Join for Free
Log In

艺术与人文
商务
计算机科学
数据科学
信息技术
健康
数学和逻辑
个人发展
物理科学与工程
社会科学
语言学习
學位
證書

瀏覽 Coursera 的全部課程

瀏覽
搜索
企業版
登錄
免費加入

Proofs 1

Loading...

Advanced Algorithms and Complexity

加州大学圣地亚哥分校

4.6（328 個評分） | 32K 名學生已註冊

課程 5（共 6 門，数据结构与算法專項課程）

You've learned the basic algorithms now and are ready to step into the area of more complex problems and algorithms to solve them. Advanced algorithms build upon basic ones and use new ideas. We will start with networks flows which are used in more typical applications such as optimal matchings, finding disjoint paths and flight scheduling as well as more surprising ones like image segmentation in computer vision. We then proceed to linear programming with applications in optimizing budget allocation, portfolio optimization, finding the cheapest diet satisfying all requirements and many others. Next we discuss inherently hard problems for which no exact good solutions are known (and not likely to be found) and how to solve them in practice. We finish with a soft introduction to streaming algorithms that are heavily used in Big Data processing. Such algorithms are usually designed to be able to process huge datasets without being able even to store a dataset. Do you have technical problems? Write to us: coursera@hse.ru

查看授課大綱

您將學習的技能

Python Programming, Linear Programming (LP), Np-Completeness, Dynamic Programming

審閱

4.6（328 個評分）

5 stars
232 ratings
4 stars
66 ratings
3 stars
17 ratings
2 stars
6 ratings
1 star
7 ratings

NN

Jan 04, 2017

Loved what I learnt, I also implemented a project using Google MAP API for the organization I'm working at

TS

Jun 08, 2019

Very good collection of advanced topics, even useful for the 6th course in the specialization!

從本節課中

Streaming Algorithms (Optional)

In most previous lectures we were interested in designing algorithms with fast (e.g. small polynomial) runtime, and assumed that the algorithm has random access to its input, which is loaded into memory. In many modern applications in big data analysis, however, the input is so large that it cannot be stored in memory. Instead, the input is presented as a stream of updates, which the algorithm scans while maintaining a small summary of the stream seen so far. This is precisely the setting of the streaming model of computation, which we study in this lecture. The streaming model is well-suited for designing and reasoning about small space algorithms. It has received a lot of attention in the literature, and several powerful algorithmic primitives for computing basic stream statistics in this model have been designed, several of them impacting the practice of big data analysis. In this lecture we will see one such algorithm (CountSketch), a small space algorithm for finding the top k most frequent items in a data stream.

Introduction5:23

Heavy Hitters Problem7:33

Reduction 14:35

Reduction 26:30

Basic Estimate 18:04

Basic Estimate 27:14

Final Algorithm 15:27

Final Algorithm 212:57

教學方

Alexander S. Kulikov
Visiting Professor
Michael Levin
Lecturer
Daniel M Kane
Assistant Professor
Neil Rhodes
Adjunct Faculty

腳本

In this part of the lecture, we will prove our main Lemma on the precision guarantees

of the final version of our proximal point query primitive.

Now this Lemma is shown on the slide.

It says that if the number of hash buckets that we hash into is chosen to

be sufficiently large and if the number of repetitions is logarithmic in the length

of the stream then our approximate point query primitive gives an approximation

to the frequency count of every data item in the universe at every position in

the stream that is correct up to an additive epsilon times fk term.

Where fk is the frequency of the kth most frequent item in the stream.

Okay, well the failure probability in the statement will be inverse polynomial

in the length of the stream.

And because of that, instead of proving the Lemma for every point p between 1 and

N in the stream, it in fact suffices to prove it for

the last position in the stream.

Because after that we can apply the Lemma to every prefix of the original stream and

get the Lemma from the previous slide.

Okay, so this is the statement that we would like to prove.

Well, let us start by observing that in fact for every row i of our rEC.

That is for every of the key hash functions, h r,

that we chose by the basic analysis of our basic version of the approximate

point query primitive we have that every single one of the estimators

that we're taking the median off has the right mean.

That is for our data item i, which we consider fixed from now on.

For every r between 1 and t,

the expectation of the rth estimator is exactly the frequency of the ith item.

And now this is very good, it is reassuring,

but it doesn't tell us that this estimator is actually better than the basic version

of the estimator that we had before.

And the distinguishing feature of our new estimator counts in reduced variants.

Specifically by the analysis of our basic estimator applied to every single one of

the little t estimators that were taking the median off we have that for every i.

And for every fixed r from 1 to t, the variants of our rth estimator for

the frequency f i is equal to the following quantity.

It is the summation over all other data items j different from i, that hashed

into the same bucket under the r hash function of the frequency, fj squared.

And note, that the summation, those

only over those j's in the data universe that hash to the same bucket as i.

Now okay, so our hash functions or hashing the data universe into b buckets.

So we should expect the right-hand side to be about a factor of 1 over b

smaller in expectation than in our basic analysis.

And this is where our gains will come from.

Okay, so we would like to show that the variance is small.

And we would like to show that it at least reduces by roughly a factor of b.

In fact, we'll be able to do a bit better than that.

Now specifically, we consider the variance of our estimator.

And we separately consider the contribution of the head element

of the stream to this variance and the contribution of the tail elements.

So we write the sum over all data items j different from i that hashed under

the same bucket, under the r hashing, and we think r is fixed from now on.

We write the summation as two separate summations.

The first one is a summation over data items j in the head of the stream.

Think of the heavy hitters that are different from i and

again hash into the same buckets of f j squared.

So recall that the head of the stream

is exactly the top k most frequent elements in the stream.

The second term is a summation over the tail of the string, so

the less frequent elements that occur in the stream of fj squared.

And this is again only over those j that hash to the same bucket as i

under the r hush function.

Now, our plan is as follows.

We would like to bound these two terms.

What we will do is for each fixed i and for each data item i,

we consider both fixed from now on, we will define the following three events.

The first event is an event that we will call no collisions of r and i.

This event occurs if and

only if our data item i does not collide with any of the top k

most frequent elements in the stream that is the head of the stream under hashing r.

The second event is

a small variance event, again it indexed by the hash function index r and i itself.

This event happens, if and only if,

i does not collide with too many of the tail items under hashing r.

So note, that the first event, if it occurs,

implies that the first summation above is just 0.

There are not other events in the head that i collides with.

And the second event will insure that essentially and

the second summation is fairly small.

After that we will also need a third event which we denote small

deviation of r and i.

And this is essentially the success event from our

basic analysis from the outcome of probabilities inequalities.

What we will show in the next few minutes is that in fact for every r and

every data item i, all three of these event on the slide hold

simultaneously with probabilities strictly better than one half.

And after that we'll later show that this implies that if

we take a logarithmic number of independent trials,

the median estimate will give us a good estimate with high probability.

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera 致力於普及全世界最好的教育，它與全球一流大學和機構合作提供在線課程。

© 2019 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取

Coursera
關於
管理團隊
工作機會
目錄
證書
MasterTrack™ 證書
學位
企業版
政府版

社區
Learners
Partners
Developers
Beta Testers
Translators

連接
博客
Facebook
領英
Twitter
Instagram
YouTube
技術博客

更多
條款
隱私
幫助
內容訪問
媒體
聯繫我們
目錄
附屬公司