CS246: Mining Massive Datasets Jure Leskovec, Stanford University
|
|
- Andra McKenzie
- 5 years ago
- Views:
Transcription
1 CS246: Mining Massive Datasets Jure Leskovec, Stanford University
2 Web advertising We discussed how to match advertisers to queries in real-time But we did not discuss how to estimate the CTR (Click-Through Rate) Recommendation engines We discussed how to build recommender systems But we did not discuss the cold start problem 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 2
3 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 3 What do CTR and cold start have in common? With every ad we show/ product we recommend we gather more data about the ad/product Theme: Learning through experimentation
4 Google s goal: Maximize revenue The old way: Pay by impression (CPM) Best strategy: Go with the highest bidder But this ignores effectiveness of an ad The new way: Pay per click! (CPC) Best strategy: Go with expected revenue What s the expected revenue of ad a for query q? E[revenue a,q ] = P(click a q) * amount a,q Prob. user will click on ad a given that she issues query q (Unknown! Need to gather information) Bid amount for ad a on query q (Known) 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 4
5 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 5 Clinical trials: Investigate effects of different treatments while minimizing patient losses Adaptive routing: Minimize delay in the network by investigating different routes Asset pricing: Figure out product prices while trying to make most money
6 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 6
7 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 7
8 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 8 Each arm a Wins (reward=1) with fixed (unknown) prob. μ a Loses (reward=0) with fixed (unknown) prob. 1-μ a All draws are independent given μ 1 μ k How to pull arms to maximize total reward?
9 How does this map to our setting? Each query is a bandit Each ad is an arm We want to estimate the arm s probability of winning μ a (i.e., ad s the CTR μ a ) Every time we pull an arm we do an experiment 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 9
10 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 10 The setting: Set of k choices (arms) Each choice a is associated with unknown probability distribution P a supported in [0,1] We play the game for T rounds In each round t: (1) We pick some arm j (2) We obtain random sample X t from P j Note reward is independent of previous draws TT tt=11 XX tt Our goal is to maximize But we don t know μ a! But every time we pull some arm a we get to learn a bit about μ a
11 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 11 Online optimization with limited feedback Choices X 1 X 2 X 3 X 4 X 5 X 6 a a a k 0 Like in online algorithms: Time Have to make a choice each time But we only receive information about the chosen action
12 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 12 Policy: a strategy/rule that in each iteration tells me which arm to pull Hopefully policy depends on the history of rewards How to quantify performance of the algorithm? Regret!
13 Let be μμ aa the mean of PP aa Payoff/reward of best arm: μμ = mmmmmm aa μμ aa Let ii 11, ii 22 ii TT be the sequence of arms pulled Instantaneous regret at time tt: rr tt = μμ μμ aatt Total regret: TT RR TT = rr tt tt=11 Typical goal: Want a policy (arm allocation strategy) that guarantees: RR TT TT 00 as TT 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 13
14 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 14 If we knew the payoffs, which arm would we pull? PPPPPPPP aaaaaa mmmmmm μμ aa aa What if we only care about estimating payoffs μμ aa? Pick each of kk arms equally often: TT kk Estimate: μμ aa = kk TT TT/kk jj=11 XX aa,jj Regret: RR TT = TT kk(μμ μμ kk aa aa )
15 Regret is defined in terms of average reward So, if we can estimate avg. reward we can minimize regret Consider algorithm: Greedy Take the action with the highest avg. reward Example: Consider 2 actions A1 reward 1 with prob. 0.3 A2 has reward 1 with prob. 0.7 Play A1, get reward 1 Play A2, get reward 0 Now avg. reward of A1 will never drop to 0, and we will never play action A2 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 15
16 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 16 The example illustrates a classic problem in decision making: We need to trade off exploration (gathering data about arm payoffs) and exploitation (making decisions based on data already gathered) The Greedy does not explore sufficiently Exploration: Pull an arm we never pulled before Exploitation: Pull an arm aa for which we currently have the highest estimate of μμ aa
17 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 17 The problem with our Greedy algorithm is that it is too certain in the estimate of μμ aa When we have seen a single reward of 0 we shouldn t conclude the average reward is 0 Greedy does not explore sufficiently!
18 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 18 Algorithm: Epsilon-Greedy For t=1:t Set εε tt = OO(11/tt) With prob. εε tt : Explore by picking an arm chosen uniformly at random With prob. 11 εε tt : Exploit by picking an arm with highest empirical mean payoff Theorem [Auer et al. 02] For suitable choice of εε tt it holds that RR TT = OO(kk log TT) RR TT TT = OO kk log TT TT 0
19 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 19 What are some issues with Epsilon Greedy? Not elegant : Algorithm explicitly distinguishes between exploration and exploitation More importantly: Exploration makes suboptimal choices (since it picks any arm equally likely) Idea: When exploring/exploiting we need to compare arms
20 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 20 Suppose we have done experiments: Arm 1: Arm 2: 1 Arm 3: Mean arm values: Arm 1: 5/10, Arm 2: 1, Arm 3: 8/10 Which arm would you pick next? Idea: Don t just look at the mean (that is, expected payoff) but also the confidence!
21 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 21 A confidence interval is a range of values within which we are sure the mean lies with a certain probability We could believe μμ aa is within [0.2,0.5] with probability 0.95 If we would have tried an action less often, our estimated reward is less accurate so the confidence interval is larger Interval shrinks as we get more information (try the action more often)
22 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 22 Assuming we know the confidence intervals Then, instead of trying the action with the highest mean we can try the action with the highest upper bound on its confidence interval This is called an optimistic policy We believe an action is as good as possible given the available evidence
23 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, % confidence interval μμ ii After more exploration μμ ii arm i arm i
24 Goal: Want to bound PP( μμ ii μμ aa,mm bb) 3/6/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 24 Suppose we fix arm a: Let YY aa,11 YY aa,mm be the payoffs of arm a in the first m trials So, YY aa,11 YY aa,mm are i.i.d. rnd. vars. taking values in [0,1] Mean payoff of arm a: μμ aa = EE[YY aa,mm ] Our estimate: μμ aa,mm = 11 mm l=11 Want to find bb such that with high probability μμ aa μμ aa,mm bb mm YY aa,l Also want bb to be as small as possible (why?)
25 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 25 Hoeffding s inequality: Let XX 11 XX mm be i.i.d. rnd. vars. taking values in [0,1] Let μμ = EE[XX] and μμ mm = 11 mm XX mm l=11 l Then: PP μμ μμ mm bb 22 eeeeee 22bb 22 mm = δδ To find out the confidence interval bb (for a given confidence level δδ) we solve 2ee 2bb2mm δδ then 2bb 2 mm ln (δδ/2) So: bb llll 22 δδ 22 mm
26 [Auer et al. 02] 3/6/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 26 UCB1 (Upper confidence sampling) algorithm Set: μμ 11 = = μμ kk = 00 and mm 11 = = mm kk = 00 μμ aa is our estimate of payoff of arm ii mm aa is the number of pulls of arm ii so far For t = 1:T Upper confidence interval (Hoeffding s inequality) For each arm a calculate: UUUUUU aa = μμ aa + αα Pick arm jj = aaaaaa mmmmxx aa UUUUUU aa 22 ln tt mm aa Pull arm jj and observe yy tt Set: mm jj mm jj + 11 and μμ jj 11 (yy mm tt + jj mm jj 11 μμ ) jj
27 3/6/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 28 UUUUUU aa = μμ aa + αα 22 ln tt mm aa Confidence interval grows with the total number of actions tt we have taken But shrinks with the number of times mm aa we have tried arm aa This ensures each arm is tried infinitely often but still balances exploration and exploitation αα plays the role of δδ: αα = ff 22 δδ Optimism in face of uncertainty The algorithm believes that it can obtain extra rewards by reaching the unexplored parts of the state space bb llll 22 δδ 22 mm
28 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 29 Theorem [Auer et al. 2002] Suppose optimal mean payoff is μμ = mmmmmm aa And for each arm let ΔΔ aa = μμ μμ aa Then it holds that llll TT EE RR TT = ΔΔ aa aa:μμ aa <μμ ππ kk ΔΔ aa ii=aa μμ aa So: OO RR TT TT OO(kk ln TT) = kk llll TT TT OO(kk)
29 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 30 k-armed bandit problem as a formalization of the exploration-exploitation tradeoff Analog of online optimization (e.g., SGD, BALANCE), but with limited feedback Simple algorithms are able to achieve no regret (in the limit) Epsilon-greedy UCB (Upper Confidence Sampling)
30 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 31 Every round receive context [Li et al., WWW 10] Context: User features, articles view before Model for each article s click-through rate
31 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 32 Feature-based exploration: Select articles to serve users based on contextual information about the user and the articles Simultaneously adapt article selection strategy based on user-click feedback to maximize total number of user clicks
32 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 33 Contextual bandit algorithm in round t (1) Algorithm observes user u t and a set A of arms together with their features x t,a Vector x t,a summarizes both the user u t and arm a We call vector x t,a the context (2) Based on payoffs from previous trials, algorithm chooses arm a A and receives payoff r t,a Note only feedback for the chosen a is observed (3) Algorithm improves arm selection strategy with each observation (xx tt,aa, aa, rr tt,aa )
33 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 34 Payoff of arm a: EE rr tt,aa xx tt,aa = xx T tt,aa x t,a d-dimensional feature vector θθ aa θθ aa unknown coefficient vector we aim to learn Note that θθ aa are not shared between different arms! What s the difference between LinUCB, UCB1? UCB2 directly estimates μμ aa through experimentation (without any knowledge about arm aa) LinUCB estimates μμ aa by regression μμ aa = xx TT tt,aa The hope is that we will be able to learn faster as we consider the context xx aa (user, ad) of arm aa θθ aa
34 Payoff of arm a: EE rr tt,aa xx tt,aa = xx T tt,aa θθ aa x t,a d-dimensional feature vector θθ aa unknown coefficient vector we aim to learn How to estimate θθ aa? DD aa mm dd matrix of mm training inputs [xx aa,tt ] bb aa mm-dim. vector of responses to a (click/no-click) Linear regression solution to θθ aa is then θθ aa = aaaaaa mmmmnn θθ mm DD aa Which is solved by: θθ aa = xx T (mm) tt,aa θθ aa bb 22 aa DD aa T DD aa + II dd 1 DDaa TT bb aa I d is d x d identity matrix 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 35
35 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 36 One can then show (using similar techniques as we used for UCB) that So LinUCB arm selection rule is: Estimated μμ aa Confidence interval: Standard deviation
36 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 37 Initialization: For each arm aa: AA aa = II dd bb aa = 00 dd identity matrix m x m vector of zeros Online algorithm: For t = 1, 2, 3, T: Observe features of all arms aa xx tt,aa RR dd For each arm aa: 11 θθ aa = AA aa bb aa regression coefficients pp tt,aa = θθ TT TT 11 aa xx tt,aa + αα xx tt,aa AA aa xx tt,aa Choose arm aa tt = aaaaaa mmmmmm pp tt,aa choose arm aa confidence bound TT AA aatt = AA aatt + xx tt,aatt xx tt,aatt update AA for the chosen arm aa tt bb aatt = bb aa tt + rr tt xx tt,aatt updated bb for the chosen arm aa tt
37 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 38 LinUCB computational complexity is Linear in the number of arms and At most cubic in the number of features LinUCB works well for a dynamic arm set (arms come and go): For example, in news article recommendation, for instance, editors add/remove articles to/from a pool
38 What to put in slots F1, F2, F3, F4 to make the user click? 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 40
39 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 41
40 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 42 Want to choose a set that caters to as many users as possible Users may have different interests, queries may be ambiguous Want to optimize both the relevance and diversity
41 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 43
42 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 44 Alternate final: Tue 3/14 7:00-10:00pm in Cubberly Auditorium Register here: We have 100 slots. First come first serve! Final: Mon 3/17 12:15-3:15pm in NVidia (lastname starting with A-J), GatesB01 (K-S), and Packard 101 (T-Z) See Practice finals are posted on Piazza! SCPD students can take the exam at Stanford!
43 3/5/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, 45 Exam protocol for SCPD students: On Friday 3/14 your exam proctor will receive the PDF of the final exam from SCPD If you take the exam at Stanford: Ask the exam monitor to delete the SCP If you don t take the exam at Stanford: Arrange 3h slot with your exam monitor You can take the exam anytime but return it in time exam PDF to cs246.mmds@gmail.com by Tuesday 3/15 11:59pm Pacific time
An introduction to multi-armed bandits
An introduction to multi-armed bandits Henry WJ Reeve (Manchester) (henry.reeve@manchester.ac.uk) A joint work with Joe Mellor (Edinburgh) & Professor Gavin Brown (Manchester) Plan 1. An introduction to
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu Can we identify node groups? (communities, modules, clusters) 2/13/2014 Jure Leskovec, Stanford C246: Mining
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 2/24/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 2 High dim. data
More information1.2 Round-off Errors and Computer Arithmetic
1.2 Round-off Errors and Computer Arithmetic 1 In a computer model, a memory storage unit word is used to store a number. A word has only a finite number of bits. These facts imply: 1. Only a small set
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu [Kumar et al. 99] 2/13/2013 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu
More informationMonte Carlo Tree Search PAH 2015
Monte Carlo Tree Search PAH 2015 MCTS animation and RAVE slides by Michèle Sebag and Romaric Gaudel Markov Decision Processes (MDPs) main formal model Π = S, A, D, T, R states finite set of states of the
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
We need your help with our research on human interpretable machine learning. Please complete a survey at http://stanford.io/1wpokco. It should be fun and take about 1min to complete. Thanks a lot for your
More informationA Contextual-Bandit Approach to Personalized News Article Recommendation
A Contextual-Bandit Approach to Personalized News Article Recommendation Lihong Li,WeiChu, Yahoo! Labs lihong,chuwei@yahooinc.com John Langford Yahoo! Labs jl@yahoo-inc.com Robert E. Schapire + + Dept
More informationLecture 27: Review. Reading: All chapters in ISLR. STATS 202: Data mining and analysis. December 6, 2017
Lecture 27: Review Reading: All chapters in ISLR. STATS 202: Data mining and analysis December 6, 2017 1 / 16 Final exam: Announcements Tuesday, December 12, 8:30-11:30 am, in the following rooms: Last
More informationUsing Reinforcement Learning to Optimize Storage Decisions Ravi Khadiwala Cleversafe
Using Reinforcement Learning to Optimize Storage Decisions Ravi Khadiwala Cleversafe Topics What is Reinforcement Learning? Exploration vs. Exploitation The Multi-armed Bandit Optimizing read locations
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 3/12/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 2 3/12/2014 Jure
More informationSimilar Polygons Date: Per:
Math 2 Unit 6 Worksheet 1 Name: Similar Polygons Date: Per: [1-2] List the pairs of congruent angles and the extended proportion that relates the corresponding sides for the similar polygons. 1. AA BB
More informationReliability Considerations in Cyber-Power Dependent Systems
Reliability Considerations in Cyber-Power Dependent Systems Visvakumar Aravinthan Wichita State University (visvakumar.aravinthan@wichita.edu) PSERC Webinar April 17, 2018 1 Acknowledgement This work was
More informationWhere we are. Exploratory Graph Analysis (40 min) Focused Graph Mining (40 min) Refinement of Query Results (40 min)
Where we are Background (15 min) Graph models, subgraph isomorphism, subgraph mining, graph clustering Eploratory Graph Analysis (40 min) Focused Graph Mining (40 min) Refinement of Query Results (40 min)
More informationLecture 27: Learning from relational data
Lecture 27: Learning from relational data STATS 202: Data mining and analysis December 2, 2017 1 / 12 Announcements Kaggle deadline is this Thursday (Dec 7) at 4pm. If you haven t already, make a submission
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 3/6/2012 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 2 In many data mining
More informationCME323 Report: Distributed Multi-Armed Bandits
CME323 Report: Distributed Multi-Armed Bandits Milind Rao milind@stanford.edu 1 Introduction Consider the multi-armed bandit (MAB) problem. In this sequential optimization problem, a player gets to pull
More informationHow to Register for Summer Camp. A Tutorial
How to Register for Summer Camp A Tutorial Table of Contents 1. Logging In 2 2. OU student faculty, or staff or Previous visitor 3 3. New User Account 4 4. Summer Camp Offerings 5 5. Summer Camp Page 6
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 2/25/2013 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 3 In many data mining
More informationHow to Register for Summer Camp. A Tutorial
How to Register for Summer Camp A Tutorial 1. Upon arriving at our website (https://flightcamp.ou.edu/), the very first step is logging in. Please click the Login link in the top left corner of the page
More informationONLINE websites currently provide a vast number
1 Online Learning in Large-scale Contextual Recommender Systems Linqi Song, Cem Tekin, and Mihaela van der Schaar, Fellow, IEEE Abstract In this paper, we propose a novel large-scale, context-aware recommender
More informationGaussian Mixture Models For Clustering Data. Soft Clustering and the EM Algorithm
Gaussian Mixture Models For Clustering Data Soft Clustering and the EM Algorithm K-Means Clustering Input: Observations: xx ii R dd ii {1,., NN} Number of Clusters: kk Output: Cluster Assignments. Cluster
More informationMarkov Decision Processes (MDPs) (cont.)
Markov Decision Processes (MDPs) (cont.) Machine Learning 070/578 Carlos Guestrin Carnegie Mellon University November 29 th, 2007 Markov Decision Process (MDP) Representation State space: Joint state x
More informationSection 1: Introduction to Geometry Points, Lines, and Planes
Section 1: Introduction to Geometry Points, Lines, and Planes Topic 1: Basics of Geometry - Part 1... 3 Topic 2: Basics of Geometry Part 2... 5 Topic 3: Midpoint and Distance in the Coordinate Plane Part
More informationBland-Altman Plot and Analysis
Chapter 04 Bland-Altman Plot and Analysis Introduction The Bland-Altman (mean-difference or limits of agreement) plot and analysis is used to compare two measurements of the same variable. That is, it
More informationMonte Carlo Tree Search
Monte Carlo Tree Search 2-15-16 Reading Quiz What is the relationship between Monte Carlo tree search and upper confidence bound applied to trees? a) MCTS is a type of UCB b) UCB is a type of MCTS c) both
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS6: Mining Massive Datasets Jure Leskovec, Stanford University http://cs6.stanford.edu /7/0 Jure Leskovec, Stanford CS6: Mining Massive Datasets, http://cs6.stanford.edu High dim. data Graph data Infinite
More informationKDD 10 Tutorial: Recommender Problems for Web Applications. Deepak Agarwal and Bee-Chung Chen Yahoo! Research
KDD 10 Tutorial: Recommender Problems for Web Applications Deepak Agarwal and Bee-Chung Chen Yahoo! Research Agenda Focus: Recommender problems for dynamic, time-sensitive applications Content Optimization
More informationEureka Math. Grade 7, Module 6. Student File_A. Contains copy-ready classwork and homework
A Story of Ratios Eureka Math Grade 7, Module 6 Student File_A Contains copy-ready classwork and homework Published by the non-profit Great Minds. Copyright 2015 Great Minds. No part of this work may be
More informationMonte Carlo Tree Search
Monte Carlo Tree Search Branislav Bošanský PAH/PUI 2016/2017 MDPs Using Monte Carlo Methods Monte Carlo Simulation: a technique that can be used to solve a mathematical or statistical problem using repeated
More informationInformation Filtering for arxiv.org
Information Filtering for arxiv.org Bandits, Exploration vs. Exploitation, & the Cold Start Problem Peter Frazier School of Operations Research & Information Engineering Cornell University with Xiaoting
More informationApplying Multi-Armed Bandit on top of content similarity recommendation engine
Applying Multi-Armed Bandit on top of content similarity recommendation engine Andraž Hribernik andraz.hribernik@gmail.com Lorand Dali lorand.dali@zemanta.com Dejan Lavbič University of Ljubljana dejan.lavbic@fri.uni-lj.si
More informationInstructor: Dr. Mehmet Aktaş. Mining of Massive Datasets Jure Leskovec, Anand Rajaraman, Jeff Ullman Stanford University
Instructor: Dr. Mehmet Aktaş Mining of Massive Datasets Jure Leskovec, Anand Rajaraman, Jeff Ullman Stanford University J. Leskovec, A. Rajaraman, J. Ullman: Mining of Massive Datasets, http://www.mmds.org
More informationPolynomial Functions I
Name Student ID Number Group Name Group Members Polnomial Functions I 1. Sketch mm() =, nn() = 3, ss() =, and tt() = 5 on the set of aes below. Label each function on the graph. 15 5 3 1 1 3 5 15 Defn:
More informationBandit-based Search for Constraint Programming
Bandit-based Search for Constraint Programming Manuel Loth 1,2,4, Michèle Sebag 2,4,1, Youssef Hamadi 3,1, Marc Schoenauer 4,2,1, Christian Schulte 5 1 Microsoft-INRIA joint centre 2 LRI, Univ. Paris-Sud
More informationSequence alignment is an essential concept for bioinformatics, as most of our data analysis and interpretation techniques make use of it.
Sequence Alignments Overview Sequence alignment is an essential concept for bioinformatics, as most of our data analysis and interpretation techniques make use of it. Sequence alignment means arranging
More informationTransform Extreme Point Multi-Objective Linear Programming Problem to Extreme Point Single Objective Linear Programming Problem by Using Harmonic Mean
Applied Mathematics 2016, 6(5): 95-99 DOI: 10.5923/j.am.20160605.01 Transform Extreme Point Multi-Objective Linear Programming Problem to Extreme Point Single Objective Linear Programming Problem by Using
More informationCS224W: Social and Information Network Analysis Jure Leskovec, Stanford University, y http://cs224w.stanford.edu Due in 1 week: Oct 4 in class! The idea of the reaction papers is: To familiarize yourselves
More informationMulti-view object segmentation in space and time. Abdelaziz Djelouah, Jean Sebastien Franco, Edmond Boyer
Multi-view object segmentation in space and time Abdelaziz Djelouah, Jean Sebastien Franco, Edmond Boyer Outline Addressed problem Method Results and Conclusion Outline Addressed problem Method Results
More informationCS5112: Algorithms and Data Structures for Applications
CS5112: Algorithms and Data Structures for Applications Lecture 3: Hashing Ramin Zabih Some figures from Wikipedia/Google image search Administrivia Web site is: https://github.com/cornelltech/cs5112-f18
More informationMining of Massive Datasets Jure Leskovec, Anand Rajaraman, Jeff Ullman Stanford University Infinite data. Filtering data streams
/9/7 Note to other teachers and users of these slides: We would be delighted if you found this our material useful in giving your own lectures. Feel free to use these slides verbatim, or to modify them
More informationQ.4 Properties of Quadratic Function and Optimization Problems
384 Q.4 Properties of Quadratic Function and Optimization Problems In the previous section, we examined how to graph and read the characteristics of the graph of a quadratic function given in vertex form,
More informationCS249: ADVANCED DATA MINING
CS249: ADVANCED DATA MINING Classification Evaluation and Practical Issues Instructor: Yizhou Sun yzsun@cs.ucla.edu April 24, 2017 Homework 2 out Announcements Due May 3 rd (11:59pm) Course project proposal
More informationSection 6: Triangles Part 1
Section 6: Triangles Part 1 Topic 1: Introduction to Triangles Part 1... 125 Topic 2: Introduction to Triangles Part 2... 127 Topic 3: rea and Perimeter in the Coordinate Plane Part 1... 130 Topic 4: rea
More informationIntroduction to Data Mining
Introduction to Data Mining Lecture #7: Recommendation Content based & Collaborative Filtering Seoul National University In This Lecture Understand the motivation and the problem of recommendation Compare
More informationPracticeAdmin Identity Guide. Last Updated 4/27/2015 Created by Vanessa Street
PracticeAdmin Identity Guide Last Updated 4/27/2015 Created by Vanessa Street About PracticeAdmin Mission At PracticeAdmin, we simplify the complex process of medical billing by providing healthcare professionals
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University.
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 2 Instructor: Jure Leskovec TAs: Aditya Parameswaran Bahman Bahmani Peyman Kazemian 3 Course website: http://cs246.stanford.edu
More informationCS224W: Analysis of Networks Jure Leskovec, Stanford University
HW2 Q1.1 parts (b) and (c) cancelled. HW3 released. It is long. Start early. CS224W: Analysis of Networks Jure Leskovec, Stanford University http://cs224w.stanford.edu 10/26/17 Jure Leskovec, Stanford
More informationOPTIMAL DESIGN OF A STEPPER-DRIVEN PLANAR LINKAGE USING ENTROPY METHODS
Optimal Design of A Stepper-Driven Planar Linkage Using Entropy Methods OPTIMAL DESIGN OF A STEPPER-DRIVEN PLANAR LINKAGE USING ENTROPY METHODS J. C. Musto 1* 1 Mechanical Engineering Department, Milwaukee
More informationMining Data Streams. Outline [Garofalakis, Gehrke & Rastogi 2002] Introduction. Summarization Methods. Clustering Data Streams
Mining Data Streams Outline [Garofalakis, Gehrke & Rastogi 2002] Introduction Summarization Methods Clustering Data Streams Data Stream Classification Temporal Models CMPT 843, SFU, Martin Ester, 1-06
More informationVisit MathNation.com or search "Math Nation" in your phone or tablet's app store to watch the videos that go along with this workbook!
Topic 1: Introduction to Angles - Part 1... 47 Topic 2: Introduction to Angles Part 2... 50 Topic 3: Angle Pairs Part 1... 53 Topic 4: Angle Pairs Part 2... 56 Topic 5: Special Types of Angle Pairs Formed
More informationHomework 2. Due: March 2, 2018 at 7:00PM. p = 1 m. (x i ). i=1
Homework 2 Due: March 2, 2018 at 7:00PM Written Questions Problem 1: Estimator (5 points) Let x 1, x 2,..., x m be an i.i.d. (independent and identically distributed) sample drawn from distribution B(p)
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS6: Mining Massive Datasets Jure Leskovec, Stanford University http://cs6.stanford.edu Customer X Buys Metalica CD Buys Megadeth CD Customer Y Does search on Metalica Recommender system suggests Megadeth
More informationCS224W: Social and Information Network Analysis Jure Leskovec, Stanford University
CS224W: Social and Information Network Analysis Jure Leskovec, Stanford University http://cs224w.stanford.edu Setting from the last class: AB-A : gets a AB-B : gets b AB-AB : gets max(a, b) Also: Cost
More informationMore Summer Program t-shirts
ICPSR Blalock Lectures, 2003 Bootstrap Resampling Robert Stine Lecture 2 Exploring the Bootstrap Questions from Lecture 1 Review of ideas, notes from Lecture 1 - sample-to-sample variation - resampling
More informationExam 2 is Tue Nov 21. Bring a pencil and a calculator. Discuss similarity to exam1. HW3 is due Tue Dec 5.
Stat 100a: Introduction to Probability. Outline for the day 1. Bivariate and marginal density. 2. CLT. 3. CIs. 4. Sample size calculations. 5. Review for exam 2. Exam 2 is Tue Nov 21. Bring a pencil and
More informationME5286 Robotics Spring 2015 Quiz 1
Page 1 of 7 ME5286 Robotics Spring 2015 Quiz 1 Total Points: 30 You are responsible for following these instructions. Please take a minute and read them completely. 1. Put your name on this page, any other
More informationFast Convergence of Regularized Learning in Games
Fast Convergence of Regularized Learning in Games Vasilis Syrgkanis Alekh Agarwal Haipeng Luo Robert Schapire Microsoft Research NYC Microsoft Research NYC Princeton University Microsoft Research NYC Strategic
More informationCS224W: Social and Information Network Analysis Jure Leskovec, Stanford University
CS224W: Social and Information Network Analysis Jure Leskovec Stanford University Jure Leskovec, Stanford University http://cs224w.stanford.edu [Morris 2000] Based on 2 player coordination game 2 players
More informationThe History of I/O Models Erik Demaine
The History of I/O Models Erik Demaine MASSACHUSETTS INSTITUTE OF TECHNOLOGY Memory Hierarchies in Practice CPU 750 ps Registers 100B Level 1 Cache 100KB Level 2 Cache 1MB 10GB 14 ns Main Memory 1EB-1ZB
More informationReal-world bandit applications: Bridging the gap between theory and practice
Real-world bandit applications: Bridging the gap between theory and practice Audrey Durand EWRL 2018 The bandit setting (Thompson, 1933; Robbins, 1952; Lai and Robbins, 1985) Action Agent Context Environment
More informationThanks to Jure Leskovec, Anand Rajaraman, Jeff Ullman
Thanks to Jure Leskovec, Anand Rajaraman, Jeff Ullman http://www.mmds.org Overview of Recommender Systems Content-based Systems Collaborative Filtering J. Leskovec, A. Rajaraman, J. Ullman: Mining of Massive
More informationGaussian Mixture Models method and applications
Gaussian Mixture Models method and applications Jesús Zambrano PostDoctoral Researcher School of Business, Society and Engineering www.mdh.se FUDIPO project. Machine Learning course. Oct.-Dec. 2017 Outline
More informationConcept of Curve Fitting Difference with Interpolation
Curve Fitting Content Concept of Curve Fitting Difference with Interpolation Estimation of Linear Parameters by Least Squares Curve Fitting by Polynomial Least Squares Estimation of Non-linear Parameters
More informationReinforcement Learning: A brief introduction. Mihaela van der Schaar
Reinforcement Learning: A brief introduction Mihaela van der Schaar Outline Optimal Decisions & Optimal Forecasts Markov Decision Processes (MDPs) States, actions, rewards and value functions Dynamic Programming
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS6: Mining Massive Datasets Jure Leskovec, Stanford University http://cs6.stanford.edu Training data 00 million ratings, 80,000 users, 7,770 movies 6 years of data: 000 00 Test data Last few ratings of
More informationBox-Cox Transformation
Chapter 190 Box-Cox Transformation Introduction This procedure finds the appropriate Box-Cox power transformation (1964) for a single batch of data. It is used to modify the distributional shape of a set
More informationarxiv: v1 [cs.cv] 2 Sep 2018
Natural Language Person Search Using Deep Reinforcement Learning Ankit Shah Language Technologies Institute Carnegie Mellon University aps1@andrew.cmu.edu Tyler Vuong Electrical and Computer Engineering
More informationHomework 2: Search and Optimization
Scott Chow ROB 537: Learning Based Control October 16, 2017 Homework 2: Search and Optimization 1 Introduction The Traveling Salesman Problem is a well-explored problem that has been shown to be NP-Complete.
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu 1/8/2014 Jure Leskovec, Stanford CS246: Mining Massive Datasets, http://cs246.stanford.edu 2 Supermarket shelf
More informationIn section 8.1, we began by introducing the sine function using a circle in the coordinate plane:
Chapter 8.: Degrees and Radians, Reference Angles In section 8.1, we began by introducing the sine function using a circle in the coordinate plane: y (3,3) θ x We now return to the coordinate plane, but
More informationD-Optimal Designs. Chapter 888. Introduction. D-Optimal Design Overview
Chapter 888 Introduction This procedure generates D-optimal designs for multi-factor experiments with both quantitative and qualitative factors. The factors can have a mixed number of levels. For example,
More informationBox-Cox Transformation for Simple Linear Regression
Chapter 192 Box-Cox Transformation for Simple Linear Regression Introduction This procedure finds the appropriate Box-Cox power transformation (1964) for a dataset containing a pair of variables that are
More informationFormal Model. Figure 1: The target concept T is a subset of the concept S = [0, 1]. The search agent needs to search S for a point in T.
Although this paper analyzes shaping with respect to its benefits on search problems, the reader should recognize that shaping is often intimately related to reinforcement learning. The objective in reinforcement
More informationInternational Journal of Industrial Engineering Computations
International Journal of Industrial Engineering Computations 7 (2016) 295 308 Contents lists available at GrowingScience International Journal of Industrial Engineering Computations homepage: www.growingscience.com/ijiec
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS46: Mining Massive Datasets Jure Leskovec, Stanford University http://cs46.stanford.edu /7/ Jure Leskovec, Stanford C46: Mining Massive Datasets Many real-world problems Web Search and Text Mining Billions
More informationThe ABC s of Web Site Evaluation
Aa Bb Cc Dd Ee Ff Gg Hh Ii Jj Kk Ll Mm Nn Oo Pp Qq Rr Ss Tt Uu Vv Ww Xx Yy Zz The ABC s of Web Site Evaluation by Kathy Schrock Digital Literacy by Paul Gilster Digital literacy is the ability to understand
More informationstriatum Documentation
striatum Documentation Release 0.0.1 Y.-Y. Yang, Y.-A. Lin November 11, 2016 Contents 1 API Reference 3 1.1 striatum.bandit package......................................... 3 1.2 striatum.storage package.........................................
More informationInvitation to fixed-parameter algorithms
Invitation to fixed-parameter algorithms Jisu Jeong (Dept. of Math, KAIST) joint work with Sigve Hortemo Sæther and Jan Arne Telle (Univ. of Bergen, Norway), Eun Jung Kim (CNRS / Univ. Paris-Dauphine)
More informationLarge Scale Data Analysis Using Deep Learning
Large Scale Data Analysis Using Deep Learning Machine Learning Basics - 1 U Kang Seoul National University U Kang 1 In This Lecture Overview of Machine Learning Capacity, overfitting, and underfitting
More informationZero-Inflated Poisson Regression
Chapter 329 Zero-Inflated Poisson Regression Introduction The zero-inflated Poisson (ZIP) regression is used for count data that exhibit overdispersion and excess zeros. The data distribution combines
More informationNCSS Statistical Software
Chapter 327 Geometric Regression Introduction Geometric regression is a special case of negative binomial regression in which the dispersion parameter is set to one. It is similar to regular multiple regression
More informationCS3600 SYSTEMS AND NETWORKS
CS3600 SYSTEMS AND NETWORKS NORTHEASTERN UNIVERSITY Lecture 15: Networking overview Prof. (amislove@ccs.neu.edu) What is a network? 2 What is a network? Wikipedia: A telecommunications network is a network
More informationMODEL SELECTION AND REGULARIZATION PARAMETER CHOICE
MODEL SELECTION AND REGULARIZATION PARAMETER CHOICE REGULARIZATION METHODS FOR HIGH DIMENSIONAL LEARNING Francesca Odone and Lorenzo Rosasco odone@disi.unige.it - lrosasco@mit.edu June 3, 2013 ABOUT THIS
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS6: Mining Massive Datasets Jure Leskovec, Stanford University http://cs6.stanford.edu //8 Jure Leskovec, Stanford CS6: Mining Massive Datasets High dim. data Graph data Infinite data Machine learning
More informationReminder: Homework 4. Due: Friday at the beginning of class
Reminder: Homework 4 Due: Friday at the beginning of class 1 Cryptography CS 555 Topic 33: Digital Signatures Part 2 2 Recap El-Gamal/RSA-OAEP Digital Signatures Similarities and differences with MACs
More informationCS246: Mining Massive Datasets Jure Leskovec, Stanford University
CS246: Mining Massive Datasets Jure Leskovec, Stanford University http://cs246.stanford.edu /2/8 Jure Leskovec, Stanford CS246: Mining Massive Datasets 2 Task: Given a large number (N in the millions or
More informationMath 381 Discrete Mathematical Modeling
Math 381 Discrete Mathematical Modeling Sean Griffin Today: -Equipment Replacement Problem -Min Spanning Tree Problem -Clustering Friday s Plan Intro to LPSolve software Download before class (link on
More informationLesson 12: Angles Associated with Parallel Lines
Lesson 12 Lesson 12: Angles Associated with Parallel Lines Classwork Exploratory Challenge 1 In the figure below, LL 1 is not parallel to LL 2, and mm is a transversal. Use a protractor to measure angles
More informationRobust Linear Regression (Passing- Bablok Median-Slope)
Chapter 314 Robust Linear Regression (Passing- Bablok Median-Slope) Introduction This procedure performs robust linear regression estimation using the Passing-Bablok (1988) median-slope algorithm. Their
More informationIntroduction to Data Mining
Introduction to Data Mining Lecture #1: Course Introduction U Kang Seoul National University U Kang 1 In This Lecture Motivation to study data mining Administrative information for this course U Kang 2
More informationEureka Math. Geometry, Module 4. Student File_B. Contains Exit Ticket, and Assessment Materials
A Story of Functions Eureka Math Geometry, Module 4 Student File_B Contains Exit Ticket, and Assessment Materials Published by the non-profit Great Minds. Copyright 2015 Great Minds. No part of this work
More informationDS504/CS586: Big Data Analytics Big Data Clustering Prof. Yanhua Li
Welcome to DS504/CS586: Big Data Analytics Big Data Clustering Prof. Yanhua Li Time: 6:00pm 8:50pm Thu Location: AK 232 Fall 2016 High Dimensional Data v Given a cloud of data points we want to understand
More informationUnit 1. Name. Basics of Geometry Part 1. 2 Section 1: Introduction to Geometry Points, Lines and Planes
Name Period Date Points, lines, and planes are the building blocks of geometry. What is geometry? Unit 1 Basics of Geometry Part 1 Geometry means, and it involves the properties of points, lines, planes
More informationECS289: Scalable Machine Learning
ECS289: Scalable Machine Learning Cho-Jui Hsieh UC Davis Sept 22, 2016 Course Information Website: http://www.stat.ucdavis.edu/~chohsieh/teaching/ ECS289G_Fall2016/main.html My office: Mathematical Sciences
More informationTabu Search for the Dynamic Bipartite Drawing Problem
Tabu Search for the Dynamic Bipartite Drawing Problem RAFAEL MARTÍ Departamento de Estadística e Investigación Operativa, Universidad de Valencia, Spain Rafael.Marti@uv.es ANNA MARTÍNEZ-GAVARA Departamento
More informationIntroduction to Deep Learning
ENEE698A : Machine Learning Seminar Introduction to Deep Learning Raviteja Vemulapalli Image credit: [LeCun 1998] Resources Unsupervised feature learning and deep learning (UFLDL) tutorial (http://ufldl.stanford.edu/wiki/index.php/ufldl_tutorial)
More informationWisconsin Retirement Testing Preparation
Wisconsin Retirement Testing Preparation The Wisconsin Retirement System (WRS) is changing its reporting requirements from annual to every pay period starting January 1, 2018. With that, there are many
More informationTABLE OF CONTENTS. 3 Intro. 4 Foursquare Logo. 6 Foursquare Icon. 9 Colors. 10 Copy & Tone Of Voice. 11 Typography. 13 Crown Usage.
BRANDBOOK TABLE OF CONTENTS 3 Intro 4 Foursquare Logo 6 Foursquare Icon 9 Colors 10 Copy & Tone Of Voice 11 Typography 13 Crown Usage 14 Badge Usage 15 Iconography 16 Trademark Guidelines 2011 FOURSQUARE
More informationrecruitment Logo Typography Colourways Mechanism Usage Pip Recruitment Brand Toolkit
Logo Typography Colourways Mechanism Usage Primary; Secondary; Silhouette; Favicon; Additional Notes; Where possible, use the logo with the striped mechanism behind. Only when it is required to be stripped
More information