神经符号学及其应用研究

doi:10.11959/j.issn.2096-6652.202234

智能科学与技术学报 ›› 2022, Vol. 4 ›› Issue (4): 560-570.doi: 10.11959/j.issn.2096-6652.202234

• 专栏：人工智能3.0中的机器学习方法 • 上一篇下一篇

神经符号学及其应用研究

蔡莹皓¹^,², 杨华³, 安璇¹^,², 王文硕¹, 杜沂东¹, 张嘉韬³, 王志刚³

¹ 中国科学院自动化研究所复杂系统管理与控制国家重点实验室，北京 100190
² 中国科学院大学人工智能学院，北京 100049
³ 英特尔中国研究院，北京 100190

修回日期:2022-07-10 出版日期:2022-12-15 发布日期:2022-12-01
作者简介:蔡莹皓（1983− ），女，中国科学院自动化研究所复杂系统管理与控制国家重点实验室副研究员，主要研究方向为机器人操作技能学习、具身智能
杨华（1975− ），女，博士，就职于英特尔中国研究院，主要研究方向为个性化知识图谱、基于人类常识的推理
安璇（2001− ），女，中国科学院自动化研究所复杂系统管理与控制国家重点实验室硕士生，主要研究方向为机器人操作技能学习
王文硕（1999− ），男，就职于中国科学院自动化研究所复杂系统管理与控制国家重点实验室，主要研究方向为机器人抓取
杜沂东（1997− ），男，就职于中国科学院自动化研究所复杂系统管理与控制国家重点实验室，主要研究方向为仿生机器人
张嘉韬（1997− ），男，就职于英特尔中国研究院，主要研究方向为知识图谱与不确定性推理
王志刚（1978− ），男，博士，英特尔中国研究院研究员，主要研究方向为机器人时空智能、基于统计方法和符号方法的混合 AI系统、系统优化

Study on NeuroSymbolic learning and its applications

Yinghao CAI¹^,², Hua YANG³, Xuan AN¹^,², Wenshuo WANG¹, Yidong DU¹, Jiatao ZHANG³, Zhigang WANG³

¹ The State Key Laboratory for Management and Control of Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
² School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China
³ Intel Labs China, Beijing 100190, China

Revised:2022-07-10 Online:2022-12-15 Published:2022-12-01

摘要/Abstract

摘要：

深度学习在感知智能上的不断突破推动了人工智能在各领域的广泛应用。但在实际落地过程中，只有把感知智能提高到更高层的认知智能，才能更好地满足日益复杂的应用需求。神经符号学将擅长感知任务的神经网络方法和擅长推理任务的逻辑符号学有机地融合在一起，是实现高层认知智能的途径之一。基于此，提出了一套神经符号学的实用框架NSFOL，并基于NSFOL实现了机器人任务规划、自学习机器人运动规划和教育实验视频评估3个典型应用。实验结果表明，尽管NSFOL尚未完善，但是它已经能够很好地支持相关应用，在可学习、可推理、可解释和可泛化方面具备一定的优势。希望通过阐述神经符号学的阶段性研究成果，激发更多的思考和研究，共同推动神经符号学的发展。

关键词: 人工智能, 神经符号学, 机器人任务规划, 机器人运动规划, 教育实验视频评估

Abstract:

The continuous breakthrough of deep learning in perception has promoted the application of AI in various fields.It is found that we can not meet the requirements without improving the intelligence from perception level to higher cognition level.NeuroSymbolic learning can seamlessly integrate neural network methods, that are good at perception tasks, and logical symbolic methods, that are good at reasoning tasks.Therefore, it is one of the best candidates to achieve high-level cognitive intelligence.A practical framework for NeuroSymbolic learning:NSFOL was proposed.Moreover, three typical applications based on NSFOL: robot motion planning, robot task planning and video evaluation for educational experiment were presented.Experiments show that NSFOL can support these three specific applications successfully.Moreover, these implementations have advantages in learn ability, reasonability, interpretability and generalizability.Hope to stimulate more thinking and research to jointly promote research in NeuroSymbolic learning by sharing our preliminary studies in this direction.

Key words: AI, NeuroSymbolic learning, robot motion planning, robot task planning, video evaluation for educational experiment

中图分类号:

TP242.6

蔡莹皓, 杨华, 安璇, 等. 神经符号学及其应用研究[J]. 智能科学与技术学报, 2022, 4(4): 560-570.

Yinghao CAI, Hua YANG, Xuan AN, et al. Study on NeuroSymbolic learning and its applications[J]. Chinese Journal of Intelligent Science and Technology, 2022, 4(4): 560-570.

图/表 9

图1

图2

图3

图4

图5

图6

图7

图8

图9

参考文献 33

[1]	YANN L C , . Deep learning has outlived its usefulness as a buzz-phrase[Z]. 2018.
[2]	BENGIO Y , . From system 1 deep learning to system 2 deep learning[C]// Proceedings of the 33rd Conference on Neural Information Processing Systems.[S.l.:s.n.], 2019.
[3]	GARCEZ A A , LAMB L C . Neurosymbolic AI:the 3rd wave[J]. arXiv preprint, 2020,arXiv:2012.05876.
[4]	张钹, 朱军, 苏航 . 迈向第三代人工智能[J]. 中国科学:信息科学, 2020,50(9): 1281-1302.
	ZHANG B , ZHU J , SU H . Toward the third generation of artificial intelligence[J]. Scientia Sinica (Informationis), 2020,50(9): 1281-1302.
[5]	GALLI E , KAHNEMAN D . Thinking,fast and slow[J]. Journal of Public Finance and Public Choice, 2011,29(1/2/3): 214-215.
[6]	MAO J Y , GAN C , KOHLI P ,et al. The neuro-symbolic concept learner:interpreting scenes,words,and sentences from natural supervision[J]. arXiv preprint,2019, 2019,arXiv:1904.12584.
[7]	RAEDT L , MANHAEVE R , DUMANCIC S ,et al. Neuro-symbolic =neural+logical+ probabilistic[C]// Proceedings of the 14th International Workshop on Neural-Symbolic Learning and Reasoning.[S.l.:s.n.], 2019.
[8]	KAMPFFMEYER M , CHEN Y B , LIANG X D ,et al. Rethinking knowledge graph propagation for zero-shot learning[C]// Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019: 11479-11488.
[9]	EVANS R , GREFENSTETTE E . Learning explanatory rules from noisy data[J]. Journal of Artificial Intelligence Research, 2018,61: 1-64.
[10]	SI X J , RAGHOTHAMAN M , HEO K ,et al. Synthesizing datalog programs using numerical relaxation[J]. arXiv preprint, 2019,arXiv:1906.00163.
[11]	KALYAN A , MOHTA A , POLOZOV O ,et al. Neural-guided deductive search for real-time program synthesis from examples[J]. arXiv preprint, 2018,arXiv:1804.01186.
[12]	ZHOU Z H . Abductive learning:towards bridging machine learning and logical reasoning[J]. Science China Information Sciences, 2019,62(7): 1-3.
[13]	XU J Y , ZHANG Z L , FRIEDMAN T ,et al. A semantic loss function for deep learning with symbolic knowledge[J]. arXiv preprint, 2017,arXiv:1711.11157.
[14]	CHEN R Q , CHEN T S , HUI X L ,et al. Knowledge graph transfer network for few-shot recognition[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020,34(7): 10575-10582.
[15]	LI A X , LUO T G , LU Z W ,et al. Large-scale few-shot learning:knowledge transfer with class hierarchy[C]// Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2019: 7205-7213.
[16]	VAN EMDEN M H , KOWALSKI R A . The semantics of predicate logic as a programming language[J]. Journal of the ACM, 1976,23(4): 733-742.
[17]	GARRETT C R , CHITNIS R , HOLLADAY R ,et al. Integrated task and motion planning[J]. Annual Review of Control,Robotics,and Autonomous Systems, 2021,4: 265-293.
[18]	CASTAMAN N , PAGELLO E , MENEGATTI E ,et al. Receding horizon task and motion planning in changing environments[J]. arXiv preprint, 2020,arXiv:200903139.
[19]	HASLUM P , LIPOVETZKY N , MAGAZZENI D ,et al. An introduction to the planning domain definition language[J]. Synthesis Lectures on Artificial Intelligence and Machine Learning, 2019,13(2): 1-187.
[20]	FOX M , LONG D . PDDL2.1:an extension to PDDL for expressing temporal planning domains[J]. Journal of Artificial Intelligence Research, 2003,20: 61-124.
[21]	LITTMAN M L . PPDDL1.0:an extension to PDDL for expressing planning domains with probabilistic effects[Z]. 2004.
[22]	SERRANO S A , SANTIAGO E , MARTINEZ-CARRANZA J , ,et al. Knowledge-based hierarchical POMDPs for task planning[J]. Journal of Intelligent ＆ Robotic Systems, 2021,101(4): 1-30.
[23]	WANG Y F . Hierarchical task planning for space manipulator with multi-constraint[J]. Journal of Mechanical Engineering, 2017,53(11): 104.
[24]	MANHAEVE R , DUMAN?I? S , KIMMIG A ,et al. Deepproblog:neural probabilistic logic programming[J]. arXiv preprint, 2018,arXiv:180510872.
[25]	GU S X , HOLLY E , LILLICRAP T ,et al. Deep reinforcement learning for robotic manipulation with asynchronous off-policy updates[C]// Proceedings of 2017 IEEE International Conference on Robotics and Automation. Piscataway:IEEE Press, 2017: 3389-3396.
[26]	ZHANG F Y , LEITNER J , MILFORD M ,et al. Towards vision-based deep reinforcement learning for robotic motion control[J]. arXiv preprint, 2015,arXiv:151103791.
[27]	DIXON J , BELL K . Electric vehicles:battery capacity,charger power,access to charging and the impacts on distribution networks[J]. eTransportation, 2020,4:100059.
[28]	KE Q D , ZHANG P , ZHANG L ,et al. Electric vehicle battery disassembly sequence planning based on frame-subgroup structure combined with genetic algorithm[J]. Frontiers in Mechanical Engineering, 2020,6:576642.
[29]	MAHARSHI S , REDDY K J . Cloud based disassembly of electric vehicle battery[J]. Procedia Manufacturing, 2019,30: 136-142.
[30]	BLANKEMEYER S , WIENS D , WIESE T ,et al. Investigation of the potential for an automated disassembly process of BEV batteries[J]. Procedia CIRP, 2021,98: 559-564.
[31]	HARPER G , SOMMERVILLE R , KENDRICK E ,et al. Recycling lithium-ion batteries from electric vehicles[J]. Nature, 2019,575(7781): 75-86.
[32]	任伟, 王志刚, 杨华 ,等. 基于神经符号的动力电池拆解任务与运动规划[J]. 计算机研究与发展, 2021(12): 2604-2617.
	REN W , WANG Z G , YANG H ,et al. NeuroSymbolic task and motion planner for disassembly electric vehicle batteries[J]. Journal of Computer Research and Development, 2021(12): 2604-2617.
[33]	DU Y D , WANG W S , WANG Z G ,et al. Learning symbolic operators:a neurosymbolic solution for autonomous disassembly of electric vehicle battery[J]. arXiv preprint, 2022,arXiv:2206.03027.

神经符号学及其应用研究

Study on NeuroSymbolic learning and its applications

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 33

相关文章 15

Metrics

推荐阅读 0

[1]	王飞跃, 缪青海, 张军平, 郑文博, 丁文文. 探讨AI for Science的影响与意义：现状与展望[J]. 智能科学与技术学报, 2023, 5(1): 1-6.
[2]	张军欢, 朱正一, 蔡可玮. 人工智能与量化交易的课程建设[J]. 智能科学与技术学报, 2023, 5(1): 104-112.
[3]	田永林, 陈苑文, 杨静, 王雨桐, 王晓, 缪青海, 王子然, 王飞跃. 元宇宙与平行系统：发展现状、对比及展望[J]. 智能科学与技术学报, 2023, 5(1): 121-132.
[4]	卢经纬, 程相, 王飞跃. 求解微分方程的人工智能与深度学习方法：现状及展望[J]. 智能科学与技术学报, 2022, 4(4): 461-476.
[5]	武强, 季雪庭, 吕琳媛. 元宇宙中的人工智能技术与应用[J]. 智能科学与技术学报, 2022, 4(3): 324-334.
[6]	赖文柱, 陈德旺, 何振峰, 邓新国, GIUSEPPE CARLO Marano. 地铁列车驾驶技术发展综述：从人工驾驶到智能无人驾驶[J]. 智能科学与技术学报, 2022, 4(3): 335-343.
[7]	胡东伟, 冯晓璐. 大脑建模的理论框架及热点问题[J]. 智能科学与技术学报, 2021, 3(4): 412-434.
[8]	李亚玲, 杨林瑶, 葛俊, 覃缘琪, 王晓. 博弈5.0：基于平行系统和机器博弈的社会认知平行博弈[J]. 智能科学与技术学报, 2021, 3(4): 507-520.
[9]	王飞跃. 平行哲学与智能技术：平行产业与智慧社会的对偶方程与测试基础[J]. 智能科学与技术学报, 2021, 3(3): 245-255.
[10]	刘文, 胡琨林, 李岩, 刘钊. 移动目标轨迹预测方法研究综述[J]. 智能科学与技术学报, 2021, 3(2): 149-160.
[11]	吴军, 张俊. 长江智城智慧平台框架的规划与实践[J]. 智能科学与技术学报, 2021, 3(2): 211-217.
[12]	刘晨熙, 刘炳宏, 张贤, 龙新南, 彭木根. 面向智能服务的雾无线接入网络：原理、技术与挑战[J]. 智能科学与技术学报, 2021, 3(1): 10-17.
[13]	吕宝粮, 张亚倩, 郑伟龙. 情感脑机接口研究综述[J]. 智能科学与技术学报, 2021, 3(1): 36-48.
[14]	刘朝阳, 穆朝絮, 孙长银. 深度强化学习算法与应用研究现状综述[J]. 智能科学与技术学报, 2020, 2(4): 314-326.
[15]	沈宇,韩金朋,李灵犀,王飞跃. 游戏智能中的AI——从多角色博弈到平行博弈[J]. 智能科学与技术学报, 2020, 2(3): 205-213.