課程資訊
課程名稱
預測、學習、與賽局
Prediction, Learning, and Games 
開課學期
107-2 
授課對象
電機資訊學院  資訊工程學研究所  
授課教師
李彥寰 
課號
CSIE5002 
課程識別碼
922 U4550 
班次
 
學分
3.0 
全/半年
半年 
必/選修
選修 
上課時間
星期一7,8,9(14:20~17:20) 
上課地點
資105 
備註
Theory course, requiring math maturity.
總人數上限:20人 
 
課程簡介影片
 
核心能力關聯
核心能力與課程規劃關聯圖
課程大綱
為確保您我的權利,請尊重智慧財產權及不得非法影印
課程概述

*This is an advanced course in learning theory.*

The probably approximately correct (PAC) theory has been the standard framework of machine learning for decades, but its underlying “i.i.d. data” assumption results in a significant theory-practice gap. This course introduces online learning theory, whose probability-free nature naturally avoids the aforementioned theory-practice gap. The main focuses are:
- Learning with expert advice.
- Online convex optimization.
- Individual sequence prediction.
- Convergence to equilibria in repeated games.
The tentative schedule can be found below.

The exact contents of this course may change with respect to the latest advances
in learning theory. 

課程目標
本課程的目標在於讓修課同學:
● Be able to read state-of-the literature on learning theory.
● Be able to analyze basic online (learning) algorithms.
● Be able to work on online learning research topics.
● Be able to think beyond the statistical and PAC learning frameworks. 
課程要求
Prerequisites: Calculus, linear algebra, and probability.

Knowledge in convex optimization, learning theory, and/or statistics can be helpful but not necessary. 
預期每週課後學習時數
 
Office Hours
備註: Every week after class or by appointment.  
指定閱讀
None.  
參考書目
1. N. Cesa-Bianchi and G. Lugosi. 2006. Prediction, Learning, and Games.
2. S. Shalev-Shwartz. 2011. Online Learning and Online Convex Optimization.
3. S. Bubeck. 2011. Introduction to Online Optimization.
4. V. V. V’yugin. 2012. Lecture Notes on Machine Learning and Prediction.
5. S. Hart and A. Mas-Colell. 2013. Simple Adaptive Strategies.
6. A. Rakhlin and K. Sridharan. 2014. Statistical Learning and Sequential Prediction.
7. E. Hazan. 2015. Introduction to Online Convex Optimization.
8. A. Slivkins. 2018. Introduction to multi-armed bandits.
9. T. Lattimore and C. Szepesvári. 2018. Bandit Algorithms. 
評量方式
(僅供參考)
   
課程進度
週次
日期
單元主題
第1週
2/18  Course organization. Learning with and without probabilities. 
第2週
2/25  Statistical learning theory. Rademacher complexity. PAC-Bayes. 
第3週
3/04  Decision theoretic online learning. Hedge.  
第4週
3/11  Multiplicative weight update.  
第5週
3/18  Learning with expert advice. Mixability. Aggregating algorithms. 
第6週
3/25  Online density estimation. Online portfolio selection. 
第7週
4/01  Online linear regression. The many faces of exponential weights. 
第8週
4/08  Midterm.  
第9週
4/15  Follow the leader. Follow the regularized leader. Follow the perturbed leader. 
第10週
4/22  Online convex optimization. Basic convex analysis. 
第11週
4/29  Online convex optimization. Online-to-batch conversion. 
第12週
5/06  Online bandit convex optimization.  
第13週
5/13  Equilibria in games. Blackwell approachability. 
第14週
5/20  Individual sequence prediction.  
第15週
5/27  Probability forecasting. Calibrated forecasting. 
第16週
6/03  Project presentations & disucssions. 
第17週
6/10  Project presentations & discussions.