Stochastic methods for Statistical Learning
Statistical learning is the core application of statistics in the fields of artificial intelligence, machine learning, and data analysis. This course will focus on the stochastic methods of statistical learning, taking Markov chains and stochastic differential equations as the theoretical foundation to delve into important contents such as contrastive divergence algorithm, score matching algorithm, noise contrastive estimation, and diffusion models. The course will be supported by algorithm implementation practice, work exhibition and exchange, and interpretation of cutting-edge papers, aiming to help students fully grasp relevant knowledge and skills. Additionally, the course will also briefly introduce the use of relevant computer languages and software packages.
Lecturer
Date
8th April ~ 8th July, 2024
Location
| Weekday | Time | Venue | Online | ID | Password |
|---|---|---|---|---|---|
| Monday,Wednesday | 15:35 - 17:10 | A3-1-301 | ZOOM 04 | 482 240 1589 | BIMSA |
Prerequisite
Real analysis, probability theory, statistics, stochastic processes, and stochastic differential equations
Syllabus
**I. Nature and Objectives of the Course**
Course Nature:
"Statistical Learning Random Methods" is a general machine learning course that focuses on summarizing and introducing random methods in statistical learning. It primarily covers contrastive divergence, score matching, noise-contrastive estimation, diffusion models, generative adversarial networks, and more.
Objectives:
Through this course, students will acquire a fundamental understanding of the ideas and methods centered on random simulations in statistical learning and their applications across various fields. They will learn to implement algorithms in statistical learning using familiar computer languages or apply statistical learning theories to areas of interest. It is hoped that students will be inspired to design better algorithms and models.
- Reviewing basic statistical learning foundations
- Primarily discussing statistical learning methods based on random simulations; introducing advanced statistical learning models; delving into principles and related papers
- Algorithm design. Most computer languages provide libraries for statistical learning. Probabilistic programming is also popular. Particularly recommended Python third-party libraries include scikit-learn, pymc3, statsmodels; rpy2 implements an R interface. Other programming languages such as Matlab, Julia, Go, Nim will also be introduced.
**II. Design of Course Teaching Methods**
The course will primarily adopt online teaching, combined with classroom discussions, literature explanations, and algorithm exercises.
Online teaching will mainly focus on explaining newer random simulation methods in statistical learning. Through analyzing classic papers, students will understand experts' research approaches and be able to design their own models/algorithms.
Classroom discussions will facilitate interaction between teachers and students, improve teaching efficiency, promptly address students' doubts, and broaden their perspectives.
Course Nature:
"Statistical Learning Random Methods" is a general machine learning course that focuses on summarizing and introducing random methods in statistical learning. It primarily covers contrastive divergence, score matching, noise-contrastive estimation, diffusion models, generative adversarial networks, and more.
Objectives:
Through this course, students will acquire a fundamental understanding of the ideas and methods centered on random simulations in statistical learning and their applications across various fields. They will learn to implement algorithms in statistical learning using familiar computer languages or apply statistical learning theories to areas of interest. It is hoped that students will be inspired to design better algorithms and models.
- Reviewing basic statistical learning foundations
- Primarily discussing statistical learning methods based on random simulations; introducing advanced statistical learning models; delving into principles and related papers
- Algorithm design. Most computer languages provide libraries for statistical learning. Probabilistic programming is also popular. Particularly recommended Python third-party libraries include scikit-learn, pymc3, statsmodels; rpy2 implements an R interface. Other programming languages such as Matlab, Julia, Go, Nim will also be introduced.
**II. Design of Course Teaching Methods**
The course will primarily adopt online teaching, combined with classroom discussions, literature explanations, and algorithm exercises.
Online teaching will mainly focus on explaining newer random simulation methods in statistical learning. Through analyzing classic papers, students will understand experts' research approaches and be able to design their own models/algorithms.
Classroom discussions will facilitate interaction between teachers and students, improve teaching efficiency, promptly address students' doubts, and broaden their perspectives.
Reference
Christian P. Robert, George Casella. Monte Carlo Statistical Methods, New York: Springer-Verlag, 1999.
Audience
Undergraduate
, Advanced Undergraduate
, Graduate
, Postdoc
, Researcher
Video Public
Yes
Notes Public
Yes
Language
Chinese
, English
Lecturer Intro
Congwei Song received the master degree in applied mathematics from the Institute of Science in Zhejiang University of Technology, and the Ph.D. degree in basic mathematics from the Department of Mathematics, Zhejiang University, worked in Zhijiang College of Zhejiang University of Technology as an assistant from 2014 to 2021, from 2021 on, worked in BIMSA as asistant researcher. His research interests include machine learning, as well as wavelet analysis and harmonic analysis.