (dLLMs) generate text through iterative denoising, filling multiple masked positions at each step. Positions filled in the same step are predicted without conditioning on one another's newly filled values and can therefore be mutually inconsistent; once retained, these inconsistencies become context for later predictions. We introduce \emph{Token-to-Mask} (T2M), a training-free inference-time correction method that identifies low-confidence positions using the model's probability of the current token, remasks them, and reconstructs them in later denoising steps. On dLLMs equipped with correction mechanisms, a single T2M configuration transfers across tasks and models without retuning and broadly improves task metrics over each model's native correction mechanism. In controlled experiments, we decompose correction methods into a detector that identifies suspicious tokens and an action that determines how to revise them. Holding the detector fixed, remasking yields higher task metrics than replacement; across the tested detector--action combinations, current-token-probability detection paired with remasking performs best. Compared with direct editing, T2M converts additional inference compute into performance gains more effectively and, on most tasks, retains a sequential-step advantage over autoregressive token-by-token decoding.

arXiv

Doc-V*:Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA

Yuanlei Zheng, Pei Fu, Hang Li, Ziyang Wang, Yuyi Zhang, Wenyu Ruan, Xiaojin Zhang, Zhongyu Wei, Zhenbo Luo, Jian Luan, Wei Chen, Xiang Bai

Multi-page Document Visual Question Answering requires reasoning over semantics, layouts, and visual elements in long, visually dense documents. Existing OCR-free methods face a trade-off between capacity and precision: end-to-end models scale poorly with document length, while visual retrieval-based pipelines are brittle and passive. We propose Doc-$V^*$, an \textbf{OCR-free agentic} framework that casts multi-page DocVQA as sequential evidence aggregation. Doc-$V^*$ begins with a thumbnail overview, then actively navigates via semantic retrieval and targeted page fetching, and aggregates evidence in a structured working memory for grounded reasoning. Trained by imitation learning from expert trajectories and further optimized with Group Relative Policy Optimization, Doc-$V^*$ balances answer accuracy with evidence-seeking efficiency. Across five benchmarks, Doc-$V^*$ outperforms open-source baselines and approaches proprietary models, improving out-of-domain performance by up to \textbf{47.9\%} over RAG baseline. Other results reveal effective evidence aggregation with selective attention, not increased input pages.

arXiv

Disentangling Speaker Traits for Deepfake Source Verification via Chebyshev Polynomial and Riemannian Metric Learning

Xi Xuan, Wenxin Zhang, Zhiyu li, Jennifer Williams, Ville Hautamäki, Tomi H. Kinnunen

Speech deepfake source verification systems aims to determine whether two synthetic speech utterances originate from the same source generator, often assuming that the resulting source embeddings are independent of speaker traits. However, this assumption remains unverified. In this paper, we first investigate the impact of speaker factors on source verification. We propose a speaker-disentangled metric learning (SDML) framework incorporating two novel loss functions. The first leverages Chebyshev polynomial to mitigate gradient instability during disentanglement optimization. The second projects source and speaker embeddings into hyperbolic space, leveraging Riemannian metric distances to reduce speaker information and learn more discriminative source features. Experimental results on MLAAD benchmark, evaluated under four newly proposed protocols designed for source-speaker disentanglement scenarios, demonstrate the effectiveness of SDML framework. The code, evaluation protocols and demo website are available at https://github.com/xxuan-acoustics/RiemannSD-Net.

arXiv

When Contextual Inference Fails: Cancelability in Interactive Instruction Following

Natalia Bila, Kata Naszádi, Alexandra Mayn, Christof Monz

We investigate the separation of literal interpretation from contextual inference in a collaborative block-building tasks, where an agent must resolve underspecified instructions using context. We adapt an existing two-speaker psycholinguistic paradigm into an interactive benchmark called Build What I Mean (BWIM). This setup contrasts a pragmatically cooperative speaker with one who is only literally reliable. In BWIM, models face underspecified instructions and must choose between making a contextual inference or requesting clarification at a small communication cost. Evaluating several state-of-the-art LLMs, we find a clear dissociation between judgment and action. Although models successfully detect speaker unreliability in explicit confidence ratings, they fail to leverage this awareness when taking action. Instead of deploying efficient clarification strategies, models default to suboptimal behaviors. These include partner-blind over-clarification and question-averse guessing under uncertainty. BWIM provides a controlled environment to evaluate online partner adaptation and contextual reasoning in interactive settings.

arXiv

Towards Audio Token Compression in Large Audio Language Models

Saurabhchand Bhati, Samuel Thomas, Hilde Kuehne, Rogerio Feris, James Glass

Large Audio Language Models (LALMs) deliver strong performance across speech and audio tasks, but their audio encoders generate high-rate token sequences (e.g., 25 tokens/s), making attention computation costly and limiting scalability. In this paper, we explore techniques such as unsupervised segmentation, uniform average pooling, etc., to reduce the number of audio tokens before they are consumed by the LLM decoder. To mitigate potential performance degradation, we employ low-rank adapters during finetuning. We evaluate our proposed models on two tasks, automatic speech recognition and speech-to-speech translation tasks, that are dependent on effectively uncovering the underlying lexical content of the input signal, and study the effect of downsampling on these tasks. Experimental results show that compressed LALMs can achieve performance closer to frame-level LALMs while reducing the input audio token count up to three times before the LLM backbone.

arXiv

DiverValue-Bench: A Benchmark and Fine-Tuning Framework for Aligning Large Language Models with Diverse Human Values

Yao Liang, Dongcheng Zhao, Feifei Zhao, Guobin Shen, Yuwei Wang, Dongqi Liang, Yi Zeng

Aligning large language models (LLMs) with diverse human values is essential for safe and effective deployment, yet existing benchmarks often overlook cultural and demographic variation. We introduce DiverValue-Bench, a population-aware benchmark for evaluating multi-dimensional value alignment across 74 countries/regions. It contains 23,763 quality-controlled instances derived from PRISM user feedback and audited through large-scale human validation, with fine-grained value labels, personalized questions, contrastive reference answers, and rich demographic metadata. Using DiverValue-Bench, we evaluate representative LLMs and reveal substantial geographic and demographic disparities that are masked by aggregate performance. We further show that lightweight preference-based fine-tuning with Low-Rank Adaptation (LoRA) and Direct Preference Optimization (DPO) substantially improves in-domain value alignment while yielding consistent out-of-domain gains. These results highlight the need for population-aware alignment evaluation and demonstrate the utility of DiverValue-Bench as a practical foundation for global alignment, personalized value modeling, and equitable AI development.

arXiv

ContestTrade: A Multi-Agent Trading System Based on Internal Contest Mechanism

Li Zhao, Rui Sun, Zuoyou Jiang, Bo Yang, Yuxiao Bai, Mengting Chen, Jing Li, Zuo Bai

In financial trading, large language model (LLM)-based agents demonstrate significant potential, but their decisions can be sensitive to noisy and non-stationary market information. We propose ContestTrade, a multi-agent trading system with an internal competitive mechanism inspired by institutional investment workflows. The system consists of two specialized teams: (1) a Data Team that processes and condenses massive market data into diversified textual factors optimized for constrained LLM context windows, and (2) a Research Team that produces parallelized multipath trading decisions via tool-augmented deep research. The core design is a "Quantify-Predict-Allocate" contest mechanism within each team: agent outputs are scored only after market outcomes become observable, future utility is predicted from historical scores, and resources are allocated to agents with positive predicted utility. In a post-2024 A-share backtest, ContestTrade achieves higher backtested return and risk-adjusted performance than the evaluated baselines. We further describe the temporal protocol, implementation choices, and limitations to clarify the scope of these results.

arXiv

HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings

Rasmus T. Aavang, Giovanni Rizzi, Rasmus Tjalk-Bøggild, Alexandre Iolov, Mike Zhang, Johannes Bjerva

Accurate tagging of earnings reports can yield significant short-term returns for stakeholders. The machine-readable inline eXtensible Business Reporting Language (iXBRL) is mandated for public financial filings. Yet, its complex, fine-grained taxonomy limits the cross-company transferability of tagged Key Performance Indicators (KPIs). To address this, we introduce the Hierarchical Financial Key Performance Indicator (HiFi-KPI) dataset, a large-scale corpus of 1.65M paragraphs and 198k unique, hierarchically organized labels linked to iXBRL taxonomies. HiFi-KPI supports multiple tasks and we evaluate three: KPI classification, KPI extraction, and structured KPI extraction. For rapid evaluation, we also release HiFi-KPI-Lite, a manually curated 8K paragraph subset. Baselines on HiFi-KPI-Lite show that encoder-based models achieve over 0.906 macro-F1 on classification, while Large Language Models (LLMs) reach 0.440 F1 on structured extraction. Finally, a qualitative analysis reveals that extraction errors primarily relate to dates. We open-source all code and data at https://github.com/aaunlp/HiFi-KPI.

arXiv

Explainable Multimodal Depression Recognition in Clinical Interviews via PHQ-Aligned Symptom Summarization

Wenjie Zheng, Qiming Xie, Jianfei Yu, Yang Wang, Lei Cao, Fei Wang, Shijin Wang, Rui Xia, Chengqing Zong

Recent advances in multimodal depression recognition for clinical interviews (MDRC) have demonstrated the potential of AI systems by integrating textual, acoustic, and facial cues. However, existing methods pay limited attention to interpretability, thereby constraining reproducibility and clinician review. To address this, we introduce Explain-MDRC, an explainable MDRC framework that mirrors clinical workflows by generating structured symptom summaries from text and integrating them with nonverbal cues for recognition. Specifically, we construct Explain-DAIC, a dataset based on DAIC-WOZ and enriched with PHQ-8-aligned summary annotations, providing a foundation for developing models with built-in interpretability. We further propose PhqCML, a model that combines PHQ-8-aligned symptom summarization with PHQ-aware contrastive learning and summary-informed multimodal fusion. Automated metrics and expert evaluations show that Explain-MDRC improves recognition performance and provides more interpretable, clinician-readable intermediate evidence, suggesting a promising direction for transparent AI-assisted depression recognition research.

arXiv

Mitigating GenAI-Powered Evidence Pollution for Out-Of-Context Misinformation Detection

Zehong Yan, Peng Qi, Wynne Hsu, Mong Li Lee

While generative artificial intelligence (GenAI) models have achieved significant success, their misuse for generating deceptive content raises growing concerns about online information security. Out-of-context (OOC) multimodal misinformation detection systems typically rely on Web-retrieved evidence to identify images repurposed in false contexts, but they are increasingly challenged by the presence of GenAI-polluted evidence. Existing work mainly focuses on verifying claims that have undergone stylistic rewriting at the claim level and assume a clean evidence corpus. In this work, we remove this assumption and systematically study the impact of GenAI-driven evidence pollution threat on OOC detection. We show that polluted evidence can degrade the performance of state-of-the-art detectors by more than 9 percentage points. We propose two mitigating strategies, cross-modal evidence reranking and cross-modal claim-evidence reasoning, to address the challenge posed by polluted evidence. Extensive experiments on two benchmark datasets demonstrate that our approaches effectively enhance the robustness of existing OOC detectors amidst polluted evidence. The source code and data are publicly available at https://github.com/YanZehong/GenAI-Evidence-Pollution.

arXiv

Beyond Visual Similarity: Entity-Aligned Retrieval for Knowledge-Based Visual Question Answering

Hangrui Xu, Zhengxian Wu, Yunyao Yu, Zhuohong Chen, Rui Cong, Xiangwen Deng, Zhifang Liu, Peng Jiao, Haoqian Wang

Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, existing retrieval pipelines predominantly employ CLIP-style dual encoders, which prioritize surface-level visual similarity over entity-level semantic alignment. This paradigm often fails when semantically identical concepts exhibit large visual variations or when distinct entities appear visually similar. To address this, we propose KBMR, the first MLLM-based embedding retriever tailored for KB-VQA. Leveraging the robust autoregressive capabilities of MLLMs, KBMR maps images into a semantic space that better preserves concept identity. To tackle the challenge of noisy supervision in Wikipedia-scale retrieval, we introduce an MLLM-based semantic discriminator that generates continuous entity-consistency weights. These weights guide a novel continuous semantic distillation objective, enabling effective hard negative sampling and soft supervision beyond rigid binary labels. Extensive experiments demonstrate that KBMR significantly outperforms CLIP baselines, yielding up to a 14.7% improvement in retrieval Recall@1 and a 9.4% gain in end-to-end VQA accuracy. Code is available at https://github.com/realHarryX/KBMR.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.