paperlist.ai

AI 顶会论文

AI 顶会论文库

检索 30+ 个顶会的论文，包括 NeurIPS、ICLR、ICML、CVPR、ACL、AAAI 等。输入关键词筛选，打开任意一篇查看摘要、作者和链接。

arXiv

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

Jeonghye Kim, Xufang Luo, Minbeom Kim, Sangmook Lee, Dohyung Kim, Jiwon Jeon, Dongsheng Li, Yuqing Yang

Self-distillation has emerged as an effective post-training paradigm for LLMs, often improving performance while shortening reasoning traces. However, in mathematical reasoning, we find that it can reduce response length while degrading performance. We trace this degradation to the suppression of epistemic verbalization - the model's expression of uncertainty during reasoning. Through controlled experiments varying conditioning context richness and task coverage, we show that conditioning the teacher on rich information suppresses uncertainty expression, enabling rapid in-domain optimization with limited task coverage but harming OOD performance, where unseen problems benefit from expressing uncertainty and adjusting accordingly. Across Qwen3-1.7B/8B, DeepSeek-Distill-Qwen-7B, and Olmo3-7B-Instruct, we observe performance drops of up to 40%. Our findings highlight that exposing appropriate levels of uncertainty is crucial for robust reasoning and underscore the importance of optimizing reasoning behavior beyond merely reinforcing correct answer traces.

arXiv

Parametric Knowledge in RAG-SFT for Domain-Specific Document Generation

Julian Oestreich, Maximilian Bley, Frank Binder, Lydia Müller, André Alcalde, Maksym Sydorenkoq

Retrieval-Augmented Generation (RAG) fine-tuning has shown substantial improvements over vanilla RAG, yet most studies target document question answering, leaving open whether these gains transfer to specialized tasks. We study supervised RAG fine-tuning (RAG-SFT) for requirements document generation in the electronics engineering domain, adapting two 7B models under two different training data strategies. Because Rouge and BertScore poorly capture factuality on long technical text, we introduce C-FEX, a claim-based evaluation pipeline that attributes each response claim to its origin (augmented prompt or reference response), and propose Parametric Knowledge Precision (PKP), which isolates claims originating from the model's weights and measures their correctness. We show that a prior metric to assess parametric knowledge decomposes as PKP $\times$ PR, separating the rate of parametric output (PR) from its quality (PKP). Empirically, fine-tuned 7B models match or exceed a 72B baseline; standard metrics disagree with claim-based factuality and can mislead about fine-tuning gains; and, fine-tuning does not reinforce correct parametric knowledge but suppresses hallucination---models speak from their weights less often but far more reliably.

arXiv

N-gram-like Language Models Predict Naturalistic Reading Time Best

James A. Michaelov, Roger P. Levy

Recent work has found that contemporary language models such as transformers can become so good at next-word prediction that the probabilities they calculate become worse for predicting naturalistic reading time. In this paper, we propose that this can be explained by reading time being shaped by simple n-gram statistics rather than the more complex statistics learned by state-of-the-art transformer language models. We demonstrate that the neural language models whose predictions are most correlated with n-gram probability are also those that calculate probabilities that are the most correlated with eye-tracking-based metrics of reading time on naturalistic text.

arXiv

Speak in Context: Multilingual ASR with Speech Context Alignment via Contrastive Learning

Yuchen Zhang, Haralambos Mouratidis, Ravi Shekhar

Automatic speech recognition (ASR) has benefited from advances in pretrained speech and language models, yet most systems remain constrained to monolingual settings and short, isolated utterances. While recent efforts in context-aware ASR show promise, two key challenges persist: limited multilingual support and the absence of principled alignment between speech and contextual representations. In this paper, we introduce a context-aware multilingual ASR framework that supports diverse languages and accents while preserving the modularity of pretrained models. Our approach combines a frozen speech encoder and a decoder-only language model via a lightweight projection module, allowing structured context prompts, including dialogue history and biasing words, to guide transcription. To improve interaction between speech and context, we employ a contrastive learning objective that aligns their representations in a shared embedding space. Evaluations on over 1,500 hours of real-world conversational speech across 11 languages and 5 English dialects show that contextual input consistently improves recognition quality. Contrastive alignment provides additional gains when applied to different context types, with an overall performance gain of over 5%. These results highlight the importance of both contextual modeling and cross-modal alignment in multilingual ASR.

arXiv

Eval4Sim: An Evaluation Framework for Persona Simulation

Eliseo Bao, Anxo Perez, Javier Parapar, Xi Wang

Large Language Model personas, explicit profiles specifying a user's attributes, preferences, and behavioural tendencies, are increasingly used to simulate human conversations for user modelling, social reasoning, and behavioural analysis. Evaluating whether such simulations faithfully reflect human conversational behaviour is critical, yet current practice often relies on LLM-as-a-judge approaches that provide limited grounding in observable behaviour and produce opaque scalar scores. We present Eval4Sim, an evaluation framework that measures alignment between simulated and human conversations across three dimensions: adherence, whether persona traits are recoverable from dialogue via dense retrieval; consistency, whether a persona maintains a distinguishable stylistic identity via authorship verification; and naturalness, whether conversations exhibit human-like turn-to-turn flow via dialogue NLI. Unlike optimization-oriented metrics, each dimension takes a human corpus as a reference baseline and penalizes deviations in both directions, distinguishing insufficient persona encoding from over-optimized, unnatural behaviour. The framework is corpus-agnostic: any persona-annotated conversational dataset can serve as the reference. Evaluated over ten simulation corpora, Eval4Sim surfaces systematic trade-offs invisible to single-score methods.

arXiv

The Emergence of Lab-Driven Alignment Signatures: A Psychometric Framework for Auditing Latent Bias and Compounding Risk in Generative AI

Dusan Bosnjakovic

Large language models increasingly serve as reasoning layers in multi-agent systems, where one provider's models may generate, judge, and summarize within a single pipeline. This raises the question of whether developer organizations impart durable behavioral tendencies that could compound across such stacks. We apply a scenario-based forced-choice instrument to 18 governance-relevant behavioral dimensions across 18 models from six developer organizations. Items are model-generated, filtered by independent judges, and administered with probe blanks embedded among semantically orthogonal decoys under deterministic option shuffling. Findings are declared on effect size, requiring both Holm-corrected significance and |d| >= 0.2. Across the 14 dimensions on which one scale pole denotes a defined response failure -- sycophancy, false balance, overconfidence, and others -- organizations occupy consistent relative positions (Kendall's W = 0.527, p = 6e-6), with Anthropic ranking first or second on 13 of 14 and Meta fifth on 10 of 14. On dimensions measuring directional valence without a normatively correct pole, no such concordance appears (W = 0.289, p = 0.48): organizations differ reliably in resistance to defined failures, not in ideological lean. Model-level variation within an organization is comparable in magnitude to variation between organizations, and is reported in full. Secondarily, across four major-version transitions, later generations scored lower on deficiency-poled dimensions in 26 of 30 comparisons. All 18 dimensions are reported, including three showing no organization-level differences, and a controlled test of the decoy manipulation returns a null result.

arXiv

Language Family Matters: Evaluating LLM-Based ASR Across Linguistic Boundaries

Yuchen Zhang, Ravi Shekhar, Haralambos Mouratidis

Large Language Model (LLM)-powered Automatic Speech Recognition (ASR) systems achieve strong performance with limited resources by linking a frozen speech encoder to a pretrained LLM via a lightweight connector. Prior work trains a separate connector per language, overlooking linguistic relatedness. We propose an efficient and novel connector-sharing strategy based on linguistic family membership, enabling one connector per family, and empirically validate its effectiveness across two multilingual LLMs and two real-world corpora spanning curated and crowd-sourced speech. Our results show that family-based connectors reduce parameter count while improving generalization across domains, offering a practical and scalable strategy for multilingual ASR deployment.

arXiv

Measuring Narrative Polarization in Online Discourse

Jan Elfes, Marco Bastos, Luca Maria Aiello

Polarization research has demonstrated how people cluster in homogeneous groups with opposing opinions. However, this effect emerges not only through interaction between people, limiting communication between groups, but also between narratives, shaping opinions and partisan identities. Yet, how polarized information environments portray opposing interpretations of reality, and whether narratives move between content environments despite limited interactions, remains unexplored. To address this gap, we formalize the concept of narrative polarization and demonstrate its measurement in 212 YouTube videos and 90,029 comments on the Israeli-Palestinian conflict. Based on structural narrative theory and implemented through a large language model, we extract the narrative roles assigned to central actors in two partisan information environments. We find that while videos produce highly polarized narratives, comments exhibit significantly lower levels of narrative polarization, converging on shared core narrative structures. However, recurring narrative motifs capturing more complex actor constellations reveal additional differences between partisan environments.

arXiv

Dripper: Token-Efficient Main HTML Extraction with a Lightweight LM

Mengjie Liu, Jiahui Peng, Wenchang Ning, Pei Chu, Jiantao Qiu, Ren Ma, He Zhu, Rui Min, Lindong Lu, Linfeng Hou, Kaiwen Liu, Yuan Qu, Zhenxiang Li, Chao Xu, Zhongying Tu, Wentao Zhang, Conghui He

High-quality main content extraction from web pages is a critical prerequisite for constructing large-scale training corpora. While traditional heuristic extractors are efficient, they lack the semantic reasoning required to handle the structural heterogeneity of the modern web. Conversely, well-pretrained generative Large Language Models (LLMs) offer superior document comprehension but are prohibited by excessive computational costs, limited context windows, and hallucination risks when applied at web scale. We present \textbf{Dripper}, a lightweight framework that resolves these bottlenecks through four contributions: (1) We reformulate extraction as a \textbf{constrained sequence labeling} task using SLMs (Small Language Models). This paradigm eliminates generative hallucinations and achieves exceptional efficiency, reaching a throughput of 3.08 pages per second on a single A100 GPU. (2) We construct \textbf{WebMainBench}, a rigorous benchmark of 7,809 human-annotated pages covering 5,434 unique domains and multiple languages. Evaluations show our Dripper-0.6B model \textbf{outperforms} heuristics like Trafilatura and rivals massive models like DeepSeek-V3.2(685B), GPT-5 and Gemini-2.5-Pro, offering an optimal efficiency-accuracy trade-off. (3) We demonstrate infrastructural value by \textbf{pre-training a 1B model} on a Dripper-curated corpus (63B tokens). This model significantly outperforms baselines in downstream tasks, proving the critical role of extraction quality and the effectiveness of our framework. (4) We \textbf{open-source} the Dripper-0.6B weights and codebase to facilitate the construction of high-quality datasets.

arXiv

SimulRAG: Simulator-based RAG for Grounding LLMs in Long-form Scientific QA

Haozhou Xu, Dongxia Wu, Matteo Chinazzi, Ruijia Niu, Rose Yu, Yi-An Ma

Large Language Models (LLMs) show promise in generating long-form scientific explanations that synthesize evidence and connect multiple factors. However, in long-form scientific question answering, LLMs often hallucinate, producing unsupported or inconsistent claims. Retrieval-Augmented Generation (RAG) improves trustworthiness by grounding generation in external sources; scientific simulators are valuable because they can validate quantitative hypotheses and capture evolving dynamics. Yet simulation-based RAG is non-trivial due to two challenges: how to retrieve from scientific simulators, and how to efficiently verify and update long-form answers. To overcome these challenges, we propose SimulRAG, a simulator-based RAG framework with a generalized retrieval interface that translates between text and simulator parameters/outputs. SimulRAG further introduces claim-level generation with uncertainty estimation and simulator boundary assessment (UE+SBA) to selectively verify and update claims. Unlike tool-first or holistic answer revision, it first elicits diverse answers without retrieval and then grounds uncertain, simulator-verifiable atomic claims with simulator evidence. We also release a long-form scientific QA benchmark spanning climate science, epidemiology, and urban planning, with ground truth verified by simulations and human annotators. Experiments show SimulRAG improves informativeness by 30.4% and factuality by 16.3% over the strongest adapted RAG baselines, while UE+SBA enhances claim-level efficiency and quality.

arXiv

Marco-Voice Technical Report

Fengping Tian, Chenyang Lyu, Xuanfan Ni, Haoqin Sun, Qingjuan Li, Zhiqiang Qian, Haijun Li, Longyue Wang, Zhao Xu, Weihua Luo, Kaifu Zhang

This paper presents a multifunctional speech synthesis system that integrates voice cloning and emotion control speech synthesis within a unified framework. The goal of this work is to address longstanding challenges in achieving highly expressive, controllable, and natural speech generation that faithfully preserves speaker identity across diverse linguistic and emotional contexts. Our approach introduces an effective speaker-emotion disentanglement mechanism with in-batch contrastive learning, enabling independent manipulation of speaker identity and eemotional style, as well as rotational emotional embedding integration method for smooth emotion control. To support comprehensive training and evaluation, we construct CSEMOTIONS, a high-quality emotional speech dataset containing 10 hours of Mandarin speech from ten professional speakers across seven emotional categories. Extensive experiments demonstrate that our system, Marco-Voice, achieves substantial improvements in both objective and subjective metrics. Comprehensive evaluations and analysis were conducted, results show that MarcoVoice delivers competitive performance in terms of speech clarity and emotional richness, representing a substantial advance in the field of expressive neural speech synthesis. Our code and dataset are publicly available at https://github.com/AIDC-AI/Marco-Voice and https://huggingface.co/datasets/AIDC-AI/CSEMOTIONS respectively.

arXiv

SAKE: Structured Agentic Knowledge Extrapolation for Complex LLM Reasoning via Reinforcement Learning

Jiashu He, Jinxuan Fan, Bowen Jiang, Ignacio Hounie, Dan Roth, Alejandro Ribeiro

Knowledge extrapolation is the process of inferring novel information by combining and extending existing knowledge that is explicitly available. It is essential for solving complex questions in specialized domains where retrieving comprehensive external knowledge is impractical. We propose SAKE (Structured Agentic Knowledge Extrapolation), a RL powered agentic framework that trains LLMs to autonomously retrieve and extrapolate structured knowledge through tool-augmented reinforcement learning. SAKE defines two external KG tools: entity group construction and cross-group triplet retrieval. The model learns to interleave these 2 retrieval tools during a three-turn rollout: extracting key entities, filtering relevant concept groups, and associative reasoning by constructing new triplets through analogy. The entire pipeline is optimized end-to-end with GRPO using a reward that combines output format and answer correctness, teaching the model what to retrieve and how to reason over it. Our experiments proved that SAKE fine-tuned Qwen2.5-7B model surpasses GPT-3.5-Turbo with state-of-the-art agentic KG reasoning on both biomedical (75.4% vs. 70.1%) and commonsense (81.3% vs. 74.7%) benchmarks, while reducing token usage by over 90%. These results demonstrate that associative reasoning over incomplete structured knowledge does not require large models with complex, multi-step prompting, thus can be learned end-to-end by small, open-weight models through reinforcement learning with the right tools and training signal. Our code is available at https://github.com/jxfan99/SAKE.

arXiv

Evidence of conceptual mastery in the application of rules by Large Language Models

José Luiz Nunes, Guilherme FCF Almeida, Brian Flanagan

Background. Evidence that large language models (LLMs) reproduce human judgments does not establish conceptual mastery: the correspondence may reflect memorisation or be sensitivite to incidental task features. Objective. Across five experiments, we test whether 13 LLMs possess a generalisable competence in applying rules, including cases in which a rule's text and purpose point towards different outcomes. Method. Study 1A compared LLM judgments with newly collected human data on published stimuli and matched vignettes created after the models' training cut-offs. Studies 2A/2B compared responses to time-pressure instructions, a manipulation with a mechanistic route to human judgment blocked for LLMs. Study 3 varied reasoning effort, as an analogue for time constrained human judgements. Studies 1B/2B alsovaried system prompt wording and numerical scale anchors. Results LLM judgments closely tracked human judgments for both stimulus sets, while responding in the same unanticipated purposivist direction in the new set as humans did. Sensitivity to text and purpose was robust across prompt variations. Responses to time-pressure instructions were model-specific, suggesting a distinction between conceptual competence and human alignment. Replication of the human pattern was most apparent in models with fewer parameters, and these effects were susceptible to prompt variation. Increasing reasoning effort produced no detectable change in rule application for most models though a significant purposivist trend was observed in higher-effort for GPT-oss and Claude Sonnet 5. Response variance remained lower for LLMs than humans despite our per-model temperature calibration to match human sample variance. Conclusions. Overall, the findings suggest that LLM rule application reflects a generalisable, standing semantic competence that does not typically depend on expanded deliberation.

arXiv

MCTS-KBQA: Monte Carlo Tree Search with Information Gain Rewards for Knowledge Base Question Answering

Guanming Xiong, Haochen Li, Zonghong Dai, Liqiang Wen, Wen Zhao

This work investigates how to improve large language model (LLM)-based reasoning for knowledge base question answering (KBQA) via Monte Carlo Tree Search (MCTS). Applying MCTS to LLM-based KBQA remains challenging because reward design is difficult and rollout-based search is computationally expensive. Existing MCTS-style methods either rely on direct LLM scoring or require substantial data to train separate reward models, and they often provide rewards only at terminal states. To address these limitations, we propose Fast MCTS, which replaces terminal rollouts with an information gain (IG) reward for intermediate states. The IG reward is implemented as a question-conditioned PPL-ratio proxy over sanitized interaction histories, computed by forward passes of an open-source instruction LLM without additional reward-model training. Experiments on four KBQA benchmarks show that Fast MCTS consistently outperforms linear baselines and generally improves the accuracy-cost trade-off relative to rollout-based Classic MCTS. Code and data are available at https://github.com/JimXiongGM/MCTS-KBQA.

arXiv

Bactrainus: Optimizing Large Language Models for Multi-hop Complex Question Answering Tasks

Iman Barati, Arash Ghafouri, Behrouz Minaei-Bidgoli

Multi-hop question answering requires a system to identify and integrate evidence distributed across documents, yet large language models remain vulnerable to irrelevant context. We investigate this evidence bottleneck in the English HotpotQA distractor setting and introduce Bactrainus, a modular selector-reader framework that separates paragraph selection, supporting-sentence identification, and answer generation. Optional question decomposition and teacher-generated rationale supervision make it possible to test where additional reasoning structure is useful. The evaluation combines foundation-model screening, controlled context and prompting ablations, parameter-efficient adaptation of Llama 3.1 8B Instruct and Llama 3.1 70B Instruct readers, and integrated selector-reader experiments. Supplying the full candidate context instead of gold supporting facts reduces answer token-overlap F1 by 17-21 points, showing that scale alone does not remove context sensitivity. The largest observed differences are associated with reader adaptation and sentence-level evidence control. The strongest reported configuration obtains 89.01 answer F1 and 79.70 joint F1, whereas decomposition and rationale-supervision variants yield smaller, recipe-dependent changes. These findings support auditable, explicitly supervised evidence interfaces for fixed-candidate multi-hop QA and motivate blind, matched, multi-seed evaluation of the remaining small differences.

arXiv

Quantum Large Language Models via Tensor Network Disentanglers

Borja Aizpurua, Fernando Loren, Saeed S. Jahromi, Sukhbinder Singh, Roman Orus

We introduce a framework for seamlessly integrating quantum computing into pretrained large language models (LLMs). The key idea is to construct a hybrid quantum-classical representation that exactly reproduces the original model, providing a principled starting point from which quantum resources can only improve performance. Our approach replaces the weight matrices in self-attention and multilayer perceptron layers with two variational quantum circuits coupled to a matrix product operator (MPO). Tensor network disentanglers transfer much of each layer's information into the quantum circuits, enabling the remaining tensor network to be compressed to a bond-dimension-one MPO with over three orders of magnitude fewer classical parameters (in our experiments, from 110,592 to approximately 36 for the replaced layer) and less than a 0.3\% increase in perplexity. Training an added unitary adapter on top of this representation then surpasses the original model, reducing perplexity by up to 1.6\%. Finally, we validate the hybrid architecture on a real quantum processor, demonstrating a practical route towards quantum-enhanced language models.

arXiv

Assessing AI-Generated vs. Human-Authored Spear Phishing SMS Attacks: An Empirical Study

Jerson Francia, Derek Hansen, Benjamin Schooley, Matthew Taylor, Shydra Valynn Murray, Rebekah Cornelius, Greg Snow

Personalized phishing is difficult to defend against because messages can be tailored to a target's work, interests, and social context. Large language models may make such tailoring faster and easier, but it remains unclear whether messages produced from simple prompts are more convincing than those written by people. This 25-target pilot study compared personalized smishing messages generated by GPT-4 with messages written by novice student authors working under time constraints. Using the proposed Threshold Ranking Approach for Personalized Deception (TRAPD), participants ranked 12 messages written for them, indicated the point at which they would intend to click, explained their reasoning, and judged whether each message was authored by GPT-4 or a human. GPT-4-generated messages elicited an intention to click more often than student-authored messages (28% versus 21%), although the difference was uncertain. More broadly, our findings suggest that a simple prompt can produce personalized messages that participants found comparably convincing within the uncertainty of this pilot study. Job-related messages were significantly more likely to elicit an intention to click than hobby- or social-media-related messages. When asked whether a message was written by a human or generated by AI, participants identified the source no more accurately than chance, although the two study-specific message sets remained computationally distinguishable based on their text. Together, these findings suggest that accessible AI-assisted personalization may increase the practical scale of social-engineering threats, while also demonstrating both the value and current limitations of TRAPD for controlled and ethical comparison.

arXiv

From Monte Carlo to neural networks approximations of boundary value problems

Lucian Beznea, Iulian Cimpean, Oana Lupascu-Stamate, Ionel Popescu, Arghir Zarnescu

In this paper we study probabilistic and neural network approximations for solutions to Poisson equation subject to Holder data in general bounded domains of $\mathbb{R}^d$. We aim at two fundamental goals. The first, and the most important, we show that the solution to Poisson equation can be numerically approximated in the sup-norm by Monte Carlo methods, and that this can be done highly efficiently if we use a modified version of the walk on spheres algorithm as an acceleration method. This provides estimates which are efficient with respect to the prescribed approximation error and with polynomial complexity in the dimension and the reciprocal of the error. A crucial feature is that the overall number of samples does not not depend on the point at which the approximation is performed. As a second goal, we show that the obtained Monte Carlo solver renders in a constructive way ReLU deep neural network (DNN) solutions to Poisson problem, whose sizes depend at most polynomialy in the dimension $d$ and in the desired error. In fact we show that the random DNN provides with high probability a small approximation error and low polynomial complexity in the dimension.

arXiv

Fidelity Preference, Not Demographic Preference: A Pixel-Level Attribute-Sensitivity Audit of Image Aesthetic/Preference Scorers

Mingyang Xu

Text-to-image systems use learned aesthetic scorers to filter training data and guide generation, but whether these scores encode demographic attributes as objective quality is unclear. We audit four scorers (LAION-Aesthetics, PickScore, ImageReward, HPSv2) using pixel-level interventions on skin tone and body type in synthetic and real images. Our key finding is that along skin-lightness, the dominant effect is fidelity preference: unaltered images score highest, and perturbations in either direction are penalized (inverted-U). Placebo arms show this penalty is not an artifact of the skin operator, as applying the same CIELAB L* shift to non-skin regions yields similar penalty magnitudes. However, the penalty is operator-dependent and holds for all operators only for LAION-Aes. Critically, audits on synthetic images alone are misleading: LAION-Aes shows strong preference for darker skin on synthetic faces, but on 1470 real faces the preference reverses and becomes much smaller, and amplification becomes non-significant. Across scorers, synthetic results do not transfer -- reversing for LAION-Aes and HPSv2, attenuating for PickScore. We contribute a reproducible benchmark with artifact control and synthetic/real cross-validation, and an auditability criterion for pixel-level causal isolation (valid for skin tone, not for body type due to deformation). Population-stratified analysis shows fidelity-penalty asymmetry is not robust across groups after FDR correction except for HPSv2. Our findings show naive synthetic audits misjudge bias direction and magnitude, and only within-image causal isolation on real data can distinguish true demographic bias from fidelity preference.

arXiv

Boosting Knowledge-based Visual Question Answering with Structured Context Reasoning

Qiyou Liu, Yong Zhang, Jianjie Luo, Zhenguo Yang, Yi Yu

Knowledge-based Visual Question Answering aims to answer questions about an image by integrating external knowledge with visual and textual information. Recent approaches often rely on in-context learning to prompt Large Language Models (LLMs) with multimodal context in a zero-shot or few-shot manner. However, we observe that directly concatenating heterogeneous visual descriptions and retrieved knowledge into long, unstructured prompts often degrades reasoning performance, due to both excessive irrelevant context and the lack of explicit relational structure. In this paper, we propose an LLM-based Structured Context Reasoning (SCoRe) framework that infers both explicit and implicit relationships for prediction. SCoRe consists of three stages: Context Acquisition, which generates diverse visual notes and retrieves explicit knowledge via an efficient two-stage multimodal retrieval strategy; Context Selection, which filters relevant visual, explicit, and implicit knowledge using LLM-guided selection; and Context Compression, which performs Relational Logic Distillation (RLD) to transform raw text into explicit entity-relation triplets. These relational triplets serve as a concise and structured prompt for final answer prediction. Extensive experiments on the OK-VQA and A-OKVQA benchmarks demonstrate that SCoRe consistently outperforms state-of-the-art methods.

继续浏览

更多 AI 顶会论文

30+ 个顶会，按会议、年份和主题浏览。

浏览全部会议

进入任一会议和年份，查看完整论文列表。

全部会议 →

搜索论文

按标题、作者或关键词，在 21 万+ 篇论文里搜索。

搜索论文 →