Network Performance Tomography
|
|
- Amberlynn Hunter
- 6 years ago
- Views:
Transcription
1 Network Performance Tomography Hung X. Nguyen TeleTraffic Research Center University of Adelaide
2 Network Performance Tomography Inferring link performance using end-to-end probes 2
3 Network Performance Tomography Inferring link performance using end-to-end probes Concentrate on only loss rates in this talk 3
4 End-to-End Loss Rates Are not Enough Path transmission rates and link transmission rates are linearly related φ: path transmission rate A tr: link transmission rate e 1 φ 1 φ 2 A->B A->C e 1 e 2 e * log(tr 1 ) log(tr 2 ) log(tr 3 ) = log(φ 1 ) log(φ 2 ) B e 2 e 3 C Under-determined Routing matrix : does not have full column rank Link transmission rates cannot be calculated using only end-to-end transmission rates 4
5 Network Performance Tomography Performance tomography: is all about obtaining additional information to solve the inverse problem and inference methods given the data. Assumptions: Fixed topology: Known routing matrix all columns are distinct all columns have at least one 1 entry Loss independence: Link losses are independent (both temporally and spatially) tr 1 φ 1 =0.72 φ 2 =0.8 B A->B A->C A tr 2 tr 3 e 1 e 2 e C 5
6 Temporal Correlation Using Multicast Multicast multicast probes: probes are forwarded to multiple receivers at the branching point strong temporal correlation A->B A->C A->B A->C e 1 e 2 e * log(tr 1 ) log(tr 2 ) log(tr 3 ) = log(φ 1 ) log(φ 2 ) log(φ 2 1 ) Multicast router 1" 1" 0" 0" B A tr 1 " tr 2 "tr 3 " C 1" 0" 1" 0" Temporal correlation ^ tr ^ ^ 1,tr 2, tr 3 " 6
7 Inference Algorithm Maximum likelihood algorithm to infer loss rates on a multicast tree [Caceres et al 99] n : number of probes n(01) : number of 01 outcomes, etc p(01) = n(01) n,etc γ 1 = p(11) + p(10) + p(01) γ 2 = p(11) + p(10) γ 3 = p(11) + p(01) γ tr 1 = 2 γ 3,tr γ 2 +γ 3 γ 2 =1 γ γ 1 2,tr 1 γ 3 =1 γ γ γ 2 tr 2 B Can be extended to work with multiple trees [Bu et al. 01] Limitation: Multicast is not widely accessible to end users 1" 1" 0" 0" A tr 1 C tr 3 1" 0" 1" 0" 7
8 Unicast Emulation of Multicast Using back-to-back unicast packets to create temporal correlation A router β 1 B C Additional variable: the conditional probability β 1 = Pr(red _ succeeds blue _ succeeds) 8
9 Treat β as variables: Inference Algorithms Infer the conditional probabilities β together with link loss rates using the EM (expectation-maximization) method [Coates 00] n b,r m b,r Likelihood : l(m b,r n b,r, p b,r ) = n b,r m p b,r m m b,r b,r (1 p b,r b,r ) n b,r m b,r, where : Number of packet pairs (blue,red) blue succeeds : Number of successful packet pairs (blue,red)-both succeed p b,r = β 1 * tr 2 Inference: argmax tr,β l(m N,tr,β) 9
10 Inference Algorithms (cont.) Infer the conditional probabilities β together with link loss rates But the model is not identifiable The estimator is biased, loss rates of internal links are overestimated Alternatively, force perfect correlation [Duffield et al. 01] β 1 = Pr(red blueblueblue) 1 Apply the multicast based inference algorithms Less accurate than the multicast approach and highly dependent on back ground traffic 10
11 Without temporal correlation In most large scale systems, temporal correlation is hard to enforce Multicast: may not be available Packet pairs/trains: hard to scale, not accurate End-to-end transmission rates are easy to obtained Identifiability issue without temporal correlation A->B A->C e 1 e 2 e * log(tr 1 ) log(tr 2 ) log(tr 3 ) = log(φ 1 ) log(φ 2 ) 11
12 Pragmatic Goals Pragmatic goals Loss rates of groups of consecutive links [Zhao et al. 06] tr 1 *tr 2 and tr 1 *tr 3 In many cases, can compute loss rates for small sets of links Identifying only the worst performing links [Padmanabhan et al. 03, Duffield 06] e 1 is congested (tr 1 <t l ), e 2 and e 3 are good(tr 2,tr 3 >t l ) where t l is the link threshold 12
13 Boolean Tomography Objective is to locate congested links (links with transmission rate tr<t l ) Input: - Routing topologies - Probing results - The threshold t l Output - The set of congested links X 2 =1 X 1 =0 A A->B A->C X 3 =0 e 1 e 2 e B C
14 Assumptions T1: Known routing matrix all columns are distinct all columns have at least one 1 entry S1: Loss independence: all losses are temporally and spatially independent S2: All flows traverse a link e i have transmission rate tr i on that link tr 1 φ 1 =0.72 φ 2 =0.8 B A->B A->C A tr 2 tr 3 e 1 e 2 e C 14
15 Explore All Possible Loss Rates Simulate all possible link transmission probabilities that are consistent with the end-to-end measurements (Padmanabhan et al. 03) Random sampling Monte Carlo Markov Chain Simulation: Construct a Markov chain whose stationary distribution is Pr( tr 1,tr 2,tr 3 probes) Slow to converge, why does it work? A tr B tr 2 tr 3 C (1,0.99, 0.75) (0.99,1, 0.76) (0.999,0.99, 0.75) Lossy links = links with majority of sample rates <0.8 Lossy links = {e 3 }
16 Boolean System of Equations Transform the linear equations into Boolean equations [Duffield 03]: A->B A->C e 1 e 2 e * log(tr 1 ) log(tr 2 ) log(tr 3 ) = log(φ 1 ) log(φ 2 ) (+,x) (max,min) Y i =1 if φ 1 < t p, Y i =0 otherwise X i =1 if tr 1 < t l, X i =0 otherwise e 1 e 2 e 3 A->B A->C X 1 X 2 X 3 = Y 1 Y 2 16
17 Smallest Consistence Failure Set (SCFS) Assumptions: 1. (Performance separability) a path is bad (φ 1 < t p ) if and only if at least one link is bad (tr 1 < t l ) 2. Bad links are rare On a tree topology [Duffield 03]?? x x 0?? 0??
18 Performance of SCFS DR: detection rate, FPR: false positive rate Tree with 1000 nodes, 1000 probes, bad links have loss rate uniformly in [0.05,1], good links: [0,0.01 ] 18
19 Back to SCFS Objective is to locate congested links (links with transmission rate tr<t l ) End-to-end measurements link e k is congested iff tr k < t l =0.9 e 1 φ 1 =0.72 φ 2 =0.8 e 2 e 3 congested congested (Y 1 =1) (Y 2 =1) congested (X 1 =1) Not congested Not congested (X 2 =0) (X 3 =0) Assumptions: congested links are rare, links are equally likely to be congested. Can be very inaccurate: biases in favor of shared links 19
20 Time-Varying Boolean Tomography Link quality changes with time
21 Time-Varying Boolean Solution To overcome the biases, link state probabilities (Nguyen 07): Methodology: p k = P(X k =1) = P(tr k <t l ) Step 1 (learning phase): Take multiple snapshots to learn the link state probabilities Step 2 (diagnosis phase): Then determine the most probable set of congested links in the current snapshot p 1 = P(X 1 =1) p 2 = P(X 2 =1) p 3 =P(X 3 =1) Y 1 =1 Y 2 =1 21
22 Identifiability of Link State Probability Theorem (Identifiability): with the previous assumptions, link state probabilities p k can be uniquely learnt from endto-end measurements if and only if 0 p k < 1 p 1 = P(X 1 =1) =0 p 2 = P(X 2 =1) =0.3 p 3 =P(X 3 =1) =0.6 Follow the proof technique of Vardi, JASA,
23 Step 1: Estimating Link State Probabilities Theorem (Identifiability): link state probabilities p k can be uniquely learnt from end-to-end measurements if 0 p k <1 p 1 =? p 2 =? Method of moments: -Vardi, JASA, E[Y 1 ] = P(Y 1 =1) = 1-(1-p 1 )*(1-p 2 ); - E[Y 2 ] = P(Y 2 =1) = 1-(1-p 1 )*(1-p 3 ); - E[Y 1 *Y 2 ] = E[max(X 1,X 2,X 3 )] = p 1 *(1-p 2 )*(1-p 3 ) +p 1 *p 2 *(1-p 3 ) +p 1 *(1-p 2 )*p 3 +(1-p 1 )*p 2 *p 3 p 3 =? [ Y 1 Y 1 Y 1... ] [Y 2 Y 2 Y 2... ] Non linear equations, intractable!!!
24 Step 1: Method of Moments Estimator Taking the second order moments in Boolean algebra: E[max(Y 1,Y 2 )]=P(max(Y 1,Y 2 )=1)= P(max(X 1,X 2,X 3 )=1) =1-P(X 1 =0)*P(X 2 =0)*P(X 3 =0) =1-(1-p 1 )*(1-p 2 )*(1-p 3 ) * log(1-p 1 ) log(1-p 2 ) log(1-p 3 ) = log(1-e[y 1 ]) log(1-e[y 2 ]) log(1-e[max(y 1,Y 2 )]) Full rank linear system!!! We can go up to third, fourth, etc. order moments. Conjecture: Second order moments are enough to obtain a system of full rank in all networks.
25 Step 2: Identifying congested links A simple optimization problem to find the set of congested links: ( ) k arg min X log 1 p k pk Xk X k = 1 subject to: at least one bad link is on a bad path End-to-end measurements link e k is congested iff tr k < t l =0.9 e 1 p 1 =0, p 2 =0.3, p 3 =0.6 Not congested tr=0.72 tr =0.72 congested congested e 2 e 3 congested congested
26 Performance DR: detection rate, FPR: false positive rate Tree with 1000 nodes, 1000 probes, bad links have loss rate uniformly in [0.05,1], good links: [0,0.01 ] 26
27 Loss Rates on the Internet Observation 1: Most links have negligible loss rates (tr k = 1) 0n PlanetLab, more than 80% of end-to-end paths have zero loss rates Observation 2: congested links have high loss rate variances [Paxson Sigcomm 97, Zhang IMC 01] 27
28 Time-Varying Loss Tomography Calculate the link loss rates from end-to-end loss rates without using probe temporal correlation Input: - Routing matrix - End-to-end transmission rates Output - Link transmission rates tr 2 tr 1 e 2 e 1 e 3 tr 3 snapshot 28
29 Assumptions T1: Known routing matrix - all columns are distinct - all columns have at least one 1 entry S1: Link independence: tr k are independent S2: Identical sample rates: φ i,k = tr k (a.s) for all path i and link e k A->B A->C S3: Monotonic relationship between mean and variance of link loss rates: v k = VAR(log(tr k ) ) is a non-decreasing function of E(1-tr k ) e 1 e 2 e
30 Time-Varying Tomography Solution To overcome the ill-posed problem, identify links with negligible loss rate variances: v k ~ 0 tr k ~ 1 Methodology: Step 1 (learning phase): Take multiple snapshots to learn the link variances Step 2 (diagnosis phase): Recursively assigning zero loss rates to links with smallest variances to reduce the diagnosis equations to a linear system of full-rank 30
31 Step 1: Estimating Link Variances Calculate link variances: A * v 1 v 2 v 3 = VAR(log(φ 1 )) VAR(log(φ 2 )) COV(log(φ 1 ), log(φ 2 )) v 1 [ φ 1 φ 1 φ 1 ] [φ 2 φ 2 φ 2 ] v 2 v 3 Full-rank linear system B C Sort links according to their variances v 3 v 2 v 1 Links with smaller variances are less congested 31
32 Step 2: Calculating Link Loss Rates Recursively eliminating good links from the first order equations * log(tr 1 ) log(tr 2 ) log(tr 3 ) = log(φ 1 ) log(φ 2 ) v 3 v 2 v 1 tr 1 ~ 1 Until we obtain a full column rank system Solve the resulting system using standard linear algebra technique 32
33 Summary Network performance tomography is all about finding extra information to overcome the inidentifiability problem Two major approaches: Using temporal correlation (multicast or Unicast) Using prior information about link loss rate Boolean Tomography Time-varying tomography There are still much work to be done before large scale diagnosis systems can be built using these techniques: Topology changes Inaccurate end-to-end measurements 33
Active Measurement for Multiple Link Failures Diagnosis in IP Networks
Active Measurement for Multiple Link Failures Diagnosis in IP Networks Hung X. Nguyen and Patrick Thiran EPFL CH-1015 Lausanne, Switzerland Abstract. Simultaneous link failures are common in IP networks
More informationTWO PROBLEMS IN NETWORK PROBING
TWO PROBLEMS IN NETWORK PROBING DARRYL VEITCH The University of Melbourne 1 Temporal Loss and Delay Tomography 2 Optimal Probing in Convex Networks Paris Networking 27 Juin 2007 TEMPORAL LOSS AND DELAY
More informationMachine Learning for Data Science (CS4786) Lecture 24
Machine Learning for Data Science (CS4786) Lecture 24 Graphical Models: Approximate Inference Course Webpage : http://www.cs.cornell.edu/courses/cs4786/2016sp/ BELIEF PROPAGATION OR MESSAGE PASSING Each
More informationProbabilistic Graphical Models
Probabilistic Graphical Models Brown University CSCI 2950-P, Spring 2013 Prof. Erik Sudderth Lecture 13: Learning in Gaussian Graphical Models, Non-Gaussian Inference, Monte Carlo Methods Some figures
More informationAlgorithmisches Lernen/Machine Learning
Algorithmisches Lernen/Machine Learning Part 1: Stefan Wermter Introduction Connectionist Learning (e.g. Neural Networks) Decision-Trees, Genetic Algorithms Part 2: Norman Hendrich Support-Vector Machines
More informationECEN 689 Special Topics in Data Science for Communications Networks
ECEN 689 Special Topics in Data Science for Communications Networks Nick Duffield Department of Electrical & Computer Engineering Texas A&M University Lecture 13 Measuring and Inferring Traffic Matrices
More informationLayer 0. α 1. Path P(0,2) Layer 1. α 2 α 3. Layer 2. α 4. α 5. Layer 3 = Receiver Set
onference on Information Sciences and Systems, The Johns Hopkins University, March 4, Least Squares Estimates of Network Link Loss Probabilities using End-to-end Multicast Measurements owei Xi, George
More informationEstimating Internal Link Loss Rates Using Active Network Tomography
Estimating Internal Link Loss Rates Using Active Network Tomography by Bowei Xi A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy (Statistics) in
More informationArtificial Intelligence
ICS461 Fall 2010 Nancy E. Reed nreed@hawaii.edu 1 Lecture #14B Outline Inference in Bayesian Networks Exact inference by enumeration Exact inference by variable elimination Approximate inference by stochastic
More informationMulticast-Based Inference of Network-Internal Loss Characteristics
Multicast-Based Inference of Network-Internal Loss Characteristics R. Cáceres y N.G. Duffield z J. Horowitz x D. Towsley { Abstract Robust measurements of network dynamics are increasingly important to
More informationDynamic resource sharing
J. Virtamo 38.34 Teletraffic Theory / Dynamic resource sharing and balanced fairness Dynamic resource sharing In previous lectures we have studied different notions of fair resource sharing. Our focus
More informationGenerative v. Discriminative classifiers Intuition
Logistic Regression (Continued) Generative v. Discriminative Decision rees Machine Learning 10701/15781 Carlos Guestrin Carnegie Mellon University January 31 st, 2007 2005-2007 Carlos Guestrin 1 Generative
More informationMarkov Chains and MCMC
Markov Chains and MCMC CompSci 590.02 Instructor: AshwinMachanavajjhala Lecture 4 : 590.02 Spring 13 1 Recap: Monte Carlo Method If U is a universe of items, and G is a subset satisfying some property,
More information7.1 Coupling from the Past
Georgia Tech Fall 2006 Markov Chain Monte Carlo Methods Lecture 7: September 12, 2006 Coupling from the Past Eric Vigoda 7.1 Coupling from the Past 7.1.1 Introduction We saw in the last lecture how Markov
More informationBayesian Networks BY: MOHAMAD ALSABBAGH
Bayesian Networks BY: MOHAMAD ALSABBAGH Outlines Introduction Bayes Rule Bayesian Networks (BN) Representation Size of a Bayesian Network Inference via BN BN Learning Dynamic BN Introduction Conditional
More informationSTA 4273H: Statistical Machine Learning
STA 4273H: Statistical Machine Learning Russ Salakhutdinov Department of Computer Science! Department of Statistical Sciences! rsalakhu@cs.toronto.edu! h0p://www.cs.utoronto.ca/~rsalakhu/ Lecture 7 Approximate
More informationHidden Markov models
Hidden Markov models Charles Elkan November 26, 2012 Important: These lecture notes are based on notes written by Lawrence Saul. Also, these typeset notes lack illustrations. See the classroom lectures
More informationThe Particle Filter. PD Dr. Rudolph Triebel Computer Vision Group. Machine Learning for Computer Vision
The Particle Filter Non-parametric implementation of Bayes filter Represents the belief (posterior) random state samples. by a set of This representation is approximate. Can represent distributions that
More informationProbabilistic Reasoning. Kee-Eung Kim KAIST Computer Science
Probabilistic Reasoning Kee-Eung Kim KAIST Computer Science Outline #1 Acting under uncertainty Probabilities Inference with Probabilities Independence and Bayes Rule Bayesian networks Inference in Bayesian
More informationUnsupervised Anomaly Detection for High Dimensional Data
Unsupervised Anomaly Detection for High Dimensional Data Department of Mathematics, Rowan University. July 19th, 2013 International Workshop in Sequential Methodologies (IWSM-2013) Outline of Talk Motivation
More informationMCMC and Gibbs Sampling. Kayhan Batmanghelich
MCMC and Gibbs Sampling Kayhan Batmanghelich 1 Approaches to inference l Exact inference algorithms l l l The elimination algorithm Message-passing algorithm (sum-product, belief propagation) The junction
More informationBayesian networks: approximate inference
Bayesian networks: approximate inference Machine Intelligence Thomas D. Nielsen September 2008 Approximative inference September 2008 1 / 25 Motivation Because of the (worst-case) intractability of exact
More informationProbabilistic Machine Learning
Probabilistic Machine Learning Bayesian Nets, MCMC, and more Marek Petrik 4/18/2017 Based on: P. Murphy, K. (2012). Machine Learning: A Probabilistic Perspective. Chapter 10. Conditional Independence Independent
More informationMarkov Networks. l Like Bayes Nets. l Graph model that describes joint probability distribution using tables (AKA potentials)
Markov Networks l Like Bayes Nets l Graph model that describes joint probability distribution using tables (AKA potentials) l Nodes are random variables l Labels are outcomes over the variables Markov
More informationSteps towards Decentralized Deterministic Network Coding
Steps towards Decentralized Deterministic Network Coding BY Oana Graur o.graur@jacobs-university.de Ph.D. Proposal in Electrical Engineering Ph.D Proposal Committee: Prof. Dr.-Ing. Werner Henkel, Dr. Mathias
More informationApproximate Inference
Approximate Inference Simulation has a name: sampling Sampling is a hot topic in machine learning, and it s really simple Basic idea: Draw N samples from a sampling distribution S Compute an approximate
More informationdistribution tree. Thus multicast trac introduces a well structured correlation in the end-to-end behavior observed by the receiver that share the sam
Network Delay Tomography from End-to-end Unicast Measurements? N.G. Dueld 1, J. Horowitz 2, F. Lo Presti 1;3, and D. Towsley 3 1 AT&T Labs{Research, 18 Park Avenue, Florham Park, NJ 7932, USA fduffield,loprestig@research.att.com
More informationComputer Intensive Methods in Mathematical Statistics
Computer Intensive Methods in Mathematical Statistics Department of mathematics johawes@kth.se Lecture 16 Advanced topics in computational statistics 18 May 2017 Computer Intensive Methods (1) Plan of
More informationMulticast-Based Inference of Network-Internal Delay Distributions
Multicast-Based Inference of Network-Internal Delay Distributions F Lo Presti NG Duffield J Horowitz D Towsley AT&T Labs Research Dept Math & Statistics Dept of Computer Science 18 Park Avenue University
More informationUnicast-Based Inference of Network Link Delay Distributions with Finite Mixture Models (Revision)
Unicast-Based Inference of Network Link Delay Distributions with Finite Mixture Models (Revision) Meng-Fu Shih, Alfred O Hero Submitted to IEEE Transactions on Signal Processing, Special Issue on Signal
More informationCS 188: Artificial Intelligence. Bayes Nets
CS 188: Artificial Intelligence Probabilistic Inference: Enumeration, Variable Elimination, Sampling Pieter Abbeel UC Berkeley Many slides over this course adapted from Dan Klein, Stuart Russell, Andrew
More informationCS 343: Artificial Intelligence
CS 343: Artificial Intelligence Bayes Nets: Sampling Prof. Scott Niekum The University of Texas at Austin [These slides based on those of Dan Klein and Pieter Abbeel for CS188 Intro to AI at UC Berkeley.
More informationRandomized Load Balancing:The Power of 2 Choices
Randomized Load Balancing: The Power of 2 Choices June 3, 2010 Balls and Bins Problem We have m balls that are thrown into n bins, the location of each ball chosen independently and uniformly at random
More informationMaximum Likelihood Estimation of the Flow Size Distribution Tail Index from Sampled Packet Data
Maximum Likelihood Estimation of the Flow Size Distribution Tail Index from Sampled Packet Data Patrick Loiseau 1, Paulo Gonçalves 1, Stéphane Girard 2, Florence Forbes 2, Pascale Vicat-Blanc Primet 1
More informationArtificial Intelligence
Artificial Intelligence Roman Barták Department of Theoretical Computer Science and Mathematical Logic Summary of last lecture We know how to do probabilistic reasoning over time transition model P(X t
More informationthe tree till a class assignment is reached
Decision Trees Decision Tree for Playing Tennis Prediction is done by sending the example down Prediction is done by sending the example down the tree till a class assignment is reached Definitions Internal
More informationLecture 2: Randomized Algorithms
Lecture 2: Randomized Algorithms Independence & Conditional Probability Random Variables Expectation & Conditional Expectation Law of Total Probability Law of Total Expectation Derandomization Using Conditional
More informationGraphical Models - Part I
Graphical Models - Part I Oliver Schulte - CMPT 726 Bishop PRML Ch. 8, some slides from Russell and Norvig AIMA2e Outline Probabilistic Models Bayesian Networks Markov Random Fields Inference Outline Probabilistic
More informationParsimonious Tomography: Optimizing Cost-Identifiability Trade-off for Probing-based Network Monitoring
Parsimonious Tomography: Optimizing Cost-Identifiability Trade-off for Probing-based Network Monitoring Diman Zad Tootaghaj, Ting He, Thomas La Porta The Pennsylvania State University {dxz149, tzh58, tlp}@cse.psu.edu
More informationAn Overview of Traffic Matrix Estimation Methods
An Overview of Traffic Matrix Estimation Methods Nina Taft Berkeley www.intel.com/research Problem Statement 1 st generation solutions 2 nd generation solutions 3 rd generation solutions Summary Outline
More informationDecision Trees. Machine Learning 10701/15781 Carlos Guestrin Carnegie Mellon University. February 5 th, Carlos Guestrin 1
Decision Trees Machine Learning 10701/15781 Carlos Guestrin Carnegie Mellon University February 5 th, 2007 2005-2007 Carlos Guestrin 1 Linear separability A dataset is linearly separable iff 9 a separating
More informationCPSC 540: Machine Learning
CPSC 540: Machine Learning Mark Schmidt University of British Columbia Winter 2018 Last Time: Monte Carlo Methods If we want to approximate expectations of random functions, E[g(x)] = g(x)p(x) or E[g(x)]
More informationAnnealing Between Distributions by Averaging Moments
Annealing Between Distributions by Averaging Moments Chris J. Maddison Dept. of Comp. Sci. University of Toronto Roger Grosse CSAIL MIT Ruslan Salakhutdinov University of Toronto Partition Functions We
More informationStephen Scott.
1 / 28 ian ian Optimal (Adapted from Ethem Alpaydin and Tom Mitchell) Naïve Nets sscott@cse.unl.edu 2 / 28 ian Optimal Naïve Nets Might have reasons (domain information) to favor some hypotheses/predictions
More informationHidden Markov models 1
Hidden Markov models 1 Outline Time and uncertainty Markov process Hidden Markov models Inference: filtering, prediction, smoothing Most likely explanation: Viterbi 2 Time and uncertainty The world changes;
More informationCS 7180: Behavioral Modeling and Decision- making in AI
CS 7180: Behavioral Modeling and Decision- making in AI Bayesian Networks for Dynamic and/or Relational Domains Prof. Amy Sliva October 12, 2012 World is not only uncertain, it is dynamic Beliefs, observations,
More informationAnnouncements. CS 188: Artificial Intelligence Fall Causality? Example: Traffic. Topology Limits Distributions. Example: Reverse Traffic
CS 188: Artificial Intelligence Fall 2008 Lecture 16: Bayes Nets III 10/23/2008 Announcements Midterms graded, up on glookup, back Tuesday W4 also graded, back in sections / box Past homeworks in return
More informationGaussian Mixture Models
Gaussian Mixture Models Pradeep Ravikumar Co-instructor: Manuela Veloso Machine Learning 10-701 Some slides courtesy of Eric Xing, Carlos Guestrin (One) bad case for K- means Clusters may overlap Some
More informationCPSC 540: Machine Learning
CPSC 540: Machine Learning Mark Schmidt University of British Columbia Winter 2019 Last Time: Monte Carlo Methods If we want to approximate expectations of random functions, E[g(x)] = g(x)p(x) or E[g(x)]
More informationSAMPLING AND INVERSION
SAMPLING AND INVERSION Darryl Veitch dveitch@unimelb.edu.au CUBIN, Department of Electrical & Electronic Engineering University of Melbourne Workshop on Sampling the Internet, Paris 2005 A TALK WITH TWO
More informationBayesian Inference and MCMC
Bayesian Inference and MCMC Aryan Arbabi Partly based on MCMC slides from CSC412 Fall 2018 1 / 18 Bayesian Inference - Motivation Consider we have a data set D = {x 1,..., x n }. E.g each x i can be the
More informationProbabilistic Reasoning. (Mostly using Bayesian Networks)
Probabilistic Reasoning (Mostly using Bayesian Networks) Introduction: Why probabilistic reasoning? The world is not deterministic. (Usually because information is limited.) Ways of coping with uncertainty
More informationLecture 4: State Estimation in Hidden Markov Models (cont.)
EE378A Statistical Signal Processing Lecture 4-04/13/2017 Lecture 4: State Estimation in Hidden Markov Models (cont.) Lecturer: Tsachy Weissman Scribe: David Wugofski In this lecture we build on previous
More informationProbabilistic Graphical Models
2016 Robert Nowak Probabilistic Graphical Models 1 Introduction We have focused mainly on linear models for signals, in particular the subspace model x = Uθ, where U is a n k matrix and θ R k is a vector
More informationPhylogenetics: Likelihood
1 Phylogenetics: Likelihood COMP 571 Luay Nakhleh, Rice University The Problem 2 Input: Multiple alignment of a set S of sequences Output: Tree T leaf-labeled with S Assumptions 3 Characters are mutually
More information26 : Spectral GMs. Lecturer: Eric P. Xing Scribes: Guillermo A Cidre, Abelino Jimenez G.
10-708: Probabilistic Graphical Models, Spring 2015 26 : Spectral GMs Lecturer: Eric P. Xing Scribes: Guillermo A Cidre, Abelino Jimenez G. 1 Introduction A common task in machine learning is to work with
More informationBayesian Networks Inference with Probabilistic Graphical Models
4190.408 2016-Spring Bayesian Networks Inference with Probabilistic Graphical Models Byoung-Tak Zhang intelligence Lab Seoul National University 4190.408 Artificial (2016-Spring) 1 Machine Learning? Learning
More informationEfficient MCMC Samplers for Network Tomography
Efficient MCMC Samplers for Network Tomography Martin Hazelton 1 Institute of Fundamental Sciences Massey University 7 December 2015 1 Email: m.hazelton@massey.ac.nz AUT Mathematical Sciences Symposium
More informationOn the Relationship between Sum-Product Networks and Bayesian Networks
On the Relationship between Sum-Product Networks and Bayesian Networks International Conference on Machine Learning, 2015 Han Zhao Mazen Melibari Pascal Poupart University of Waterloo, Waterloo, ON, Canada
More informationPerfect simulation of repulsive point processes
Perfect simulation of repulsive point processes Mark Huber Department of Mathematical Sciences, Claremont McKenna College 29 November, 2011 Mark Huber, CMC Perfect simulation of repulsive point processes
More informationLink Prediction. Eman Badr Mohammed Saquib Akmal Khan
Link Prediction Eman Badr Mohammed Saquib Akmal Khan 11-06-2013 Link Prediction Which pair of nodes should be connected? Applications Facebook friend suggestion Recommendation systems Monitoring and controlling
More informationCS6375: Machine Learning Gautam Kunapuli. Decision Trees
Gautam Kunapuli Example: Restaurant Recommendation Example: Develop a model to recommend restaurants to users depending on their past dining experiences. Here, the features are cost (x ) and the user s
More information4 : Exact Inference: Variable Elimination
10-708: Probabilistic Graphical Models 10-708, Spring 2014 4 : Exact Inference: Variable Elimination Lecturer: Eric P. ing Scribes: Soumya Batra, Pradeep Dasigi, Manzil Zaheer 1 Probabilistic Inference
More informationLecture 8: Bayesian Networks
Lecture 8: Bayesian Networks Bayesian Networks Inference in Bayesian Networks COMP-652 and ECSE 608, Lecture 8 - January 31, 2017 1 Bayes nets P(E) E=1 E=0 0.005 0.995 E B P(B) B=1 B=0 0.01 0.99 E=0 E=1
More informationModeling Residual-Geometric Flow Sampling
Modeling Residual-Geometric Flow Sampling Xiaoming Wang Joint work with Xiaoyong Li and Dmitri Loguinov Amazon.com Inc., Seattle, WA April 13 th, 2011 1 Agenda Introduction Underlying model of residual
More informationFactor Graphs and Message Passing Algorithms Part 1: Introduction
Factor Graphs and Message Passing Algorithms Part 1: Introduction Hans-Andrea Loeliger December 2007 1 The Two Basic Problems 1. Marginalization: Compute f k (x k ) f(x 1,..., x n ) x 1,..., x n except
More informationBayesian Congestion Control over a Markovian Network Bandwidth Process
Bayesian Congestion Control over a Markovian Network Bandwidth Process Parisa Mansourifard 1/30 Bayesian Congestion Control over a Markovian Network Bandwidth Process Parisa Mansourifard (USC) Joint work
More informationComputer Intensive Methods in Mathematical Statistics
Computer Intensive Methods in Mathematical Statistics Department of mathematics johawes@kth.se Lecture 5 Sequential Monte Carlo methods I 31 March 2017 Computer Intensive Methods (1) Plan of today s lecture
More informationMarkov Networks. l Like Bayes Nets. l Graphical model that describes joint probability distribution using tables (AKA potentials)
Markov Networks l Like Bayes Nets l Graphical model that describes joint probability distribution using tables (AKA potentials) l Nodes are random variables l Labels are outcomes over the variables Markov
More informationSequence labeling. Taking collective a set of interrelated instances x 1,, x T and jointly labeling them
HMM, MEMM and CRF 40-957 Special opics in Artificial Intelligence: Probabilistic Graphical Models Sharif University of echnology Soleymani Spring 2014 Sequence labeling aking collective a set of interrelated
More informationSandwiching the marginal likelihood using bidirectional Monte Carlo. Roger Grosse
Sandwiching the marginal likelihood using bidirectional Monte Carlo Roger Grosse Ryan Adams Zoubin Ghahramani Introduction When comparing different statistical models, we d like a quantitative criterion
More informationHidden Markov Models. AIMA Chapter 15, Sections 1 5. AIMA Chapter 15, Sections 1 5 1
Hidden Markov Models AIMA Chapter 15, Sections 1 5 AIMA Chapter 15, Sections 1 5 1 Consider a target tracking problem Time and uncertainty X t = set of unobservable state variables at time t e.g., Position
More informationOn the Quality of Wireless Network Connectivity
Globecom 2012 - Ad Hoc and Sensor Networking Symposium On the Quality of Wireless Network Connectivity Soura Dasgupta Department of Electrical and Computer Engineering The University of Iowa Guoqiang Mao
More informationEfficient Information Planning in Graphical Models
Efficient Information Planning in Graphical Models computational complexity considerations John Fisher & Giorgos Papachristoudis, MIT VITALITE Annual Review 2013 September 9, 2013 J. Fisher (VITALITE Annual
More informationSampling Algorithms for Probabilistic Graphical models
Sampling Algorithms for Probabilistic Graphical models Vibhav Gogate University of Washington References: Chapter 12 of Probabilistic Graphical models: Principles and Techniques by Daphne Koller and Nir
More informationInformatics 2D Reasoning and Agents Semester 2,
Informatics 2D Reasoning and Agents Semester 2, 2018 2019 Alex Lascarides alex@inf.ed.ac.uk Lecture 25 Approximate Inference in Bayesian Networks 19th March 2019 Informatics UoE Informatics 2D 1 Where
More informationVariational Learning : From exponential families to multilinear systems
Variational Learning : From exponential families to multilinear systems Ananth Ranganathan th February 005 Abstract This note aims to give a general overview of variational inference on graphical models.
More informationCSCE 478/878 Lecture 6: Bayesian Learning and Graphical Models. Stephen Scott. Introduction. Outline. Bayes Theorem. Formulas
ian ian ian Might have reasons (domain information) to favor some hypotheses/predictions over others a priori ian methods work with probabilities, and have two main roles: Naïve Nets (Adapted from Ethem
More informationCS711008Z Algorithm Design and Analysis
.. Lecture 6. Hidden Markov model and Viterbi s decoding algorithm Institute of Computing Technology Chinese Academy of Sciences, Beijing, China . Outline The occasionally dishonest casino: an example
More informationOutline. CSE 573: Artificial Intelligence Autumn Agent. Partial Observability. Markov Decision Process (MDP) 10/31/2012
CSE 573: Artificial Intelligence Autumn 2012 Reasoning about Uncertainty & Hidden Markov Models Daniel Weld Many slides adapted from Dan Klein, Stuart Russell, Andrew Moore & Luke Zettlemoyer 1 Outline
More informationIntroduction to Machine Learning
Introduction to Machine Learning Brown University CSCI 1950-F, Spring 2012 Prof. Erik Sudderth Lecture 25: Markov Chain Monte Carlo (MCMC) Course Review and Advanced Topics Many figures courtesy Kevin
More informationPhylogenetics: Bayesian Phylogenetic Analysis. COMP Spring 2015 Luay Nakhleh, Rice University
Phylogenetics: Bayesian Phylogenetic Analysis COMP 571 - Spring 2015 Luay Nakhleh, Rice University Bayes Rule P(X = x Y = y) = P(X = x, Y = y) P(Y = y) = P(X = x)p(y = y X = x) P x P(X = x 0 )P(Y = y X
More informationA Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement
A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement Simon Leglaive 1 Laurent Girin 1,2 Radu Horaud 1 1: Inria Grenoble Rhône-Alpes 2: Univ. Grenoble Alpes, Grenoble INP,
More informationFinal Exam, Machine Learning, Spring 2009
Name: Andrew ID: Final Exam, 10701 Machine Learning, Spring 2009 - The exam is open-book, open-notes, no electronics other than calculators. - The maximum possible score on this exam is 100. You have 3
More informationTarget Tracking and Classification using Collaborative Sensor Networks
Target Tracking and Classification using Collaborative Sensor Networks Xiaodong Wang Department of Electrical Engineering Columbia University p.1/3 Talk Outline Background on distributed wireless sensor
More informationLinear Dynamical Systems (Kalman filter)
Linear Dynamical Systems (Kalman filter) (a) Overview of HMMs (b) From HMMs to Linear Dynamical Systems (LDS) 1 Markov Chains with Discrete Random Variables x 1 x 2 x 3 x T Let s assume we have discrete
More informationShortest Paths & Link Weight Structure in Networks
Shortest Paths & Link Weight Structure in etworks Piet Van Mieghem CAIDA WIT (May 2006) P. Van Mieghem 1 Outline Introduction The Art of Modeling Conclusions P. Van Mieghem 2 Telecommunication: e2e A ETWORK
More informationCISC 889 Bioinformatics (Spring 2004) Hidden Markov Models (II)
CISC 889 Bioinformatics (Spring 24) Hidden Markov Models (II) a. Likelihood: forward algorithm b. Decoding: Viterbi algorithm c. Model building: Baum-Welch algorithm Viterbi training Hidden Markov models
More informationComputer Vision Group Prof. Daniel Cremers. 10a. Markov Chain Monte Carlo
Group Prof. Daniel Cremers 10a. Markov Chain Monte Carlo Markov Chain Monte Carlo In high-dimensional spaces, rejection sampling and importance sampling are very inefficient An alternative is Markov Chain
More information16.4 Multiattribute Utility Functions
285 Normalized utilities The scale of utilities reaches from the best possible prize u to the worst possible catastrophe u Normalized utilities use a scale with u = 0 and u = 1 Utilities of intermediate
More informationChapter 4 Dynamic Bayesian Networks Fall Jin Gu, Michael Zhang
Chapter 4 Dynamic Bayesian Networks 2016 Fall Jin Gu, Michael Zhang Reviews: BN Representation Basic steps for BN representations Define variables Define the preliminary relations between variables Check
More informationSequential Importance Sampling for Rare Event Estimation with Computer Experiments
Sequential Importance Sampling for Rare Event Estimation with Computer Experiments Brian Williams and Rick Picard LA-UR-12-22467 Statistical Sciences Group, Los Alamos National Laboratory Abstract Importance
More informationDecentralized Control of Stochastic Systems
Decentralized Control of Stochastic Systems Sanjay Lall Stanford University CDC-ECC Workshop, December 11, 2005 2 S. Lall, Stanford 2005.12.11.02 Decentralized Control G 1 G 2 G 3 G 4 G 5 y 1 u 1 y 2 u
More informationSymbolic Variable Elimination in Discrete and Continuous Graphical Models. Scott Sanner Ehsan Abbasnejad
Symbolic Variable Elimination in Discrete and Continuous Graphical Models Scott Sanner Ehsan Abbasnejad Inference for Dynamic Tracking No one previously did this inference exactly in closed-form! Exact
More informationBayes Nets III: Inference
1 Hal Daumé III (me@hal3.name) Bayes Nets III: Inference Hal Daumé III Computer Science University of Maryland me@hal3.name CS 421: Introduction to Artificial Intelligence 10 Apr 2012 Many slides courtesy
More informationBayesian Congestion Control over a Markovian Network Bandwidth Process: A multiperiod Newsvendor Problem
Bayesian Congestion Control over a Markovian Network Bandwidth Process: A multiperiod Newsvendor Problem Parisa Mansourifard 1/37 Bayesian Congestion Control over a Markovian Network Bandwidth Process:
More informationMEASUREMENTS that are telemetered to the control
2006 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 4, NOVEMBER 2004 Auto Tuning of Measurement Weights in WLS State Estimation Shan Zhong, Student Member, IEEE, and Ali Abur, Fellow, IEEE Abstract This
More informationA Brief Introduction to Graphical Models. Presenter: Yijuan Lu November 12,2004
A Brief Introduction to Graphical Models Presenter: Yijuan Lu November 12,2004 References Introduction to Graphical Models, Kevin Murphy, Technical Report, May 2001 Learning in Graphical Models, Michael
More informationOnline Bayesian Transfer Learning for Sequential Data Modeling
Online Bayesian Transfer Learning for Sequential Data Modeling....? Priyank Jaini Machine Learning, Algorithms and Theory Lab Network for Aging Research 2 3 Data of personal preferences (years) Data (non-existent)
More informationCS 188: Artificial Intelligence Fall Recap: Inference Example
CS 188: Artificial Intelligence Fall 2007 Lecture 19: Decision Diagrams 11/01/2007 Dan Klein UC Berkeley Recap: Inference Example Find P( F=bad) Restrict all factors P() P(F=bad ) P() 0.7 0.3 eather 0.7
More information