Forward and Backward Deviation Measures and Robust Optimization Peng Sun (Duke) with Xin Chen (UIUC) and Melvyn Sim (NUS)

Slides:



Advertisements
Similar presentations
Quantum t-designs: t-wise independence in the quantum world Andris Ambainis, Joseph Emerson IQC, University of Waterloo.
Advertisements

On allocations that maximize fairness Uriel Feige Microsoft Research and Weizmann Institute.
Beyond Convexity – Submodularity in Machine Learning
The Normal Distribution
Solving IPs – Cutting Plane Algorithm General Idea: Begin by solving the LP relaxation of the IP problem. If the LP relaxation results in an integer solution,
Introduction to Mathematical Programming Matthew J. Liberatore John F. Connelly Chair in Management Professor, Decision and Information Technologies.
Lecture #3; Based on slides by Yinyu Ye
Optimization of thermal processes2007/2008 Optimization of thermal processes Maciej Marek Czestochowa University of Technology Institute of Thermal Machinery.
Convex Position Estimation in Wireless Sensor Networks
by Rianto Adhy Sasongko Supervisor: Dr.J.C.Allwright
Robust Allocation of a Defensive Budget Considering an Attacker’s Private Information Mohammad E. Nikoofal and Jun Zhuang Presenter: Yi-Cin Lin Advisor:
Second order cone programming approaches for handing missing and uncertain data P. K. Shivaswamy, C. Bhattacharyya and A. J. Smola Discussion led by Qi.
INFORMS Annual Meeting, San Diego, Oct 11 – 14, 2009 A New, Intuitive and Industrial Scale Approach towards Uncertainty in Supply Chains G.N. Srinivasa.
Parameterized Timing Analysis with General Delay Models and Arbitrary Variation Sources Khaled R. Heloue and Farid N. Najm University of Toronto {khaled,
IAAC International Symposium in Systems & Control, 7-8 October 2013, Technion, Haifa, Israel P-O Gutman: Constrained control of uncertain linear time-invariant.
Robust Multi-Kernel Classification of Uncertain and Imbalanced Data
A Randomized Polynomial-Time Simplex Algorithm for Linear Programming Daniel A. Spielman, Yale Joint work with Jonathan Kelner, M.I.T.
1 2 Extreme Pathway Lengths and Reaction Participation in Genome Scale Metabolic Networks Jason A. Papin, Nathan D. Price and Bernhard Ø. Palsson.
Chapter 8 Estimating Single Population Parameters
Feasibility, uncertainty and interpolation J. A. Rossiter (Sheffield, UK)
Sampling Distributions
Linear Programming Special Cases Alternate Optimal Solutions No Feasible Solution Unbounded Solutions.
1 Absolute Stability with a Generalized Sector Condition Tingshu Hu.
Jan 6-10th, 2007VLSI Design A Reduced Complexity Algorithm for Minimizing N-Detect Tests Kalyana R. Kantipudi Vishwani D. Agrawal Department of Electrical.
Planning operation start times for the manufacture of capital products with uncertain processing times and resource constraints D.P. Song, Dr. C.Hicks.
Session 6: Introduction to cryptanalysis part 1. Contents Problem definition Symmetric systems cryptanalysis Particularities of block ciphers cryptanalysis.
Polyhedral Optimization Lecture 3 – Part 3 M. Pawan Kumar Slides available online
1 Robustness by cutting planes and the Uncertain Set Covering Problem AIRO 2008, Ischia, 10 September 2008 (work supported my MiUR and EU) Matteo Fischetti.
1 10. Joint Moments and Joint Characteristic Functions Following section 6, in this section we shall introduce various parameters to compactly represent.
1 Lecture 4 Maximal Flow Problems Set Covering Problems.
1 Abhilasha Aswal & G N S Prasanna IIIT-B INFORMS 2010 Inventory Optimization under Correlated Uncertainty Abhilasha Aswal G N S Prasanna, International.
Approximation Algorithms for Stochastic Combinatorial Optimization Part I: Multistage problems Anupam Gupta Carnegie Mellon University.
Alternative Measures of Risk. The Optimal Risk Measure Desirable Properties for Risk Measure A risk measure maps the whole distribution of one dollar.
18th Inter-Institute Seminar, September 2011, Budapest, Hungary 1 J. Lógó, D. B. Merczel and L. Nagy Department of Structural Mechanics Budapest.
Adaptive CSMA under the SINR Model: Fast convergence using the Bethe Approximation Krishna Jagannathan IIT Madras (Joint work with) Peruru Subrahmanya.
Computer Vision - A Modern Approach Set: Tracking Slides by D.A. Forsyth The three main issues in tracking.
Module 1: Statistical Issues in Micro simulation Paul Sousa.
Probabilistic Reasoning for Robust Plan Execution Steve Schaffer, Brad Clement, Steve Chien Artificial Intelligence.
Journey to New Areas of Optimization Trends in Modern Optimization Feb., 2007 Kyungchul Park.
Copyright © 2013, 2010 and 2007 Pearson Education, Inc. Section Inference about Two Means: Independent Samples 11.3.
I.4 Polyhedral Theory 1. Integer Programming  Objective of Study: want to know how to describe the convex hull of the solution set to the IP problem.
Robust Optimization and Applications Laurent El Ghaoui IMA Tutorial, March 11, 2003.
Quality of LP-based Approximations for Highly Combinatorial Problems Lucian Leahu and Carla Gomes Computer Science Department Cornell University.
1 Enclosing Ellipsoids of Semi-algebraic Sets and Error Bounds in Polynomial Optimization Makoto Yamashita Masakazu Kojima Tokyo Institute of Technology.
Point Estimation of Parameters and Sampling Distributions Outlines:  Sampling Distributions and the central limit theorem  Point estimation  Methods.
2.1 – Linear and Quadratic Equations Linear Equations.
Deterministic Algorithms for Submodular Maximization Problems Moran Feldman The Open University of Israel Joint work with Niv Buchbinder.
Joint Moments and Joint Characteristic Functions.
Optimization formulation Optimization methods help us find solutions to problems where we seek to find the best of something. This lecture is about how.
University of Toronto Power Grid Verification in the Presence of Uncertainty Farid N. Najm University of Toronto Imad A. Ferzli University.
Linear Programming Chap 2. The Geometry of LP  In the text, polyhedron is defined as P = { x  R n : Ax  b }. So some of our earlier results should.
23rd European Conference on Operational Research, Bonn, 2009 New framework for supply chain decision making under uncertainty: An analysis of the computational.
Approximation Algorithms based on linear programming.
Behavioral Finance Preferences Part I Feb 16 Behavioral Finance Economics 437.
Morphing and Shape Processing
6.5 Stochastic Prog. and Benders’ decomposition
Symmetric Allocations for Distributed Storage
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Chapter 6. Large Scale Optimization
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Distributed Stochastic Model Predictive Control
Max Z = x1 + x2 2 x1 + 3 x2  6 (1) x2  1.5 (2) x1 - x2  2 (3)
Robust Optimization and Applications in Machine Learning
I.4 Polyhedral Theory (NW)
On Approximating Covering Integer Programs
Lecture 20 Linear Program Duality
I.4 Polyhedral Theory.
6.5 Stochastic Prog. and Benders’ decomposition
Branch-and-Bound Algorithm for Integer Program
1.2 Guidelines for strong formulations
Presentation transcript:

Forward and Backward Deviation Measures and Robust Optimization Peng Sun (Duke) with Xin Chen (UIUC) and Melvyn Sim (NUS)

Agenda Overview of Robust Optimization Forward and Backward Deviation Measures Uncertainty Sets and Probability Bounds Conclusions

Uncertain linear constraints Uncertain linear constraint Chance constraint  Hard to solve when  is small

Linear constraints with uncertainty Affine Uncertainty : zero mean, independent but not necessarily identically distributed

Overview of Robust Optimization Worst case  Relies only on the distribution support  Easy to solve (Soyster 1973)  Extremely conservative

Overview of Robust Optimization Goal of Robust Optimization:  Easy to obtain feasible solutions that satisfy chance constraint  Not as conservative as worst case

Overview of Robust Optimization Design of Uncertainty Set  Subset of worst case set, W   : Budget of uncertainty Control radius of uncertainty set, S    max : Worse case budget For reasonable probability bound,  <<  max

Overview of Robust Optimization Robust Counterpart  Semi-infinite constraint Tractability Polyhedral Uncertainty Set  Linear Programming (LP) Conic Quadratic Uncertainty Set  Second Order Cone Programming (SOCP)

Overview of Robust Optimization Ellipsoidal Uncertainty Set  Ben-Tal and Nemirovski (1997), El-Ghaoui et al. (1996)  Robust Counterpart is a SOCP  Probability bound for symmetrically bounded uncertainties:

Overview of Robust Optimization Polyhedral Uncertainty Set  Bertsimas and Sim (2000)  Robust Counterpart is a LP  Probability bound for symmetrically bounded uncertainties:  More conservative compared to Ellipsoidal Uncertainty Set

Overview of Robust Optimization Norm Uncertainty Set  Robust Counterpart depends on the choice of norm  Probability bound for symmetrically bounded uncertainties:

Overview of Robust Optimization Modeling limitations of Classical Uncertainty Sets  Uncertainty set is symmetrical, but distributions are generally asymmetrical  Can we use more information on distributions to obtain less conservative solution in achieveing the same probability bound? --- Forward and Backward Deviation Measures

Forward and Backward Deviation Measures Set of forward deviation measures Set of backward deviation measures

Alternatively, Forward deviation Backward deviation

Forward and Backward Deviation Measures Key idea

Important property Subadditivity

Least conservative forward and backward deviation measures Suppose distribution is known p * = q * if distribution is symmetrical p *, q * ¸  (standard deviation) p * = q * =  if distribution is Normal

Least conservative forward and backward deviation measures p *, q * can be obtained numerically  E.g: p * = q * = 0.58 for uniform distribution in [-1,1]  E.g:

Forward and backward deviation measures – two point distribution example

Distribution not known Suppose distribution is bounded in [-1,1] (but not necessarily symmetrical)

Only know mean and support

Forward and Backward Deviation Measures: Function g(  )

General Deviation Measures

General Deviation Measure

Agenda Overview of Robust Optimization Forward and Backward Deviation Measures Uncertainty Sets and Probability Bounds Conclusions

Uncertainty Sets and Probability Bounds Model of Data Uncertainty  Recall: Affine Uncertainty : zero mean, independent but not necessarily identically distributed

Uncertainty Sets and Probability Bounds Model of Data Uncertainty

Uncertainty Sets and Probability Bounds New Uncertainty Set (Asymmetrical)

Uncertainty Sets and Probability Bounds Recall: Norm Uncertainty Set (Symmetrical)

Uncertainty Sets and Probability Bounds Generalizes Symmetrical Uncertainty Set

Uncertainty Sets and Probability Bounds

Robust Counterpart

Uncertainty Sets and Probability Bounds Probability Bound More information achieve less conservative solution while preserving the bound. E.g  p j = q j = 1 for symmetric distribution in [-1,1]  p j = q j = 0.58 for uniform distribution in [-1,1]

Conclusions RO framework  Affine uncertainty constraints  Independent r.v.’s with asymmetric distributions Forward and backward Deviation measures  Defined from moment generating functions – implying probability bound  Sub-additivity – linear combinations  Easy to calculate and approximate from support information  Advantage -- less conservative solution Next  Stochastic programming with chance constraints