‍‍

‍‍

‍‍

‍‍

北京智源人工智能研究院（简称“智源研究院”）是落实“北京智源行动计划”的重要举措，是在科技部和北京市委市政府的指导和支持下，由北京市科委和海淀区政府于2018年11月推动成立的新型研发机构。

为支持科学家勇闯人工智能科技前沿“无人区”，挑战最基础的问题和最关键的难题，推动人工智能理论、方法、工具、系统和应用取得变革性、颠覆性突破，智源研究院先后成立了「人工智能的数理基础」、「机器学习」、「智能信息检索与挖掘」、「智能体系架构与芯片」、「自然语言处理」、「人工智能的认知神经基础」等六个重大研究方向。

「人工智能的数理基础」重大方向，以北京大学张平文院士牵头，分别来自北大、清华、中科院的12位学者，围绕当前人工智能面临的可计算性、可解释性、泛化性、稳定性的重大理论挑战，以期打破基于计算机实验和神经科学的人工智能惯用建模范式，建立以数学与统计理论为第一原理的新一代人工智能方法论。

为促进该领域最前沿研究在国内广泛、快速地传播，以及提供学科交叉碰撞的平台，智源研究院「人工智能的数理基础」方向的诸位学者将不定期邀请世界范围内的顶尖学者进行系列前沿报告。

本期报告，我们分别邀请了清华大学统计科学研究中心助理教授杨朋昆、北京大学北京国际数学研究中心博士后尤翀、中国科学院理论物理研究所研究员张潘，就其最新研究做报告讲解。

时间形式

1. 会议时间：2021年5月26日14:00-17:00

2. 活动形式：腾讯会议（ID ：620 505 900）

报名入群：

（入会、交流、资料获取）

活动日程

会议详情

报告一

报告主题：Learning overparametrized neural networks and statistical models

报告人：杨朋昆

个人简介：Pengkun Yang is an assistant professor at the Center for Statistical Science at Tsinghua University. Prior to joining Tsinghua, he was a Postdoctoral Research Associate at the Department of Electrical Engineering at Princeton University. He received a Ph.D. degree (2018) and a master degree (2016) from the Department of Electrical and Computer Engineering at University of Illinois at Urbana-Champaign, and a B.E. degree (2013) from the Department of Electronic Engineering at Tsinghua University. His research interests include statistical inference, learning, optimization and systems. He is a recipient of Thomas M. Cover Dissertation Award in 2020, and a recipient of Jack Keil Wolf ISIT Student Paper Award at the 2015 IEEE International Symposium on Information Theory (semi-plenary talk).

报告摘要：Modern machine learning has constantly presented puzzling empirical properties and surprised the classical statistical theory. Learning with overparametrized models is becoming a norm in data-analytic applications, and the tension of memorization rarely bothers practitioners. In this talk, I will discuss the training of overparametrized neural networks from both the neural tangent kernel and the mean-field perspectives, which guarantees the global convergence property despite the non-convexity of the optimization landscape. I will also discuss more interesting phenomena in a series of overparametrized statistical questions.

报告二

报告主题：Variational Bayes approaches to the selection of linear models

报告人：尤翀

个人简介：尤翀博士，2015年获得悉尼大学统计学博士，北京大学北京国际数学研究中心博士后。主要从事统计方法和理论的研究工作，研究方向集中在变分近似，线性混合效应模型分析，模型选择，以及GWAS相关改进工作。近年来也在流病分析上，尤其是新冠疫情分析，开展方法学研究。

报告摘要：In applied statistics model selection is one of the most fundamental tasks. The selection of models in the linear regression context has been well-studied. However, the efficiency and effectiveness of these methods in high-dimensional settings is limited. We present a model selection method to addresses this.

Variational Bayes (VB) is known as a fast alternative to Markov chain Monte Carlo for performing approximate Bayesian inference. However, VB is often criticised, typically based on empirical grounds, for being unable to produce valid statistical inferences. We proved VB based estimators share asymptotic properties in Bayesian linear models, partially contradicting this criticism.

We further extend the VB approach to the more complicated spike and slab priors. We show under mild regularity conditions, that: (i) VB based estimators for the coefficients are consistent estimators of the true parameters; and (ii) the VB estimators of the model indicator variables shrink towards zero rapidly if the corresponding true value of the coefficient is zero and one otherwise. This property allows us to use VB estimates of indicator variables to select models.

报告三

报告主题：Computation with Tensor Networks

报告人：张潘

个人简介：Pan Zhang is working as a professor at the Institute of Theoretical Physics, Chinese Academy of Sciences (ITP, CAS). He got his Ph.D. from Lanzhou University in 2009 and did post-docs at Europe and the Santa Fe Institute in USA before joining ITP, CAS in 2015. Pan Zhang’s research is in the interdisciplinary field of statistical physics, machine learning, quantum many-body, and quantum computation.

报告摘要：Pan Zhang will present methods and algorithms for solving statistical mechanics problems, combinatorial optimization problems, and quantum circuit simulations, in an integrated framework based on tensor networks.

In statistical mechanics problems, the partition function at a finite temperature can be obtained by contracting a tensor network that is converted from the statistical mechanics problem. When equipped with the “Tropical” algebra, the tensor network contraction can be used to obtain ground state energy and entropy of the model directly at zero temperature. When the interactions in the statistical mechanics model are complex, computing the partition function acts as estimating the amplitude of a basis vector of the final state of a quantum circuit, thus tensor network contractions can be used to simulate quantum computers. Pan Zhang will introduce approximate and exact algorithms for contracting tensor networks, and their wide applications, particularly in simulating Google’s Sycamore quantum circuits.

‍

‍

‍

‍

‍

‍

‍

‍