首页 > 其他 > 详细

Online Algorithms

时间:2021-01-19 09:22:47      阅读:44      评论:0      收藏:0      [点我收藏+]

* Online algorithms到底是啥?

技术分享图片

  (取自http://www.cs.cmu.edu/afs/cs/academic/class/15750-s19/OldScribeNotes/lecture35.pdf开头)

 

**********************************↓2021春夏-本科毕业设计↓********************************

教学日历维护时间为2021年1月15日-2月28日

总References

一. Scheduling

 1. Online Virtual Machine Allocation with Predictions

 

二. Paging

1. Competitive caching with machine learned advice, Thodoris Lykouris, Sergei Vassilvitskii

First published in ICML 2018; last revised 21 Aug 2020 (this version, v4) 

 

三. Metric

1. Online Metric Algorithms with Untrusted Predictions (python代码) (两个例子,俩同学:程振威,田澳冉)

 

四. Ski rental

Online Algorithms for Multi-shop Ski Rental with Machine Learned Advicematlab代码)(一个同学:邢朗)

 

五. $k$-server

Learning-Augmented Online Algorithms for the 2-Server Problem on the Line and Generalizations (rust and python代码) (k-server on the line、2-taxi on the line,俩同学:欧阳嶂浩、冯明尧)

 

六. bin packing (施莹莹同学)

Simulation of Online Bin Packing in Practice 

Online algorithms for variants of bin packing

 

**********************************↑2021春夏-本科毕业设计↑********************************

 技术分享图片

    (取自https://www.cs.umd.edu/~samir/DCscheduling18/slides/Svitkina.pdf的第5页)

 

 

https://www2.cs.duke.edu/courses/spring15/compsci590.1/

https://www.win.tue.nl/~nikhil/AU16/index.html

http://math.ucdenver.edu/~sborgwardt/wiki/index.php/Lagrangian_Duality

 

 

2021-01-18

  • 对online algorithms感兴趣的人其实很多啊,在试图弄明白work function的时候,从这个slide发现了两个亮点:

一个亮点是“Online energy generation scheduling for microgrids with intermittent energy”这篇文章(work function相关,用到了look ahead,因此可以考虑with predictions。该文中的这段话很有启发性:Under typical settings, we show that CHASE achieves the best competitive ratio among all deterministic online algorithms, and the ratio is no larger than a small constant 3. We also extend our algorithms to intelligently leverage on limited prediction of the future, such as near-term demand or wind forecast )。

另一个亮点是:

技术分享图片 

  • “Online Algorithms - From Prediction to Decision-PhD thesis-Niangjun_Chen_2017”与“Data-driven Competitive Algorithms for Online Knapsack and Set Cover”这篇论文必须得看啊!
  • Interestingly, a very similar metric space is powerful enough to model the noted “ski rental problem”, see Karlin, Manasse, Rudolph, and Sleator [68] for details. (摘自“Advanced Techniques for Dynamic Programming”)

 

2021-01-09

  • online problem的Input包括两部分:以k-server problem为例,the first part of input = metric space + k;the online part of input = a sequence of requests。这是原来自己没有留心到的。

2021-01-08

  • 以前看过的一个blog,也讲到了online algorithm,这个blog的意义还在于为本科生算法扩展内容提供素材。

 

2021-01-05

  • 关于paging的randomized MARK算法competitive ratio分析,有图且好理解的是这儿的lecture 8。

 

2021-01-04

  • 关于Adversary Models,讲得最简略而又最容易理解的是ebook-Randomized algorithms一书第13章的13.2节。从adversary的power来看:Oblivious < Adaptive online < Adaptive offline;从竞争比的大小来看:$c^{obl}<c^{aon}<c^{aof}<c^{det}$。
  • 可以认为OPT算法由adversary设计,$\sigma$由adversary生成。上面$c^{det}$最大,是因为deterministic ALG太弱,弱在不是randomized。

 

2020-12-26

  • 关于PDLA,首先得有Linear Programming Recap,这个所见最好的是Lecture 9 — September 30, 2014 Prof. Jelani Nelson。接下来的Lecture 10 — October 2, 2014 Prof. Jelani Nelson以Ski Rental为例讲了PD为工具的online algorithm,再结合ebook-The Design of Competitive Online Algorithms via a Primal-Dual Approach中的相应部分来看,比较好理解。
  • ski rental:就ski rental的primal-dual approach来说,对于2-competitive的算法,有两种处理方式,一是increment $y$,这也最常见的方式,如lecture note1lecture note2lecture note3,分析都是用的approximate complementary slackness result;二是increment $x_{buy}$,这种方式的分析仍然是利用了approximate complementary slackness result,其提出好像是为了与基于primal dual的randomized online algorithm相对比。
  • ski rental:基于primal dual的randomized online algorithm的分析较多,但是就其由来,即算法中的参数为何那么设定,讲这个的是lecture note3。但是最透彻的是Prof. Dr. Yann Disser所给的lecture notes的第7节部分。

 

2020-12-23

巴西的一位老师,随机部分较好

 

2020-11-03

Online and other Myopic Algorithms (Fall 2019)

Algorithms and Uncertainty (Fall, 2016)

 

2020-11-01 

既然有The Design of Competitive Online Algorithms via a Primal–Dual Approach,那么是否可以考虑The Design of Competitive Online Algorithms via a Local Ratio Approach

(On the Equivalence Between the Primal-Dual Schema and the Local Ratio Technique)

http://www.cs.technion.ac.il/users/wwwb/cgi-bin/tr-get.cgi/2004/CS/CS-2004-01.pdf

(On the Equivalence between the Primal-Dual Schema and the Local Ratio Technique)

https://epubs.siam.org/doi/pdf/10.1137/050625382

(Combinatorial Interpretations of Dual Fitting and Primal Fitting)

http://www.eng.biu.ac.il/~rawitzd/Papers/fitting.pdf

 

2020-10-28 

第一个看懂的list update的MTF分析:(https://www.cs.princeton.edu/courses/archive/spring13/cos521/notes/lecture5.pdf)

 

2020-7-31

Tim Roughgarden的lecture notes,真的应该好好通读! 比如,对于online paging的competitive ratio的证明,看得懂的就是从这位大师的17年的Beyond Worst-Case Analysis(lecture 3)开始的,再具体点,就是LRU的upper bound。看懂这个之后,再看“ebook-draft-Online Algorithms-Allan Borodin & Denis Pankratov-March 14, 2019-ch1-8”的第11页,就可以理解LRU与FIFO的相似性,从而理解“ebook-An Introduction to Online Computation Determinism, Randomization, Advice”所证明的FIFO的upper bound了。

  • agnostic learing means: Make no assumptions about a specific functional relationship between $\mathbf{x}$ and $y$.  

2020-7-28

STOC‘20 Workshop on Algorithms with Predictions

Learning-Augmented Algorithms

 

 

Researchers

Dennis Komm (author of ebook-An Introduction to Online Computation Determinism, Randomization, Advice)

Teachers

Thomas Kesselheim (2017-2020的lecture notes都很好!)

Prof. Dr. Yann Disser

 

Online Algorithms

原文:https://www.cnblogs.com/liu-hong/p/13694625.html

(0)
(0)
   
举报
评论 一句话评论(0
关于我们 - 联系我们 - 留言反馈 - 联系我们:wmxa8@hotmail.com
© 2014 bubuko.com 版权所有
打开技术之扣,分享程序人生!