2602.08145v1 Feb 04, 2026 cs.LG

신뢰성과 책임성을 갖춘 기초 모델: 종합적인 연구

Reliable and Responsible Foundation Models: A Comprehensive Survey

F. Tramèr
F. Tramèr
Citations: 1,072
h-index: 14
Huaxiu Yao
Huaxiu Yao
Citations: 59
h-index: 5
Rishi Bommasani
Rishi Bommasani
Citations: 14,179
h-index: 12
Elias Stengel-Eskin
Elias Stengel-Eskin
Citations: 891
h-index: 17
Huan Zhang
Huan Zhang
Citations: 134
h-index: 4
Zhaoyang Wang
Zhaoyang Wang
Citations: 129
h-index: 4
Y. Tsvetkov
Y. Tsvetkov
Citations: 330
h-index: 6
Jaehong Yoon
Jaehong Yoon
Citations: 190
h-index: 5
A. Bibi
A. Bibi
Citations: 174
h-index: 6
Xinyu Yang
Xinyu Yang
Citations: 294
h-index: 7
Mohit Bansal
Mohit Bansal
Citations: 7
h-index: 2
Philip Torr
Philip Torr
Citations: 513
h-index: 12
Wenjie Qu
Wenjie Qu
Citations: 146
h-index: 7
Kaidi Xu
Kaidi Xu
Citations: 27
h-index: 2
Junlin Han
Junlin Han
Citations: 4
h-index: 1
Jinqi Luo
Jinqi Luo
Citations: 11
h-index: 2
Wangchunshu Zhou
Wangchunshu Zhou
Citations: 115
h-index: 2
Xiyao Wang
Xiyao Wang
Citations: 26
h-index: 2
Shengbang Tong
Shengbang Tong
Citations: 35
h-index: 3
Lingfeng Shen
Lingfeng Shen
Citations: 40
h-index: 2
Rafael Rafailov
Rafael Rafailov
Citations: 13,596
h-index: 24
Runjia Li
Runjia Li
University of Oxford
Citations: 203
h-index: 7
Yiyang Zhou
Yiyang Zhou
Citations: 1,178
h-index: 13
Chenhang Cui
Chenhang Cui
Citations: 907
h-index: 9
Yu Wang
Yu Wang
Citations: 4
h-index: 1
Wen-qing Zheng
Wen-qing Zheng
Citations: 1
h-index: 1
Huichi Zhou
Huichi Zhou
Citations: 10
h-index: 2
Jindong Gu
Jindong Gu
Citations: 66
h-index: 6
Zhaorun Chen
Zhaorun Chen
Citations: 39
h-index: 3
Peng Xia
Peng Xia
Citations: 130
h-index: 4
Tony Lee
Tony Lee
Citations: 2
h-index: 1
Thomas M. Zollo
Thomas M. Zollo
Citations: 2
h-index: 1
Vikash Sehwag
Vikash Sehwag
Citations: 16
h-index: 3
Jixuan Leng
Jixuan Leng
Citations: 5
h-index: 2
Jiuhai Chen
Jiuhai Chen
Citations: 9
h-index: 2
Yuxin Wen
Yuxin Wen
Citations: 48
h-index: 4
Zhun Deng
Zhun Deng
Citations: 2
h-index: 1
Linjun Zhang
Linjun Zhang
Citations: 26
h-index: 1
Pavel Izmailov
Pavel Izmailov
Citations: 2
h-index: 1
Pang Wei Koh
Pang Wei Koh
Citations: 6
h-index: 2
A. Wilson
A. Wilson
Citations: 8
h-index: 2
Jiaheng Zhang
Jiaheng Zhang
Citations: 0
h-index: 0
James Zou
James Zou
Citations: 51
h-index: 2
Cihang Xie
Cihang Xie
Citations: 191
h-index: 6
Hao Wang
Hao Wang
Citations: 11
h-index: 2
Julian McAuley
Julian McAuley
Citations: 454
h-index: 9
David Alvarez-Melis
David Alvarez-Melis
Citations: 103
h-index: 6
Suman Jana
Suman Jana
Citations: 119
h-index: 3
Chris Callison-Burch
Chris Callison-Burch
Citations: 50
h-index: 3
René Vidal
René Vidal
Citations: 185
h-index: 3
Filippos Kokkinos
Filippos Kokkinos
Citations: 298
h-index: 8
Beidi Chen
Beidi Chen
Citations: 65
h-index: 4

대규모 언어 모델(LLM), 다중 모달 대규모 언어 모델(MLLM), 이미지 생성 모델(텍스트-이미지 모델 및 이미지 편집 모델), 비디오 생성 모델과 같은 기초 모델은 법률, 의학, 교육, 금융, 과학 등 다양한 분야에서 광범위하게 활용되는 필수적인 도구로 자리 잡았습니다. 이러한 모델이 실제 환경에 점점 더 많이 적용됨에 따라, 학계, 산업계 및 정부는 이러한 모델의 신뢰성과 책임성을 확보하는 것이 매우 중요해졌습니다. 본 연구는 기초 모델의 신뢰성과 책임성 있는 개발에 대한 종합적인 개요를 제공합니다. 우리는 편향 및 공정성, 보안 및 개인 정보 보호, 불확실성, 설명 가능성, 데이터 분포 변화 등 중요한 문제들을 탐구합니다. 또한, 환각 현상과 같은 모델의 한계점과 정렬(alignment) 및 인공지능 생성 콘텐츠(AIGC) 탐지 방법 등을 다룹니다. 각 영역별로 현재 연구 동향을 검토하고 구체적인 향후 연구 방향을 제시합니다. 또한, 이러한 영역들 간의 상호 연관성을 논의하며, 공통적인 과제들을 강조합니다. 본 연구가 강력하면서도 윤리적이고, 신뢰할 수 있으며, 안정적이고, 사회적으로 책임 있는 기초 모델 개발에 기여하기를 바랍니다.

Original Abstract

Foundation models, including Large Language Models (LLMs), Multimodal Large Language Models (MLLMs), Image Generative Models (i.e, Text-to-Image Models and Image-Editing Models), and Video Generative Models, have become essential tools with broad applications across various domains such as law, medicine, education, finance, science, and beyond. As these models see increasing real-world deployment, ensuring their reliability and responsibility has become critical for academia, industry, and government. This survey addresses the reliable and responsible development of foundation models. We explore critical issues, including bias and fairness, security and privacy, uncertainty, explainability, and distribution shift. Our research also covers model limitations, such as hallucinations, as well as methods like alignment and Artificial Intelligence-Generated Content (AIGC) detection. For each area, we review the current state of the field and outline concrete future research directions. Additionally, we discuss the intersections between these areas, highlighting their connections and shared challenges. We hope our survey fosters the development of foundation models that are not only powerful but also ethical, trustworthy, reliable, and socially responsible.

0 Citations
0 Influential
12 Altmetric
60.0 Score

No Analysis Report Yet

This paper hasn't been analyzed by Gemini yet.

댓글

댓글을 작성하려면 로그인하세요.

아직 댓글이 없습니다. 첫 번째 댓글을 남겨보세요!