活动报名|北京大学多体智能中心研讨会

栏目:安全教育  时间:2022-11-30
手机版

  关于多智能体中心首届单日讨论会

  北京大学人工智能研究院多智能体中心(CMAR)主要研究开放环境下的通用单智能体、多智能体强化学习理论、方法与应用;智能群体机器人与工程应用;游戏环境和游戏智能体的设计、分析及相关算法研究;计算经济中的智能体建模与人工智能算法;面向供应链的人工智能决策方法;基于融入行为特征的风险度量的强化学习方法;大规模、智能化、网络化、多层次、多尺度、多模式、非线性、不确定时变动态系统的建模、分析、模拟、预测、优化和控制。

  多智能体中心首届单日讨论会将于5月28日在线举行。由北京大学人工智能研究院多智能体中心的青年教师组织,聚焦基于近期人工智能与多智能体技术在通用人工智能、智能机器人、游戏AI、智能群体决策、网络经济学等方向的前沿研究。

  本届多智能体中心日活动以特邀报告加圆桌论坛的形式进行。将邀请多智能体以及人工智能领域的国内外专家和研究人员带来多场精彩报告。

  扫码报名

  报告信息

  以报告顺序排序

  ”

  Towards Model-Agnostic Rare-Event Simulation with Guarantees

  摘要

  Quantifying and hedging against rare events is ubiquitous in risk-aware modeling and learning. In estimating rare-event probabilities, naive Monte Carlo is inefficient and techniques such as importance sampling have shown to be powerful in speeding up computation. However, to attain statistical guarantees, these techniques typically rely on model structural knowledge that is not always available or amenable to analysis, a challenge that is especially prominent for complex black-box systems such as those arising in AI safety applications. We describe some approaches towards obtaining guarantees for importance sampling with lighter requirements on model knowledge. Specifically, we propose and justify relaxations of classical efficiency notions on several fronts, to account for: 1) over-conservativeness of variance-based criteria in enforcing algorithmic properties; 2) lack of precise geometric knowledge on rare-event sets; and 3) lack of a priori knowledge on the solution when solving optimization problems or training models that are rare-event-aware. We present simple procedures to attain these relaxed efficiency notions and discuss their performances.

  报告人

  

  报告人:Henry Lam

  Henry Lam is an Associate Professor in the Department of Industrial Engineering and Operations Research at Columbia University. He received his PhD degree in statistics from Harvard University and was on the faculty of Boston University and the University of Michigan before joining Columbia in 2017. His research interests include Monte Carlo methods, uncertainty quantification, data-driven optimization and rare-event analysis. Henry's works have been recognized by several venues such as the NSF CAREER Award, NSA Young Investigator Award, JP Morgan Chase Faculty Research Award and Adobe Faculty Research Award. He serves on the editorial boards of Operations Research, INFORMS Journal on Computing, Applied Probability Journals, Stochastic Models, Manufacturing and Service Operations Management, and Queueing Systems, and as the Area Editor in Stochastic Models and Data Science in Operations Research Letters.

  Option Pricing by Neural Stochastic Differential Equations:

  A Simulation Optimization Approach

  摘要

  The classical option pricing models rely on prior assumptions on the dynamics of the underlying assets. Though empirical evidence shows that these models may partially explain the option prices, their performance may be poor when the actual situations deviate from the assumptions. Neural network models are capable of learning the underlying relationship through the data. However, they require massive amount of data to avoid over-fitting, which is typically not available for option pricing problems. Thus, we propose a new model by integrating neural networks to a classical stochastic differential equation pricing model to balance the model flexibility and the data requirement. Besides, some more specific models are also constructed by using neural network as a model calibration method of the classical models. Furthermore, we show that the training of the model can be formulated into a simulation optimization problem and can be solved in a way that is compatible to the training of neural networks as well. Preliminary numerical results show that our approach appears to work better compared with some existing models. This is a joint work with Shoudao Wang and Nifei Lin.

  报告人

  报告人:洪流

  洪流教授本科毕业于清华大学,博士毕业于美国西北大学。现任复旦大学特聘教授、弘毅讲席教授、大数据学院副院长和管理科学系系主任;曾任香港城市大学管理科学讲座教授,香港科技大学教授和金融工程实验室主任等。洪流教授的研究主要集中在随机运筹学、数据科学、供应链管理、风险管理等领域,在Operations Research和Management Science等UTD期刊上发表论文二十余篇。洪教授目前担任INFORMS仿真分会主席、中国管理现代化研究会风险管理专委会主任、中国运筹学会金融工程和风险管理分会副理事长、《Journal of Operations Research Society of China》的Associate Editor-in-Chief、《Operations Research》的Area Editor和《Management Science》的Associate Editor等。

  Actionable Machine Learning for Tackling Distribution Shift

  摘要

  To deploy machine learning algorithms in real-world applications, we must pay attention to distribution shift. When the test distribution differs from the training distribution, there will be a substantial degradation in model performance. To tackle the distribution shift, in this talk, I will present two paradigms with some instantiations. Concretely, I will first discuss how to build machine learning models that are robust to two kinds of distribution shifts, including subpopulation shift and domain shift. I will then discuss how to effectively adapt the trained model to the test distribution with minimal labeled data. The remaining challenges and promising future research directions will also be discussed.

  报告人

  

  报告人:姚骅修

  Huaxiu Yao is a Postdoctoral Scholar in Computer Science at Stanford University, working with Prof. Chelsea Finn. Currently, his research focuses on building machine learning models that are robust to distribution shifts. He is also passionate about applying these methods to solve real-world problems with limited data. He obtained his Ph.D. degree from Pennsylvania State University. The results of his work have been published in top-tier venues such as ICML, ICLR, NeurIPS. He organized the MetaLearn workshop at NeurIPS, the pre-training workshop at ICML, and he served as a tutorial speaker at KDD, IJCAI, and AAAI.

  集群漫谈——“乌合之众”的胜利

  摘要

  你是否曾在漫天的晚霞中驻足欣赏百万级规模的欧椋鸟群呼啸而过,带着似童话般美好、似沙画般灵动的图案?你是否曾经幻化成一条自由的鱼儿在大海中遨游,为偶然看到的沙丁鱼群环绕成巨大的球体以抵抗鲨鱼的侵袭而伤感落泪?生物群中的这种集群行为背后的机理到底是什么?如何能够更快地形成集群行为?如何能够使得集群行为更稳定?如何利用集群的思想,通过设计合理、高效的分布式交互和决策机制,依靠一群“乌合之众”,完成复杂协同任务?本次讲座将为您讲述集群行为背后的故事。

  报告人

  报告人:贾永楠

  贾永楠,1984年生人,河北乐亭人,北京科技大学自动化学院副教授,硕士生导师,中国仿真学会智能物联系统建模与仿真专委会秘书长。2003年考入北京工业大学电子科学与技术系,四年后保送到北京大学师从王龙教授从事水下机器人设计、运动控制及群体行为研究。博士毕业后,进入中国航天科工三院无人机总体单位的系统总体技术岗位工作。两年后,回归高校,后赴匈牙利与匈牙利科学院院士、知名物理学家Tamas Vicsek教授合作开展无人机层级化控制方面的研究。先后获得中国自然科学基金青年项目、博士后科学基金面上项目、ICTP(国际理论物理中心)、中电科集团和鞍钢集团的横向课题等资助。贾永楠在集群领域深耕十余年,发表SCI/EI学术论文十余篇,研究成果受到国内集群领域的关注,发表的集群综述文章是2020年《航空学报》《CJA》最受关注十大热点文章之一(排名第三)。

  现代人工智能系统中以数据为驱动的研究

  摘要

  近十年来,人工智能的高速发展已经渗透到各行各业。在学术界,我们看到了大量关于神经网络架构、模型、优化器或者新的学习范式的创新。相应地,工业界的关注点却往往在数据本身和部署上面:例如如何在数据采集和标注成本更低廉的情况下训练出性能更加优异的模型。在本报告中,演讲者拟就学术界和工业界人工智能算法开发中的耦合和区别进行系统性分析。并且,演讲者拟就模型自动化评估、主动学习和偏标签学习等方向介绍课题组相关工作。

  报告人

  

  报告人:赵俊博

  赵俊博(英文名Jake),现为浙江大学计算机学院百人计划教职(ZJU-100 Young Professor Program)、博士生导师,前连续创业者。目前领导硕士生、博士生20余人团队,主攻方向为人工智能数据侧驱动的算法研究。赵俊博于2019年获得纽约大学计算机博士学位。赵俊博博士师从图灵奖得主、美国两院院士、卷积神经网络发明人Yann LeCun。在博士期间,曾在Facebook人工智能实验室从事研究工作近2年。截止2022年5月,他在Google Scholar上的论文总他引数目近11000次。

  多样性强化学习:不光要赢,还要赢得精彩

  摘要

  近年来,深度强化学习技术已经攻克了越来越多的人工智能挑战,也在许多任务中击败了无数顶尖人类选手。强化学习AI以得到最高的奖励分数为目标,并能以最简单高效的方式赢得比赛。虽然AI很强,但是AI的行为却往往千篇一律,与人相比缺少了些创造性。人类不光能够赢得比赛,常常还能发明各种各样有趣的创新策略并以此为乐。那为什么有些人类觉得好玩有意思的行为,强化学习AI却从来学不会呢?在这个报告中,我们会对这个问题进行分析,并提出新的强化学习范式,多样性强化学习——即,AI不光要得高分,还要尽可能用不同的方式赢得高分。我们也会介绍两个多样性强化学习算法,并展示在不同场景中,这两个算法所发现的有趣的行为。

  报告人

  

  报告人:吴翼

  吴翼,清华大学交叉信息研究院助理教授,回国前曾任OpenAI全职研究员,研究领域为强化学习的泛化性,多智能体学习,自然语言理解,机器人学习等。2019年在美国加州大学伯克利分校获得博士学位,师从Stuart Russell教授;2014年本科毕业于清华大学交叉信息院计算机科学实验班(姚班)。其代表作包括:NIPS2016最佳论文,Value Iteration Network;多智能体深度强化学习领域最高引用论文,MADDPG算法;以及OpenAI hide-and-seek 项目等。

  社会仿真模拟与社会智能学习

  摘要

  社会是一个复杂系统,由诸多行动着的智能体组成。计算社会科学的重要研究方向之一,就是通过智能体刻画、模拟个体、组织、社会的动态运行、模式涌现。已有的社会仿真模拟研究中,个体的智能化程度较低,未能形成有效知识。基于社会智能学习算法,个体能够从有限的案例、模拟中学习到高维度的社会共同知识。基于共同知识矩阵的学习,使得个体能够实现利益最大化,并同时提升全社会的集体福利。

  报告人

  

  报告人:吕鹏

  吕鹏,中南大学公共管理学院教授、自动化学院教授、中南大学社会计算研究中心主任,教育部“青年长江学者”。清华大学社会学系博士,清华大学自动化系博士后。美国芝加哥大学联合培养博士、韩国首尔大学高级访问学者。研究范式具有跨学科特征,将自然科学方法应用到社会复杂巨系统研究。目前主要研究社会公共安全、智能社会治理、计算社会科学等。主持国家社科重大项目、国家自然科学基金面上项目(2016-2020)等。开设《计算社会科学》在线慕课。

  点击“阅读原文”,即可报名

上一篇:新东方雅思线上培训班怎么样收费
下一篇:2022版高考成绩可申请港澳大学一览表

最近更新安全教育