基础模型推理综述
人工智能
2024-01-26 v5 计算与语言
计算机视觉与模式识别
机器学习
摘要
推理作为复杂问题解决的关键能力，在谈判、医疗诊断和刑事调查等各种现实场景中发挥着举足轻重的作用。它是通用人工智能（AGI）领域的一项基础方法论。随着基础模型（如大语言模型，LLM）的不断发展，人们对其在推理任务中能力的探索兴趣日益增长。在本文中，我们介绍了为推理提出或可适配的奠基性基础模型，重点突出了各种推理任务、方法和基准测试的最新进展。随后，我们深入探讨了基础模型中推理能力涌现背后的潜在未来方向。我们还讨论了多模态学习、自主智能体和超级对齐在推理背景下的相关性。通过讨论这些未来研究方向，我们希望激励研究人员对该领域的探索，推动基础模型推理的进一步发展，并为AGI的发展做出贡献。
引用
@article{arxiv.2312.11562,
title = {A Survey of Reasoning with Foundation Models},
author = {Jiankai Sun and Chuanyang Zheng and Enze Xie and Zhengying Liu and Ruihang Chu and Jianing Qiu and Jiaqi Xu and Mingyu Ding and Hongyang Li and Mengzhe Geng and Yue Wu and Wenhai Wang and Junsong Chen and Zhangyue Yin and Xiaozhe Ren and Jie Fu and Junxian He and Wu Yuan and Qi Liu and Xihui Liu and Yu Li and Hao Dong and Yu Cheng and Ming Zhang and Pheng Ann Heng and Jifeng Dai and Ping Luo and Jingdong Wang and Ji-Rong Wen and Xipeng Qiu and Yike Guo and Hui Xiong and Qun Liu and Zhenguo Li},
journal= {arXiv preprint arXiv:2312.11562},
year = {2024}
}
备注
20 Figures, 160 Pages, 750+ References, Project Page https://github.com/reasoning-survey/Awesome-Reasoning-Foundation-Models