課程資訊
課程名稱
系統效能評估
Performance Modeling 
開課學期
109-2 
授課對象
電機資訊學院  資訊網路與多媒體研究所  
授課教師
周承復 
課號
CSIE5023 
課程識別碼
922 U0240 
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
 
備註
上課教室:資310。
總人數上限:30人 
Ceiba 課程網頁
http://ceiba.ntu.edu.tw/1092CSIE5023_ 
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

Performance Modeling: Distributed Machine-Learning System is an introduction to these system-focused aspects of machine learning, covering guiding principles and commonly used techniques for scaling up to large data sets. That is, we will cover the techniques that lie between a standard machine learning course and an efficient systems implementation.
Overview of SGD
SGD
The condition number
The kernel trick
Non-convex stochastic gradient descent
Parallelism
Algorithms other than SGD
MDP and Reinforcement Learning
Policy gradient, DDPG
NAS
Hybrid types of learning
Broad techniques

Online versus offline learning
Metaparameter optimization
 

課程目標
Performance Modeling: Distributed Machine-Learning System is an introduction to these system-focused aspects of machine learning, covering guiding principles and commonly used techniques for scaling up to large data sets. 
課程要求
Basic ML and Programming knowledge
Math/Statistics knowledge
 
預期每週課後學習時數
 
Office Hours
 
參考書目
papers 
指定閱讀
papers and slides 
評量方式
(僅供參考)
   
課程進度
週次
日期
單元主題
第1週
2/20  Introduction 
第2週
2/27  Holiday 
第3週
3/06  Overview 
第4週
3/13  Overview 
第5週
3/20  Techniques for SGD 
第6週
3/27  Techniques for SGD 
第7週
4/03  Holiday 
第8週
4/10  Variance Reduction in SGD  
第9週
4/17  The Kernel Trick, Gram Matrices, and Feature Extraction 
第10週
4/24  sample of mid-term exam
paper list 
第11週
5/01  Midterm 
第12週
5/10  Hyperparameter Optimization,
Parallelism 
第13週
5/17  Markov Decision Process and Reinforcement Learning 
第14週
5/22  Paper Presentation
1 吳禹璇 Privacy-Preserving Traffic Flow Prediction: A Federated Learning Approach
https://youtu.be/dwGzcLuv1tQ
2 黃光輝 When Edge Meets Learning: Adaptive Control for Resource-Constrained Distributed Machine Learning
https://youtu.be/RxU0kCTRc7g
3 彭梓瑄 Federated Learning with Matched Averaging
https://www.youtube.com/watch?v=JjLONixIZd4
4 許宛筑 Human-level control through deep reinforcement learning
https://youtu.be/pbzhttrIEWM
 
第15週
5/29  Paper Presentation
1 陳俊瑋 Policy Gradient Methods for Reinforcement Learning with Function Approximation
https://www.youtube.com/watch?v=cbDXk4OpTCw
2 鍾宜樺 Variational Neural Annealing
https://www.youtube.com/watch?v=qnIGt29Jrx8
3 謝宏祺 A Closer Look at Few-shot Classification
https://youtu.be/bT0X-1rIptM
4 吳旻栗 "Scalable Distributed DL Training: Batching Communication and Computation
https://drive.google.com/file/d/1l7SI2aSuaZ2q4Iyt60t1y233LMwxdAnu/view 
第16週
6/05  Paper Presentation
1 許國讚 A Survey of Actor-Critic Reinforcement Learning: Standard and Natural Policy Gradient
2 蘇泰宇 HOGWILD! A Lock-Free Approach to Parallelizing Stochastic Gradient Descent
https://www.youtube.com/watch?v=Tl7S2UHrif8&ab_channel=%E8%98%87%E6%B3%B0%E5%AE%87
3 楊晨郁 Metagan: An adversarial approach to few-shot learning
https://youtu.be/MydqE2KB64s 
第17週
6/12  Paper Presentation
1 李岳庭 Optimization as a model for few-shot learning
2 徐衍新 Model-agnostic meta-learning for fast adaptation of deep networks
https://youtu.be/c47HrCB4pgw
3 吳吉加 Mastering the game of Go with deep neural networks and tree search 
第18週
06/19  Final Project Presentation
https://docs.google.com/spreadsheets/d/15FcYhry41osPT9EBYnP9eHkVXJk7gHblFmfRELDwuwg/edit?usp=sharing