Introduction to Optimization

Size: px
Start display at page:

Download "Introduction to Optimization"

Transcription

1 Introduction to Optimization Constrained Optimization Marc Toussaint U Stuttgart

2 Constrained Optimization General constrained optimization problem: Let R n, f : R n R, g : R n R m, h : R n R l find min f() s.t. g() 0, h() = 0 In this lecture I ll focus (mostly) on inequality constraints g! Applications Find an optimal, non-colliding trajectory in robotics Optimize the shape of a turbine blade, s.t. it must not break Optimize the train schedule, s.t. consistency/possibility 2/39

3 Try to some how transform the constraint problem to a series of unconstraint problems a single but larger unconstraint problem another constraint problem, hopefully simpler (dual, conve) 3/39

4 General approaches Penalty & Barriers Associate a (adaptive) penalty cost with violation of the constraint Associate an additional force compensating the gradient into the constraint (augmented Lagrangian) Associate a log barrier with a constraint, becoming for violation (interior point method) Gradient projection methods (mostly for linear contraints) For active constraints, project the step direction to become tangantial When checking a step, always pull it back to the feasible region Lagrangian & dual methods Rewrite the constrained problem into an unconstrained one Or rewrite it as a (conve) dual problem Simple methods (linear constraints) Walk along the constraint boundaries 4/39

5 Penalties & Barries Convention: A barrier is really for g() > 0 A penalty is zero for g() 0 and increases with g() > 0 5/39

6 Log barrier method or Interior Point method 6/39

7 Log barrier method Instead of we address min f() s.t. g() 0 min f() µ i log( g i ()) 7/39

8 Log barrier For µ 0, µ log( g) converges to [g > 0] The barriers gradient log( g) = g g constraint Notation: [boolean epression] {0, 1} pushes away from the Eventually we want to have a very small µ but choosing small µ makes the barrier very non-smooth, which is bad for Gradient and 2nd order methods 8/39

9 Central Path Every µ defines a different optimal (µ) (µ) = argmin f() µ i log( g i ()) Each point on the path can be understood as the optimal compromise of minimizing f() and a repelling force of the constraints. (Which corresponds to dual variables λ (µ).) 9/39

10 Log barrier method Input: initial R n, functions f(), g(), f(), g(), tolerances θ, ɛ Output: 1: initialize µ = 1 2: repeat 3: find argmin f() µ i log( g i()) with tolerance 10θ 4: decrease µ µ/10 5: until < θ and i : g i () < ɛ Note: See Boyd & Vandenberghe for stopping criteria based on f precision (duality gap) and better choice of initial µ (which is called t there). 10/39

11 We will revisit the log barrier method later, once we introduced the Langrangian... 11/39

12 Squared Penalty Method 12/39

13 Squared Penalty Method This is perhaps the simplest approach Instead of we address min f() s.t. g() 0 min f() + µ m [g i () > 0] g i () 2 i=1 Input: initial R n, functions f(), g(), f(), g(), tol. θ, ɛ Output: 1: initialize µ = 1 2: repeat 3: find argmin f() + µ i [g i() > 0] g() 2 with tolerance 10θ 4: µ 10µ 5: until < θ and i : g i () < ɛ 13/39

14 Squared Penalty Method The method is ok, but will always lead to some violation of constraints 14/39

15 Squared Penalty Method The method is ok, but will always lead to some violation of constraints A better idea would be to add an out-pushing gradient/force g i () for every constraint g i () > 0 that is violated. Ideally, the out-pushing gradient mies with f() eactly such that the result becomes tangential to the constraint! This idea leads to the augmented Lagrangian approach. 14/39

16 Augmented Lagrangian (We can introduce this is a self-contained manner, without yet defining the Lagrangian ) 15/39

17 Augmented Lagrangian (equality constraint) We first consider an equality constraint before addressing inequalities Instead of min f() s.t. h() = 0 we address min f() + µ m i=1 h i () 2 + i=1 λ i h i () (1) Note: The gradient h i () is always orthogonal to the constraint By tuning λ i we can induce a virtual gradient λ i h i () The term µ m i=1 h i() 2 penalizes as before Here is the trick: First minimize (1) for some µ and λ i This will in general lead to a (slight) penalty µ m i=1 h i() 2 For the net iteration, choose λ i to generate eactly the gradient that was previously generated by the penalty 16/39

18 Optimality condition after an iteration: m m = argmin f() + µ h i() 2 + λ ih i() i=1 i=1 m m 0 = f( ) + µ 2h i( ) h i( ) + λ i h i( ) i=1 Update λ s for the net iteration: m λ new i h i( ) = µ i=1 λ new i i=1 i=1 2h i( ) h i( ) + i=1 = λ old i + 2µh i( ) λ old i h i( ) Input: initial R n, functions f(), h(), f(), h(), tol. θ, ɛ Output: 1: initialize µ = 1, λ i = 0 2: repeat 3: find argmin f() + µ i h i() 2 + i λ ih i () 4: i : λ i λ i + 2µh i ( ) 5: until < θ and h i () < ɛ 17/39

19 This adaptation of λ i is really elegant: We do not have to take the penalty limit µ but still can have eact constraints If f and h were linear ( f and h i constant), the updated λ i is eactly right: In the net iteration we would eactly hit the constraint (by construction) The penalty term is like a measuring device for the necessary virtual gradient, which is generated by the agumentation term in the net iteration The λ i are very meaningful: they give the force/gradient that a constraint eerts on the solution 18/39

20 Augmented Lagrangian Instead of we address min f() + µ i=1 (inequality constraint) min f() s.t. g() 0 m m [g i () 0 λ i > 0] g i () 2 + λ i g i () A constraint is either active or inactive: When active (g i () 0 λ i > 0) we aim for equality g i () = 0 When inactive (g i () < 0 λ i = 0) we don t penalize/augment λ i are zero or positive, but never negative i=1 Input: initial R n, functions f(), g(), f(), g(), tol. θ, ɛ Output: 1: initialize µ = 1, λ i = 0 2: repeat 3: find argmin f() + µ i [gi() 0 λi > 0] gi()2 + i λigi() 4: i : λ i ma(λ i + 2µg i( ), 0) 5: until < θ and g i() < ɛ 19/39

21 General approaches Penalty & Barriers Associate a (adaptive) penalty cost with violation of the constraint Associate an additional force compensating the gradient into the constraint (augmented Lagrangian) Associate a log-barrier with a constraint, becoming for violation (interior point method) Gradient projection methods (mostly for linear contraints) For active constraints, project the step direction to become tangantial When checking a step, always pull it back to the feasible region Lagrangian & dual methods Rewrite the constrained problem into an unconstrained one Or rewrite it as a (conve) dual problem Simple methods (linear constraints) Walk along the constraint boundaries 20/39

22 The Lagrangian 21/39

23 The Lagrangian Given a constraint problem we define the Lagrangian as min f() s.t. g() 0 L(, λ) = f() + m λ i g i () i=1 The λ i 0 are called dual variables or Lagrange multipliers 22/39

24 What s the point of this definition? The Lagrangian is useful to compute optima analytically, on paper that s why physicist learn it early on The Lagrangian implies the KKT conditions of optimality Optima are necessarily at saddle points of the Lagrangian The Lagrangian implies a dual problem, which is sometimes easier to solve than the primal 23/39

25 Eample: Some calculus using the Lagrangian For R 2, what is min 2 s.t = 1 Solution: L(, λ) = 2 + λ( ) 0 = L(, λ) = 2 + λ 1 1 = 2 = λ/2 1 0 = λ L(, λ) = = λ/2 λ/2 1 λ = 1 = 1 = 2 = 1/2 24/39

26 The force & KKT view on the Lagrangian At the optimum there must be a balance between the cost gradient f() and the gradient of the active constraints g i () 25/39

27 The force & KKT view on the Lagrangian At the optimum there must be a balance between the cost gradient f() and the gradient of the active constraints g i () Formally: for optimal : f() span{ g i()} [ ] Or: for optimal there must eist λ i such that f() = i ( λi gi()) 26/39

28 The force & KKT view on the Lagrangian At the optimum there must be a balance between the cost gradient f() and the gradient of the active constraints g i () Formally: for optimal : f() span{ g i()} [ ] Or: for optimal there must eist λ i such that f() = i ( λi gi()) For optimal it must hold (necessary condition): λ s.t. m f() + λ i g i () = 0 i=1 i : g i () 0 i : λ i 0 i : λ i g i () = 0 ( force balance ) (primal feasibility) (dual feasibility) (complementary) The last condition says that λ i > 0 only for active constraints. These are the Karush-Kuhn-Tucker conditions (KKT, neglecting equality constraints) 26/39

29 The force & KKT view on the Lagrangian The first condition ( force balance ), λ s.t. m f() + λ i g i () = 0 i=1 can be equivalently epressed as, λ s.t. L(, λ) = 0 In that sense, the Lagrangian can be viewed as the energy function that generates (for good choice of λ) the right balance between cost and constraint gradients This is eactly as in the augmented Lagrangian approach, where however we have an additional ( augmented ) squared penalty that is used to tune the λ i 27/39

30 Saddle point view on the Lagrangian Let s briefly consider the equality case again: with the Lagrangian Note: min f() s.t. h() = 0 L(, λ) = f() + m λ i h i () i=1 min L(, λ) 0 = L(, λ) force balance ma λ L(, λ) 0 = λl(, λ) = h i () constraint Optima (, λ ) are saddle points where L = 0 ensures force balance and λ L = 0 ensures the constraint 28/39

31 Saddle point view on the Lagrangian In the inequality case: ma L(, λ) = λ 0 ma L(, λ) λ i 0 f() if g() 0 otherwise λ i = 0 if g() < 0 0 = λi L(, λ) = g i (0) otherwise This implies either (λ i = 0 g i () < 0) or g i (0) = 0, which is eactly equivalent to the KKT conditions Again, optima (, λ ) are saddle points where min L enforces force balance and ma λ L enforces the KKT conditions 29/39

32 The Lagrange dual problem We define the Lagrange dual function as l(λ) = min L(, λ) This implies two problems min f() s.t. g() 0 primal problem ma l(λ) s.t. λ 0 dual problem λ The dual problem is conve, even if the primal is non-conve! Written more symmetric: min ma L(, λ) l 0 ma min λ 0 primal problem L(, λ) dual problem because the ma λ 0 L(, λ) ensures the constraints (previous slide). 30/39

33 The Lagrange dual problem The dual function is always a lower bound (for any λ i 0) [ ] l(λ) = min L(, λ) min f() s.t. g() 0 And consequently ma min λ 0 We say strong duality holds iff ma min λ 0 L(, λ) min L(, λ) = min ma L(, λ) l 0 ma L(, λ) l 0 If the primal is conve, and there eist an interior point : i : g i () < 0 (which is called Slater condition), then we have strong duality 31/39

34 And what about algorithms? So far we ve only introduced a whole lot of formalism, and seen that the Lagrangian sort of represents the constraint problem min L or L = 0 is related to the force balance ma λ L or λ L = 0 is related to constraints or KKT conditions This implies two dual problems, min ma λ L and ma λ min L, the second (dual) is a lower bound of the first (primal) But what are the algorithms we can get out of this? 32/39

35 Algorithmic implications of the Lagrangian view If min L(, λ) can be solved analytically, we can alternatively solve the (conve) dual problem. But more generally Optimization problem Solve KKT conditions Apply standard algos for solving an equation system r(, λ) = 0: Newton method r = r λ This leads to primal-dual algorithms that adapt and λ concurrently. Roughly, they use the curvature 2 f to estimate the right λ to push out of the constraint. We will discuss this after we ve learnt about 2nd order methods. 33/39

36 Log barrier method revisited 34/39

37 Log barrier method revisited Log barrier method: Instead of min f() s.t. g() 0 we address min f() µ i log( g i ()) For given µ the optimality condition is or equivalently f() i µ g i () g i() = 0 f() i λ i g i () = 0, λ i g i () = µ These are called modified (=approimate) KKT conditions. 35/39

38 Log barrier method revisited Centering (the unconstrained minimization) in the log barrier method is equivalent to solving the modified KKT conditions. Note also: On the central path, the duality gap is mµ: l(λ (µ)) = f( (µ)) + i λ ig i ( (µ)) = f( (µ)) mµ 36/39

39 Phase I: Finding a feasible initialization 37/39

40 Phase I: Finding a feasible initialization An elegant method for finding a feasible point : min s s.t. i : g i () s, s 0 (,s) R n+1 or min (,s) R n+m m s i s.t. i : g i () s i, s i 0 i=1 38/39

41 General approaches Penalty & Barriers Associate a (adaptive) penalty cost with violation of the constraint Associate an additional force compensating the gradient into the constraint (augmented Lagrangian) Associate a log barrier with a constraint, becoming for violation (interior point method) Gradient projection methods (mostly for linear contraints) For active constraints, project the step direction to become tangantial When checking a step, always pull it back to the feasible region Lagrangian & dual methods Rewrite the constrained problem into an unconstrained one Or rewrite it as a (conve) dual problem Simple methods (linear constraints) Walk along the constraint boundaries 39/39

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Second Order Optimization Methods Marc Toussaint U Stuttgart Planned Outline Gradient-based optimization (1st order methods) plain grad., steepest descent, conjugate grad.,

More information

LECTURE 13: SOLUTION METHODS FOR CONSTRAINED OPTIMIZATION. 1. Primal approach 2. Penalty and barrier methods 3. Dual approach 4. Primal-dual approach

LECTURE 13: SOLUTION METHODS FOR CONSTRAINED OPTIMIZATION. 1. Primal approach 2. Penalty and barrier methods 3. Dual approach 4. Primal-dual approach LECTURE 13: SOLUTION METHODS FOR CONSTRAINED OPTIMIZATION 1. Primal approach 2. Penalty and barrier methods 3. Dual approach 4. Primal-dual approach Basic approaches I. Primal Approach - Feasible Direction

More information

Programming, numerics and optimization

Programming, numerics and optimization Programming, numerics and optimization Lecture C-4: Constrained optimization Łukasz Jankowski ljank@ippt.pan.pl Institute of Fundamental Technological Research Room 4.32, Phone +22.8261281 ext. 428 June

More information

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Marc Toussaint April 19, 2016 This is a direct concatenation and reformatting of all lecture slides and eercises from the Optimization course (summer term 2015, U Stuttgart),

More information

Optimization III: Constrained Optimization

Optimization III: Constrained Optimization Optimization III: Constrained Optimization CS 205A: Mathematical Methods for Robotics, Vision, and Graphics Doug James (and Justin Solomon) CS 205A: Mathematical Methods Optimization III: Constrained Optimization

More information

Convex Optimization and Machine Learning

Convex Optimization and Machine Learning Convex Optimization and Machine Learning Mengliu Zhao Machine Learning Reading Group School of Computing Science Simon Fraser University March 12, 2014 Mengliu Zhao SFU-MLRG March 12, 2014 1 / 25 Introduction

More information

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Marc Toussaint July 2, 2014 This is a direct concatenation and reformatting of all lecture slides and eercises from the Optimization course (summer term 2014, U Stuttgart),

More information

Support Vector Machines. James McInerney Adapted from slides by Nakul Verma

Support Vector Machines. James McInerney Adapted from slides by Nakul Verma Support Vector Machines James McInerney Adapted from slides by Nakul Verma Last time Decision boundaries for classification Linear decision boundary (linear classification) The Perceptron algorithm Mistake

More information

COMS 4771 Support Vector Machines. Nakul Verma

COMS 4771 Support Vector Machines. Nakul Verma COMS 4771 Support Vector Machines Nakul Verma Last time Decision boundaries for classification Linear decision boundary (linear classification) The Perceptron algorithm Mistake bound for the perceptron

More information

Lecture Notes: Constraint Optimization

Lecture Notes: Constraint Optimization Lecture Notes: Constraint Optimization Gerhard Neumann January 6, 2015 1 Constraint Optimization Problems In constraint optimization we want to maximize a function f(x) under the constraints that g i (x)

More information

Computational Methods. Constrained Optimization

Computational Methods. Constrained Optimization Computational Methods Constrained Optimization Manfred Huber 2010 1 Constrained Optimization Unconstrained Optimization finds a minimum of a function under the assumption that the parameters can take on

More information

Demo 1: KKT conditions with inequality constraints

Demo 1: KKT conditions with inequality constraints MS-C5 Introduction to Optimization Solutions 9 Ehtamo Demo : KKT conditions with inequality constraints Using the Karush-Kuhn-Tucker conditions, see if the points x (x, x ) (, 4) or x (x, x ) (6, ) are

More information

Introduction to Constrained Optimization

Introduction to Constrained Optimization Introduction to Constrained Optimization Duality and KKT Conditions Pratik Shah {pratik.shah [at] lnmiit.ac.in} The LNM Institute of Information Technology www.lnmiit.ac.in February 13, 2013 LNMIIT MLPR

More information

Lecture 15: Log Barrier Method

Lecture 15: Log Barrier Method 10-725/36-725: Convex Optimization Spring 2015 Lecturer: Ryan Tibshirani Lecture 15: Log Barrier Method Scribes: Pradeep Dasigi, Mohammad Gowayyed Note: LaTeX template courtesy of UC Berkeley EECS dept.

More information

PRIMAL-DUAL INTERIOR POINT METHOD FOR LINEAR PROGRAMMING. 1. Introduction

PRIMAL-DUAL INTERIOR POINT METHOD FOR LINEAR PROGRAMMING. 1. Introduction PRIMAL-DUAL INTERIOR POINT METHOD FOR LINEAR PROGRAMMING KELLER VANDEBOGERT AND CHARLES LANNING 1. Introduction Interior point methods are, put simply, a technique of optimization where, given a problem

More information

Linear methods for supervised learning

Linear methods for supervised learning Linear methods for supervised learning LDA Logistic regression Naïve Bayes PLA Maximum margin hyperplanes Soft-margin hyperplanes Least squares resgression Ridge regression Nonlinear feature maps Sometimes

More information

A Truncated Newton Method in an Augmented Lagrangian Framework for Nonlinear Programming

A Truncated Newton Method in an Augmented Lagrangian Framework for Nonlinear Programming A Truncated Newton Method in an Augmented Lagrangian Framework for Nonlinear Programming Gianni Di Pillo (dipillo@dis.uniroma1.it) Giampaolo Liuzzi (liuzzi@iasi.cnr.it) Stefano Lucidi (lucidi@dis.uniroma1.it)

More information

A Short SVM (Support Vector Machine) Tutorial

A Short SVM (Support Vector Machine) Tutorial A Short SVM (Support Vector Machine) Tutorial j.p.lewis CGIT Lab / IMSC U. Southern California version 0.zz dec 004 This tutorial assumes you are familiar with linear algebra and equality-constrained optimization/lagrange

More information

Shiqian Ma, MAT-258A: Numerical Optimization 1. Chapter 2. Convex Optimization

Shiqian Ma, MAT-258A: Numerical Optimization 1. Chapter 2. Convex Optimization Shiqian Ma, MAT-258A: Numerical Optimization 1 Chapter 2 Convex Optimization Shiqian Ma, MAT-258A: Numerical Optimization 2 2.1. Convex Optimization General optimization problem: min f 0 (x) s.t., f i

More information

Kernel Methods & Support Vector Machines

Kernel Methods & Support Vector Machines & Support Vector Machines & Support Vector Machines Arvind Visvanathan CSCE 970 Pattern Recognition 1 & Support Vector Machines Question? Draw a single line to separate two classes? 2 & Support Vector

More information

Inverse KKT Motion Optimization: A Newton Method to Efficiently Extract Task Spaces and Cost Parameters from Demonstrations

Inverse KKT Motion Optimization: A Newton Method to Efficiently Extract Task Spaces and Cost Parameters from Demonstrations Inverse KKT Motion Optimization: A Newton Method to Efficiently Extract Task Spaces and Cost Parameters from Demonstrations Peter Englert Machine Learning and Robotics Lab Universität Stuttgart Germany

More information

Characterizing Improving Directions Unconstrained Optimization

Characterizing Improving Directions Unconstrained Optimization Final Review IE417 In the Beginning... In the beginning, Weierstrass's theorem said that a continuous function achieves a minimum on a compact set. Using this, we showed that for a convex set S and y not

More information

Convex Programs. COMPSCI 371D Machine Learning. COMPSCI 371D Machine Learning Convex Programs 1 / 21

Convex Programs. COMPSCI 371D Machine Learning. COMPSCI 371D Machine Learning Convex Programs 1 / 21 Convex Programs COMPSCI 371D Machine Learning COMPSCI 371D Machine Learning Convex Programs 1 / 21 Logistic Regression! Support Vector Machines Support Vector Machines (SVMs) and Convex Programs SVMs are

More information

CME307/MS&E311 Optimization Theory Summary

CME307/MS&E311 Optimization Theory Summary CME307/MS&E311 Optimization Theory Summary Yinyu Ye Department of Management Science and Engineering Stanford University Stanford, CA 94305, U.S.A. http://www.stanford.edu/~yyye http://www.stanford.edu/class/msande311/

More information

Lecture 4 Duality and Decomposition Techniques

Lecture 4 Duality and Decomposition Techniques Lecture 4 Duality and Decomposition Techniques Jie Lu (jielu@kth.se) Richard Combes Alexandre Proutiere Automatic Control, KTH September 19, 2013 Consider the primal problem Lagrange Duality Lagrangian

More information

Convex Optimization. Lijun Zhang Modification of

Convex Optimization. Lijun Zhang   Modification of Convex Optimization Lijun Zhang zlj@nju.edu.cn http://cs.nju.edu.cn/zlj Modification of http://stanford.edu/~boyd/cvxbook/bv_cvxslides.pdf Outline Introduction Convex Sets & Functions Convex Optimization

More information

Convex Optimization. Erick Delage, and Ashutosh Saxena. October 20, (a) (b) (c)

Convex Optimization. Erick Delage, and Ashutosh Saxena. October 20, (a) (b) (c) Convex Optimization (for CS229) Erick Delage, and Ashutosh Saxena October 20, 2006 1 Convex Sets Definition: A set G R n is convex if every pair of point (x, y) G, the segment beteen x and y is in A. More

More information

Contents. I Basics 1. Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

Contents. I Basics 1. Copyright by SIAM. Unauthorized reproduction of this article is prohibited. page v Preface xiii I Basics 1 1 Optimization Models 3 1.1 Introduction... 3 1.2 Optimization: An Informal Introduction... 4 1.3 Linear Equations... 7 1.4 Linear Optimization... 10 Exercises... 12 1.5

More information

California Institute of Technology Crash-Course on Convex Optimization Fall Ec 133 Guilherme Freitas

California Institute of Technology Crash-Course on Convex Optimization Fall Ec 133 Guilherme Freitas California Institute of Technology HSS Division Crash-Course on Convex Optimization Fall 2011-12 Ec 133 Guilherme Freitas In this text, we will study the following basic problem: maximize x C f(x) subject

More information

Lec 11 Rate-Distortion Optimization (RDO) in Video Coding-I

Lec 11 Rate-Distortion Optimization (RDO) in Video Coding-I CS/EE 5590 / ENG 401 Special Topics (17804, 17815, 17803) Lec 11 Rate-Distortion Optimization (RDO) in Video Coding-I Zhu Li Course Web: http://l.web.umkc.edu/lizhu/teaching/2016sp.video-communication/main.html

More information

Applied Lagrange Duality for Constrained Optimization

Applied Lagrange Duality for Constrained Optimization Applied Lagrange Duality for Constrained Optimization Robert M. Freund February 10, 2004 c 2004 Massachusetts Institute of Technology. 1 1 Overview The Practical Importance of Duality Review of Convexity

More information

Constrained optimization

Constrained optimization Constrained optimization A general constrained optimization problem has the form where The Lagrangian function is given by Primal and dual optimization problems Primal: Dual: Weak duality: Strong duality:

More information

In other words, we want to find the domain points that yield the maximum or minimum values (extrema) of the function.

In other words, we want to find the domain points that yield the maximum or minimum values (extrema) of the function. 1 The Lagrange multipliers is a mathematical method for performing constrained optimization of differentiable functions. Recall unconstrained optimization of differentiable functions, in which we want

More information

Optimal Control Techniques for Dynamic Walking

Optimal Control Techniques for Dynamic Walking Optimal Control Techniques for Dynamic Walking Optimization in Robotics & Biomechanics IWR, University of Heidelberg Presentation partly based on slides by Sebastian Sager, Moritz Diehl and Peter Riede

More information

DM6 Support Vector Machines

DM6 Support Vector Machines DM6 Support Vector Machines Outline Large margin linear classifier Linear separable Nonlinear separable Creating nonlinear classifiers: kernel trick Discussion on SVM Conclusion SVM: LARGE MARGIN LINEAR

More information

16.410/413 Principles of Autonomy and Decision Making

16.410/413 Principles of Autonomy and Decision Making 16.410/413 Principles of Autonomy and Decision Making Lecture 17: The Simplex Method Emilio Frazzoli Aeronautics and Astronautics Massachusetts Institute of Technology November 10, 2010 Frazzoli (MIT)

More information

Lecture 7: Support Vector Machine

Lecture 7: Support Vector Machine Lecture 7: Support Vector Machine Hien Van Nguyen University of Houston 9/28/2017 Separating hyperplane Red and green dots can be separated by a separating hyperplane Two classes are separable, i.e., each

More information

CME307/MS&E311 Theory Summary

CME307/MS&E311 Theory Summary CME307/MS&E311 Theory Summary Yinyu Ye Department of Management Science and Engineering Stanford University Stanford, CA 94305, U.S.A. http://www.stanford.edu/~yyye http://www.stanford.edu/class/msande311/

More information

Augmented Lagrangian Methods

Augmented Lagrangian Methods Augmented Lagrangian Methods Mário A. T. Figueiredo 1 and Stephen J. Wright 2 1 Instituto de Telecomunicações, Instituto Superior Técnico, Lisboa, Portugal 2 Computer Sciences Department, University of

More information

Comparison of Interior Point Filter Line Search Strategies for Constrained Optimization by Performance Profiles

Comparison of Interior Point Filter Line Search Strategies for Constrained Optimization by Performance Profiles INTERNATIONAL JOURNAL OF MATHEMATICS MODELS AND METHODS IN APPLIED SCIENCES Comparison of Interior Point Filter Line Search Strategies for Constrained Optimization by Performance Profiles M. Fernanda P.

More information

Lecture 18: March 23

Lecture 18: March 23 0-725/36-725: Convex Optimization Spring 205 Lecturer: Ryan Tibshirani Lecture 8: March 23 Scribes: James Duyck Note: LaTeX template courtesy of UC Berkeley EECS dept. Disclaimer: These notes have not

More information

Theoretical Concepts of Machine Learning

Theoretical Concepts of Machine Learning Theoretical Concepts of Machine Learning Part 2 Institute of Bioinformatics Johannes Kepler University, Linz, Austria Outline 1 Introduction 2 Generalization Error 3 Maximum Likelihood 4 Noise Models 5

More information

Gate Sizing by Lagrangian Relaxation Revisited

Gate Sizing by Lagrangian Relaxation Revisited Gate Sizing by Lagrangian Relaxation Revisited Jia Wang, Debasish Das, and Hai Zhou Electrical Engineering and Computer Science Northwestern University Evanston, Illinois, United States October 17, 2007

More information

4 Integer Linear Programming (ILP)

4 Integer Linear Programming (ILP) TDA6/DIT37 DISCRETE OPTIMIZATION 17 PERIOD 3 WEEK III 4 Integer Linear Programg (ILP) 14 An integer linear program, ILP for short, has the same form as a linear program (LP). The only difference is that

More information

Mathematical Programming and Research Methods (Part II)

Mathematical Programming and Research Methods (Part II) Mathematical Programming and Research Methods (Part II) 4. Convexity and Optimization Massimiliano Pontil (based on previous lecture by Andreas Argyriou) 1 Today s Plan Convex sets and functions Types

More information

Nonlinear Programming

Nonlinear Programming Nonlinear Programming SECOND EDITION Dimitri P. Bertsekas Massachusetts Institute of Technology WWW site for book Information and Orders http://world.std.com/~athenasc/index.html Athena Scientific, Belmont,

More information

Augmented Lagrangian Methods

Augmented Lagrangian Methods Augmented Lagrangian Methods Stephen J. Wright 1 2 Computer Sciences Department, University of Wisconsin-Madison. IMA, August 2016 Stephen Wright (UW-Madison) Augmented Lagrangian IMA, August 2016 1 /

More information

ISM206 Lecture, April 26, 2005 Optimization of Nonlinear Objectives, with Non-Linear Constraints

ISM206 Lecture, April 26, 2005 Optimization of Nonlinear Objectives, with Non-Linear Constraints ISM206 Lecture, April 26, 2005 Optimization of Nonlinear Objectives, with Non-Linear Constraints Instructor: Kevin Ross Scribe: Pritam Roy May 0, 2005 Outline of topics for the lecture We will discuss

More information

David G. Luenberger Yinyu Ye. Linear and Nonlinear. Programming. Fourth Edition. ö Springer

David G. Luenberger Yinyu Ye. Linear and Nonlinear. Programming. Fourth Edition. ö Springer David G. Luenberger Yinyu Ye Linear and Nonlinear Programming Fourth Edition ö Springer Contents 1 Introduction 1 1.1 Optimization 1 1.2 Types of Problems 2 1.3 Size of Problems 5 1.4 Iterative Algorithms

More information

Machine Learning for Signal Processing Lecture 4: Optimization

Machine Learning for Signal Processing Lecture 4: Optimization Machine Learning for Signal Processing Lecture 4: Optimization 13 Sep 2015 Instructor: Bhiksha Raj (slides largely by Najim Dehak, JHU) 11-755/18-797 1 Index 1. The problem of optimization 2. Direct optimization

More information

Natural Language Processing

Natural Language Processing Natural Language Processing Classification III Dan Klein UC Berkeley 1 Classification 2 Linear Models: Perceptron The perceptron algorithm Iteratively processes the training set, reacting to training errors

More information

Kernels and Constrained Optimization

Kernels and Constrained Optimization Machine Learning 1 WS2014 Module IN2064 Sheet 8 Page 1 Machine Learning Worksheet 8 Kernels and Constrained Optimization 1 Kernelized k-nearest neighbours To classify the point x the k-nearest neighbours

More information

Lecture 10: SVM Lecture Overview Support Vector Machines The binary classification problem

Lecture 10: SVM Lecture Overview Support Vector Machines The binary classification problem Computational Learning Theory Fall Semester, 2012/13 Lecture 10: SVM Lecturer: Yishay Mansour Scribe: Gitit Kehat, Yogev Vaknin and Ezra Levin 1 10.1 Lecture Overview In this lecture we present in detail

More information

Solution Methods Numerical Algorithms

Solution Methods Numerical Algorithms Solution Methods Numerical Algorithms Evelien van der Hurk DTU Managment Engineering Class Exercises From Last Time 2 DTU Management Engineering 42111: Static and Dynamic Optimization (6) 09/10/2017 Class

More information

Unconstrained Optimization Principles of Unconstrained Optimization Search Methods

Unconstrained Optimization Principles of Unconstrained Optimization Search Methods 1 Nonlinear Programming Types of Nonlinear Programs (NLP) Convexity and Convex Programs NLP Solutions Unconstrained Optimization Principles of Unconstrained Optimization Search Methods Constrained Optimization

More information

Lecture 2 September 3

Lecture 2 September 3 EE 381V: Large Scale Optimization Fall 2012 Lecture 2 September 3 Lecturer: Caramanis & Sanghavi Scribe: Hongbo Si, Qiaoyang Ye 2.1 Overview of the last Lecture The focus of the last lecture was to give

More information

Introduction to Machine Learning

Introduction to Machine Learning Introduction to Machine Learning Maximum Margin Methods Varun Chandola Computer Science & Engineering State University of New York at Buffalo Buffalo, NY, USA chandola@buffalo.edu Chandola@UB CSE 474/574

More information

Intro to Linear Programming. The problem that we desire to address in this course is loosely stated below.

Intro to Linear Programming. The problem that we desire to address in this course is loosely stated below. . Introduction Intro to Linear Programming The problem that we desire to address in this course is loosely stated below. Given a number of generators make price-quantity offers to sell (each provides their

More information

A primal-dual framework for mixtures of regularizers

A primal-dual framework for mixtures of regularizers A primal-dual framework for mixtures of regularizers Baran Gözcü baran.goezcue@epfl.ch Laboratory for Information and Inference Systems (LIONS) École Polytechnique Fédérale de Lausanne (EPFL) Switzerland

More information

9. Support Vector Machines. The linearly separable case: hard-margin SVMs. The linearly separable case: hard-margin SVMs. Learning objectives

9. Support Vector Machines. The linearly separable case: hard-margin SVMs. The linearly separable case: hard-margin SVMs. Learning objectives Foundations of Machine Learning École Centrale Paris Fall 25 9. Support Vector Machines Chloé-Agathe Azencot Centre for Computational Biology, Mines ParisTech Learning objectives chloe agathe.azencott@mines

More information

Programs. Introduction

Programs. Introduction 16 Interior Point I: Linear Programs Lab Objective: For decades after its invention, the Simplex algorithm was the only competitive method for linear programming. The past 30 years, however, have seen

More information

Convex Optimization MLSS 2015

Convex Optimization MLSS 2015 Convex Optimization MLSS 2015 Constantine Caramanis The University of Texas at Austin The Optimization Problem minimize : f (x) subject to : x X. The Optimization Problem minimize : f (x) subject to :

More information

Lecture 12: convergence. Derivative (one variable)

Lecture 12: convergence. Derivative (one variable) Lecture 12: convergence More about multivariable calculus Descent methods Backtracking line search More about convexity (first and second order) Newton step Example 1: linear programming (one var., one

More information

Chapter II. Linear Programming

Chapter II. Linear Programming 1 Chapter II Linear Programming 1. Introduction 2. Simplex Method 3. Duality Theory 4. Optimality Conditions 5. Applications (QP & SLP) 6. Sensitivity Analysis 7. Interior Point Methods 1 INTRODUCTION

More information

Support Vector Machines.

Support Vector Machines. Support Vector Machines srihari@buffalo.edu SVM Discussion Overview 1. Overview of SVMs 2. Margin Geometry 3. SVM Optimization 4. Overlapping Distributions 5. Relationship to Logistic Regression 6. Dealing

More information

Linear Programming has been used to:

Linear Programming has been used to: Linear Programming Linear programming became important during World War II: used to solve logistics problems for the military. Linear Programming (LP) was the first widely used form of optimization in

More information

Local and Global Minimum

Local and Global Minimum Local and Global Minimum Stationary Point. From elementary calculus, a single variable function has a stationary point at if the derivative vanishes at, i.e., 0. Graphically, the slope of the function

More information

Introduction to optimization methods and line search

Introduction to optimization methods and line search Introduction to optimization methods and line search Jussi Hakanen Post-doctoral researcher jussi.hakanen@jyu.fi How to find optimal solutions? Trial and error widely used in practice, not efficient and

More information

Advanced Operations Research Techniques IE316. Quiz 2 Review. Dr. Ted Ralphs

Advanced Operations Research Techniques IE316. Quiz 2 Review. Dr. Ted Ralphs Advanced Operations Research Techniques IE316 Quiz 2 Review Dr. Ted Ralphs IE316 Quiz 2 Review 1 Reading for The Quiz Material covered in detail in lecture Bertsimas 4.1-4.5, 4.8, 5.1-5.5, 6.1-6.3 Material

More information

5 Machine Learning Abstractions and Numerical Optimization

5 Machine Learning Abstractions and Numerical Optimization Machine Learning Abstractions and Numerical Optimization 25 5 Machine Learning Abstractions and Numerical Optimization ML ABSTRACTIONS [some meta comments on machine learning] [When you write a large computer

More information

Optimization under uncertainty: modeling and solution methods

Optimization under uncertainty: modeling and solution methods Optimization under uncertainty: modeling and solution methods Paolo Brandimarte Dipartimento di Scienze Matematiche Politecnico di Torino e-mail: paolo.brandimarte@polito.it URL: http://staff.polito.it/paolo.brandimarte

More information

New Directions in Linear Programming

New Directions in Linear Programming New Directions in Linear Programming Robert Vanderbei November 5, 2001 INFORMS Miami Beach NOTE: This is a talk mostly on pedagogy. There will be some new results. It is not a talk on state-of-the-art

More information

Lagrangian Relaxation: An overview

Lagrangian Relaxation: An overview Discrete Math for Bioinformatics WS 11/12:, by A. Bockmayr/K. Reinert, 22. Januar 2013, 13:27 4001 Lagrangian Relaxation: An overview Sources for this lecture: D. Bertsimas and J. Tsitsiklis: Introduction

More information

arxiv: v2 [math.oc] 12 Feb 2019

arxiv: v2 [math.oc] 12 Feb 2019 A CENTRAL PATH FOLLOWING METHOD USING THE NORMALIZED GRADIENTS Long Chen 1, Wenyi Chen 2, and Kai-Uwe Bletzinger 1 arxiv:1902.04040v2 [math.oc] 12 Feb 2019 1 Chair of Structural Analysis, Technical University

More information

Coloring 3-Colorable Graphs

Coloring 3-Colorable Graphs Coloring -Colorable Graphs Charles Jin April, 015 1 Introduction Graph coloring in general is an etremely easy-to-understand yet powerful tool. It has wide-ranging applications from register allocation

More information

Lecture 19: Convex Non-Smooth Optimization. April 2, 2007

Lecture 19: Convex Non-Smooth Optimization. April 2, 2007 : Convex Non-Smooth Optimization April 2, 2007 Outline Lecture 19 Convex non-smooth problems Examples Subgradients and subdifferentials Subgradient properties Operations with subgradients and subdifferentials

More information

LECTURE 6: INTERIOR POINT METHOD. 1. Motivation 2. Basic concepts 3. Primal affine scaling algorithm 4. Dual affine scaling algorithm

LECTURE 6: INTERIOR POINT METHOD. 1. Motivation 2. Basic concepts 3. Primal affine scaling algorithm 4. Dual affine scaling algorithm LECTURE 6: INTERIOR POINT METHOD 1. Motivation 2. Basic concepts 3. Primal affine scaling algorithm 4. Dual affine scaling algorithm Motivation Simplex method works well in general, but suffers from exponential-time

More information

An augmented Lagrangian method for equality constrained optimization with fast infeasibility detection

An augmented Lagrangian method for equality constrained optimization with fast infeasibility detection An augmented Lagrangian method for equality constrained optimization with fast infeasibility detection Paul Armand 1 Ngoc Nguyen Tran 2 Institut de Recherche XLIM Université de Limoges Journées annuelles

More information

The Alternating Direction Method of Multipliers

The Alternating Direction Method of Multipliers The Alternating Direction Method of Multipliers With Adaptive Step Size Selection Peter Sutor, Jr. Project Advisor: Professor Tom Goldstein October 8, 2015 1 / 30 Introduction Presentation Outline 1 Convex

More information

Chapter 3 Numerical Methods

Chapter 3 Numerical Methods Chapter 3 Numerical Methods Part 1 3.1 Linearization and Optimization of Functions of Vectors 1 Problem Notation 2 Outline 3.1.1 Linearization 3.1.2 Optimization of Objective Functions 3.1.3 Constrained

More information

Math 5593 Linear Programming Lecture Notes

Math 5593 Linear Programming Lecture Notes Math 5593 Linear Programming Lecture Notes Unit II: Theory & Foundations (Convex Analysis) University of Colorado Denver, Fall 2013 Topics 1 Convex Sets 1 1.1 Basic Properties (Luenberger-Ye Appendix B.1).........................

More information

Interior Point I. Lab 21. Introduction

Interior Point I. Lab 21. Introduction Lab 21 Interior Point I Lab Objective: For decades after its invention, the Simplex algorithm was the only competitive method for linear programming. The past 30 years, however, have seen the discovery

More information

3 Interior Point Method

3 Interior Point Method 3 Interior Point Method Linear programming (LP) is one of the most useful mathematical techniques. Recent advances in computer technology and algorithms have improved computational speed by several orders

More information

Chapter 3 Path Optimization

Chapter 3 Path Optimization Chapter 3 Path Optimization Background information on optimization is discussed in this chapter, along with the inequality constraints that are used for the problem. Additionally, the MATLAB program for

More information

Epigraph proximal algorithms for general convex programming

Epigraph proximal algorithms for general convex programming Epigraph proimal algorithms for general conve programming Matt Wytock, Po-Wei Wang and J. Zico Kolter Machine Learning Department Carnegie Mellon University mwytock@cs.cmu.edu Abstract This work aims at

More information

Paul's Online Math Notes Calculus III (Notes) / Applications of Partial Derivatives / Lagrange Multipliers Problems][Assignment Problems]

Paul's Online Math Notes Calculus III (Notes) / Applications of Partial Derivatives / Lagrange Multipliers Problems][Assignment Problems] 1 of 9 25/04/2016 13:15 Paul's Online Math Notes Calculus III (Notes) / Applications of Partial Derivatives / Lagrange Multipliers Problems][Assignment Problems] [Notes] [Practice Calculus III - Notes

More information

Linear Programming Problems

Linear Programming Problems Linear Programming Problems Two common formulations of linear programming (LP) problems are: min Subject to: 1,,, 1,2,,;, max Subject to: 1,,, 1,2,,;, Linear Programming Problems The standard LP problem

More information

Integer Programming Theory

Integer Programming Theory Integer Programming Theory Laura Galli October 24, 2016 In the following we assume all functions are linear, hence we often drop the term linear. In discrete optimization, we seek to find a solution x

More information

Performance Evaluation of an Interior Point Filter Line Search Method for Constrained Optimization

Performance Evaluation of an Interior Point Filter Line Search Method for Constrained Optimization 6th WSEAS International Conference on SYSTEM SCIENCE and SIMULATION in ENGINEERING, Venice, Italy, November 21-23, 2007 18 Performance Evaluation of an Interior Point Filter Line Search Method for Constrained

More information

Constrained Optimization COS 323

Constrained Optimization COS 323 Constrained Optimization COS 323 Last time Introduction to optimization objective function, variables, [constraints] 1-dimensional methods Golden section, discussion of error Newton s method Multi-dimensional

More information

Mathematical and Algorithmic Foundations Linear Programming and Matchings

Mathematical and Algorithmic Foundations Linear Programming and Matchings Adavnced Algorithms Lectures Mathematical and Algorithmic Foundations Linear Programming and Matchings Paul G. Spirakis Department of Computer Science University of Patras and Liverpool Paul G. Spirakis

More information

Search direction improvement for gradient-based optimization problems

Search direction improvement for gradient-based optimization problems Computer Aided Optimum Design in Engineering IX 3 Search direction improvement for gradient-based optimization problems S Ganguly & W L Neu Aerospace and Ocean Engineering, Virginia Tech, USA Abstract

More information

CONLIN & MMA solvers. Pierre DUYSINX LTAS Automotive Engineering Academic year

CONLIN & MMA solvers. Pierre DUYSINX LTAS Automotive Engineering Academic year CONLIN & MMA solvers Pierre DUYSINX LTAS Automotive Engineering Academic year 2018-2019 1 CONLIN METHOD 2 LAY-OUT CONLIN SUBPROBLEMS DUAL METHOD APPROACH FOR CONLIN SUBPROBLEMS SEQUENTIAL QUADRATIC PROGRAMMING

More information

Conic Optimization via Operator Splitting and Homogeneous Self-Dual Embedding

Conic Optimization via Operator Splitting and Homogeneous Self-Dual Embedding Conic Optimization via Operator Splitting and Homogeneous Self-Dual Embedding B. O Donoghue E. Chu N. Parikh S. Boyd Convex Optimization and Beyond, Edinburgh, 11/6/2104 1 Outline Cone programming Homogeneous

More information

LECTURE NOTES Non-Linear Programming

LECTURE NOTES Non-Linear Programming CEE 6110 David Rosenberg p. 1 Learning Objectives LECTURE NOTES Non-Linear Programming 1. Write out the non-linear model formulation 2. Describe the difficulties of solving a non-linear programming model

More information

Outline. CS38 Introduction to Algorithms. Linear programming 5/21/2014. Linear programming. Lecture 15 May 20, 2014

Outline. CS38 Introduction to Algorithms. Linear programming 5/21/2014. Linear programming. Lecture 15 May 20, 2014 5/2/24 Outline CS38 Introduction to Algorithms Lecture 5 May 2, 24 Linear programming simplex algorithm LP duality ellipsoid algorithm * slides from Kevin Wayne May 2, 24 CS38 Lecture 5 May 2, 24 CS38

More information

Limits. f(x) and lim. g(x) g(x)

Limits. f(x) and lim. g(x) g(x) Limits Limit Laws Suppose c is constant, n is a positive integer, and f() and g() both eist. Then,. [f() + g()] = f() + g() 2. [f() g()] = f() g() [ ] 3. [c f()] = c f() [ ] [ ] 4. [f() g()] = f() g()

More information

Constrained Optimization with Calculus. Background Three Big Problems Setup and Vocabulary

Constrained Optimization with Calculus. Background Three Big Problems Setup and Vocabulary Constrained Optimization with Calculus Background Three Big Problems Setup and Vocabulary Background Information In unit 3, you learned about linear programming, in which all constraints and the objective

More information

Convex Functions & Optimization

Convex Functions & Optimization 672 Conve Functions & Optimization Aashray Yadav Abstract - My research paper is based on the recent work in interior-point methods, specifically those methods that keep track of both the primal and dual

More information

Kernel Methods. Chapter 9 of A Course in Machine Learning by Hal Daumé III. Conversion to beamer by Fabrizio Riguzzi

Kernel Methods. Chapter 9 of A Course in Machine Learning by Hal Daumé III.   Conversion to beamer by Fabrizio Riguzzi Kernel Methods Chapter 9 of A Course in Machine Learning by Hal Daumé III http://ciml.info Conversion to beamer by Fabrizio Riguzzi Kernel Methods 1 / 66 Kernel Methods Linear models are great because

More information