logo
科言猫学术研究的AI总结
language
手机端手机端
  • Phone Feature微信小程序
简体中文简体中文
notice消息通知
博客博客
delete
arrow
返回

ECD: Efficient Contrastive Decoding with Probabilistic Hallucination Detection

delete2026-01-01
delete0
PRE
AI
L
Laura Fieback *
N
Nishilkumar Balar
J
Jakob Spiegelberg
H
Hanno Gottschalk
DOI：10.1007/978-3-032-06109-6_2delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
尽管大型视觉语言模型（LVLMs）取得了近期进展，但这些模型仍存在生成与所提供视觉输入不符的幻觉性响应。为缓解此类幻觉，我们提出了一种高效对比解码（ECD）方法，该方法利用概率性幻觉检测，在推理阶段将输出分布转向上下文准确的答案。通过对比标记概率和幻觉分数，ECD从原始分布中减去幻觉概念，有效抑制幻觉。值得注意的是，我们提出的方法可应用于任何开源LVLM，且无需额外训练LVLM。我们在多个基准数据集和不同LVLM上评估了该方法。实验表明，ECD有效缓解了幻觉问题，在LVLM基准性能和计算时间方面优于现有最优方法。
Keyword:
Multimodal large language models
Contrastive decoding
Hallucination mitigation
Hallucination detection
Meta classification

期刊

M
MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES. RESEARCH TRACK, ECML PKDD 2025, PT VII
IF:
0
论文数:
24
被引数:
0

机构

V
volkswagen
学者数:
806
论文数: 557
被引数: 2
V
volkswagen germany
学者数:
277
论文数: 236
被引数: 1
引用论文

引用论文

LLaVA-MR: Large Language-and-Vision Assistant for Video Moment Retrieval
err
IF0
err2024-12-05
err0
errOAAI
errJian Li
err分享
err收藏
Evaluating Object Hallucination in Large Vision-Language Models
errProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing
IF0
err2023-01-01
err0
errOAAI
errYifan Li; Yifan Du; Kun Zhou; Jinpeng Wang; Xin Zhao; Ji-Rong Wen
err分享
err收藏
Temporal Performance Prediction for Deep Convolutional Long Short-Term Memory Networks
errLecture Notes in Computer Science
IF0
err2023-01-01
err0
PREAI
errFieback,Laura; Dash,Bidya; Spiegelberg,Jakob; Gottschalk,Hanno
err分享
err收藏
GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering
err2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2019-06-01
err0
errOAAI
errDrew A. Hudson; Christopher D. Manning
err分享
err收藏
Prediction Error Meta Classification in Semantic Segmentation: Detection via Aggregated Dispersion Measures of Softmax Probabilities
err2020 International Joint Conference on Neural Networks (IJCNN)
IF0
err2020-07-01
err0
errOAAI
errMatthias Rottmann; Pascal Colling; Thomas Paul Hack; Robin Chan; Fabian Huger; Peter Schlicht; Hanno Gottschalk
err分享
err收藏
The relationship between Precision-Recall and ROC curves精确召回率与ROC曲线的关系
errProceedings of the 23rd international conference on Machine learning - ICML '06
IF0
err2006-01-01
err0
errOAAI
errJesse Davis; Mark Goadrich
err分享
err收藏
Overcoming language priors with self-contrastive learning for visual question answering通过自我对比学习克服语言先验知识以进行视觉问答
errMULTIMEDIA TOOLS AND APPLICATIONS
IF3
err2022-11-10
err6
PREAI
errYan, Hong; Liu, Lijun; Feng, Xupeng; Huang, Qingsong
err分享
err收藏
Contrastive Decoding: Open-ended Text Generation as Optimization
errProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
IF0
err2023-01-01
err0
errOAAI
errXiang Lisa Li; Ari Holtzman; Daniel Fried; Percy Liang; Jason Eisner; Tatsunori Hashimoto; Luke Zettlemoyer; Mike Lewis
err分享
err收藏
Logical Closed Loop: Uncovering Object Hallucinations in Large Vision-Language Models逻辑闭环：揭示大型视觉语言模型中的对象幻觉
errFindings of the Association for Computational Linguistics ACL 2024
IF0
err2024-01-01
err0
PREAI
errJunfei Wu; Qiang Liu; Ding Wang; Jinghao Zhang; Shu Wu; Liang Wang; Tieniu Tan
err分享
err收藏
Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding通过视觉对比解码减轻大型视觉语言模型中的物体幻觉
err2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
IF0
err2024-06-16
err0
PREAI
errSicong Leng; Hang Zhang; Guanzheng Chen; Xin Li; Shijian Lu; Chunyan Miao; Lidong Bing
err分享
err收藏
学者 查看更多内容