Solving Very Large Financial Planning Problems on Blue Gene
|
|
- Angelina Lester
- 8 years ago
- Views:
Transcription
1 U N I V E R S School of Mathematics T H E O I T Y H F G E D I N U R Solving Very Large Financial Planning Problems on lue Gene ndreas Grothey, University of Edinburgh joint work with Jacek Gondzio, Marco Colombo urora 2007, Vienna 1
2 Overview sset and Liability Management Mean-Variance Formulation/Stochastic Programming Structure of Problem OOPS (Interior Point Solver) Interior Point Methods Structure Exploitation through Object Oriented Design Challenges in Massively Parallel Implementation Communications Data Management Granularity of Computations Further Developments urora 2007, Vienna 2
3 Portfolio Optimization: sset and Liability Management set of assets J = {1,..., J} is given (e.g. bonds, stock, real estate). t every stage t = 0,..., T 1 we can buy or sell different assets. The return of asset j at stage t is uncertain (but distribution is known). We have to make investment decisions: what to buy or sell, at which time stage Objectives: maximize the final wealth minimize the associated risk Extensions (leading to nonlinear models) } Mean Variance formulation: max IE(X) ρvar(x) different risk measures (one-sided, expected shortfall/value at risk) (nonlinear) utility functions urora 2007, Vienna 3
4 Modelling: Multiperiod Mean-Variance Model Variables: x h j,t position in asset j at time t. x s j,t,xb j,t amount of asset j bought/sold at time t. Parameters: v j value of asset j r j,t return of asset j when held at time t L t, C t liabilities/cash contributions at time t Objective: max IE(X) ρvar(x), X = (1 c t ) j v j x h j,t Constraints: C t + (1 c t ) j x h j,t = (1 + r t 1,j )x h j,t 1 x s j,t 1 + x b j,t 1 v j x s j,t = L t + (1+c t ) j v j x b j,t (inventory) (cash balance) urora 2007, Vienna 4
5 Modelling: Multiperiod Mean-Variance Model Variables: x h j,t position in asset j at time t. x s j,t,xb j,t amount of asset j bought/sold at time t. Parameters: v j value of asset j r j,t return of asset j when held at time t L t, C t liabilities/cash contributions at time t Objective: max IE(X) ρvar(x), X = (1 c t ) j v j x h j,t Constraints: C t + (1 c t ) j x h j,t = (1 + r t 1,j )x h j,t 1 x s j,t 1 + x b j,t 1 v j x s j,t = L t + (1+c t ) j v j x b j,t (inventory) (cash balance) urora 2007, Vienna 4
6 Stochastic Programming: use event tree (t-1,a(n)) ξ 1 ξ 2 (t,n-1) (t,n) Stages represent points in time at which decisions are taken. Nodes represent possible futures (given by asset returns and probabilities): r (t,n) j return of asset j at node (t, n). p (t,n) probability of reaching node (t, n). L (t,n), C (t,n) liability payment/cash-contribution at node (t, n). urora 2007, Vienna 5
7 sset and Liability Management as Stochastic Program With every node i = (t, n) we associate variables: x h j,i the position in asset j at node i; x b j,i, x s j,i the amount of asset j bought/sold at node i. and constraints: (1 + r i,j )x h π(i),j = xh i,j x b i,j + x s i,j, j (inventory) (1 + c t )v j x b i,j + L i = (1 c t )v j x s i,j + C i (cash balance) j j Objective is sum over final stage scenarios min IE(X) ρvar(x) = IE(X) ρie[(x E(X)) 2 ] = (1 c t ) p i v j x h i,j ρ p i [ ] i L T j i LT urora 2007, Vienna 6
8 Multistage Stochastic Programming Period 1 Period 2 Period 3 Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario Tree Constraint Matrix Symmetrical event tree with K realizations at each node and T periods corresponds to K T 1 K T 1 scenarios nodes (blocks) K 1 urora 2007, Vienna 7
9 Structure of Objective IE(X) ρvar(x) = IE(X) ρ[ie(x 2 ) IE(X) 2 = [ p i v j x h ij ρ p i (v j x h ij )2 [ ] p i v j x h ij ]2 i L T j i L T j i L T j Dense, convex Hessian urora 2007, Vienna 8
10 Structure of Objective IE(X) ρvar(x) = IE(X) ρ[ie(x 2 ) IE(X) 2 = [ p i v j x h ij ρ p i (v j x h ij )2 [ ] p i v j x h ij ]2 i L T j i L T j i L T j lternatively: with y = p i v j x h ij i L T j IE(X) ρvar(x) = y ρ p i (v j x h ij) 2 y 2 i LT j Dense, convex Hessian Sparse, nonconvex Hessian 1 urora 2007, Vienna 8
11 LM: Structure of matrices and : Matrix Matrix 1 d d d d 1 urora 2007, Vienna 9
12 Stochastic Programming approach to Financial Planning Popular with academics, practitioners seem suspicious Large problem sizes Places constraints on model (?) Traditionally solved by decomposition Nested enders Decomposition aka L-shaped method Works well for LP models Unclear how to extend to quadratic/nonlinear models Or solved by Interior Point Methods Use taylored linear algebra Only certain types of constraints are allowable (?) Structure Exploitation is key to succesful implementation urora 2007, Vienna 10
13 OOPS - Object Oriented Parallel (Interior Point) Solver Mantra: Truly large scale problems are not only sparse but structured (due to e.g. dynamics, uncertainty, spatial distribution etc.) Exploiting structure is key to building efficient IPMs for large problems: Faster linear algebra Reduced memory use (by use of implicit factorization) Possibility to exploit (massive) parallelism We assume that structure is known! no automatic detection. OOPS is a general purpose (parallel) Interior Point solver Not tuned to any particular hardware Not tuned to any particular problem (structure). OOPS currently solves LP/P problems. NLP extension solves nonlinear financial planning problems urora 2007, Vienna 11
14 Interior Point Methods min c x x x s.t. x = b x 0 Optimality conditions: c + x y z = 0 x = b XZe = 0(µe) x, z 0 (P) Newton Step: I 0 0 Z 0 X x y z = ξ c ξ b r xz = c + x y z b x µe XZe (NS-P) Reduced to [ Θ 0 ] [ x y ] = [ ξc X 1 r xz ξ b ] where Θ = X 1 Z, X = diag(x), Z = diag(z) urora 2007, Vienna 12
15 Linear lgebra of IPMs Main work: solve [ Θ ] } {{ 0 } Φ [ x y ] = [ r h ] for several right-hand-sides at each iteration Two stage solution procedure factorize Φ = LDL backsolve(s) to compute direction ( x, y) + corrections Φ changes numerically but not structurally at each iteration Key to efficient implementation is exploiting structure of Φ in these two steps urora 2007, Vienna 13
16 Example: ordered lock-diagonal Structure (Schur complement) Φ Φ n n } 1 n {{ Φ 0 } Φ L 1... = L n } L 1,0 L n,0 {{ L 0 } L D 1... } {{ D 0 } D Cholesky-like factors can be obtained by Schur-complement: L i,0 = i L i Φ i = L i D i L i C = Φ 0 n i=1 L i,0d i L i,0 C = L 0 D 0 L 0 nd the system Φx = b can be solved by L 1 L 1,0.... L n L n,0 L 0 } {{ } L Di 1, i = 1,... n z i = L 1 i b i z 0 = L 1 0 (b 0 L i,0 z i ) y i = Di 1 z i x 0 = L 0 y 0 x i = L i (y i L i,0x 0 ) Operations (Cholesky, Solve, Product) are only performed on sub-blocks Can also exploit structure in sub-blocks urora 2007, Vienna 14
17 Tree Representation of Problem Structure Primal lock ngular Structure Dual lock ngular Structure Primal lock ngular Structure D 1 D 2 D 11 D 12 D D D D D C 31 C 32 D 30 Organisation of Linear lgebra, Parallelism urora 2007, Vienna 15
18 OOPS: Object-oriented linear algebra implementation Every node in block elimination tree has own linear algebra implementation (depending on its type) Implementation is realisation of an abstract linear algebra interface. Different implementations for different structures are available. Matrix Interface matrix factorize SolveL SolveLt y=mx y=mtx orderedlockdiagonal D i Implementations based on Schur complement i SparseMatrix general sparse linear algebra RankCorrector Rank corrector implementations D R Rebuild block elimination tree with matrix interface structures urora 2007, Vienna 16
19 Sources of Structure I (linear) dynamics: x t+1 = x t + u t I I uncertainty: common 1st stage decision (today) recourse action in 2nd stage (tomorrow) T x + W i y i = h i T T T W W W almost independent divisions share a common resource k i x i = b i=1 nd many others.... lso nested structures. urora 2007, Vienna 17
20 LM: Structure of matrices and : Matrix Matrix 1 d d d d 1 urora 2007, Vienna 18
21 Structures of and imply structure of Φ : d d d d 1 1 T T ( 0 T T T T ) T T T T T T T d T d T d T d T 1 d T T d T T P T T T T d ( 0 ) T T d P 1 T T d T d T d T d T T 1 1 urora 2007, Vienna 19
22 Results (LM: Mean-Variance P formulation): Problem Stages lk ssets Scenarios Constraints Variables iter time procs machine LM SunFire 15K LM LM LM UNS LM LM LM urora 2007, Vienna 20
23 HPCx (Daresbury, England) 1600 IM Power-4 Processors 1.7GHz, 800Mb R max = 6.2 TFlops #100 in top500.org list luegene/l (Edinburgh, Scotland) 2048 Processors 0.7GHz, 256Mb R max = 4.7 TFlops #148 in top500.org list urora 2007, Vienna 21
24 LM: Size of largest problem Optimization of 21 assets (stock market indices) Scenario tree needs to capture: correlations of 21 assets correlations over time periods 100 branches over 6 stages = scenarios etter: Scenario tree geometry: million scenarios. Scenario tree generated using simulated geometric brownian motion with mean reversion billion variables, 353 million constraints urora 2007, Vienna 22
25 Issues for Massively Parallel Implementation Communications Memory/Data management Granularity of Computations Sparsity of multilevel linear algebra ootstrapping How to get the data onto the processors urora 2007, Vienna 23
26 Example: ordered lock-diagonal Structure (Schur complement) Φ Φ n n } 1 n {{ Φ 0 } Φ L 1... = L n } L 1,0 L n,0 {{ L 0 } L D 1... } {{ D 0 } D Cholesky-like factors can be obtained by Schur-complement: L i,0 = i L i Φ i = L i D i L i C = Φ 0 n i=1 L i,0d i L i,0 C = L 0 D 0 L 0 nd the system Φx = b can be solved by L 1 L 1,0.... L n L n,0 L 0 } {{ } L Di 1, i = 1,... n z i = L 1 i b i z 0 = L 1 0 (b 0 L i,0 z i ) y i = Di 1 z i x 0 = L 0 y 0 x i = L i (y i L i,0x 0 ) Operations (Cholesky, Solve, Product) are only performed on sub-blocks Can also exploit structure in sub-blocks urora 2007, Vienna 24
27 Exploiting Parallelism: ordered lock-diagonal Structure: Distribution of computations: proc 1 proc n Φ 1= L D L C = L D L Φ = L D L n T n n T n 1 1 T T C = L D L n T 1 T n n n n n C = Φ 0 Σ ic i T 0 C = L D L 0 0 Factorize Solve proc 1 proc n 1 1 z = L b z n= L n bn 1 1 T 1 l = L D z 1 1 T l n= nl n D nzn l = b 0 Σi l i z = L l y = D z n y n= D n z y = D z T 0 0 x 0 = L y T T x = L (y L D x ) T T x n= L n (y n nl n D x ) n 0 Storage: Φ 1 T 1 x1 processor 1 On separate processors On all processors Φ n T n x n processor n Communications 1 n Φ 0 x 0 all processors y y y 1 n 0 urora 2007, Vienna 25
28 Memory Management Data for problem requires 2.6G of memory. need to split information between processors To each node in block-elimination tree a set of processors is assigned Linear lgebra is implemented so that processors communicate when needed Distribution of leading matrix blocks among processors implies Distribution of subordinate blocks Distribution of row/column vector contributions urora 2007, Vienna 26
29 Granularity of Computations: Parallel division of the problem In LM problems matrices up to variables can be treated as unstructured sparse matrices on one processor Problem has: 128 first level nodes with variables each second level nodes with variables each. need to decompose problem at second level (with 1280 processors 3 blocks per processor) 128 nodes 30 nodes 30 nodes urora 2007, Vienna 27
30 Sparsity of Linear lgebra (nested ordered lock-diagonal) Φ 1 1 T.... Φ = Φ n n T 1 n Φ 0 Factorize matrix by Schur complement: Factor Φ i = L i D i L T i, Φ = LLT, L = uild C = Φ 0 i (L 1 i i T)T D i (L 1 i i T) Factor C = L c D c L T c L 1... L n L 0,1 L 0,n C Work tends to be dominated by forming V i = L 1 i T i and V T i V i. Keep Vi as sparse as possible. Requires careful exploitation of sparsity when L i itself is available through Schur-complement. urora 2007, Vienna 28
31 Sparsity of Multilevel lgebra Dominant cost is forming Schur complement: n C = Φ 0 Vi T D i V i, where V i = L 1 i=1 Need to solve a system for each column of T i i T i? =? = T Φ i = L i D i Li i T L i V i = T i urora 2007, Vienna 29
32 Sparsity of Multilevel lgebra Dominant cost is forming Schur complement: n C = Φ 0 Vi T D i V i, where V i = L 1 i=1 Need to solve a system for each column of T i i T i = = T Φ i = L i D i Li i T L i V i = T i urora 2007, Vienna 29
33 ootstrapping Problem: Data for the problem is 25 G. We have 512M/node on luegene. How to get the data on the processors? Data can be represented as core: 64 variables, 22 constraints scenario tree (topology, probability, returns for each node): 2.6 G Still does not fit on one processor Scenario tree data (probability, returns) can be generated/read in on each processor as needed Scenario tree topology needs 30 M This is all that OOPS needs to decide on the allocation of blocks to processors urora 2007, Vienna 30
34 ootstrapping II 1. Read problem dimensions, generate scenario tree topology and matrix tree (30M) 2. Call OOPS: Divide matrix tree among processors, set up data structures on each processor 3. Generate/Read in scenario tree data on every processor 4. On every processor: Call-back to generator to fill in SparseMatrix data 5. Start solution process urora 2007, Vienna 31
35 ootstrapping Set up (shape of) Scenario Tree urora 2007, Vienna 32
36 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) urora 2007, Vienna 32
37 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors 1,2,3,4 5,6,7,8 urora 2007, Vienna 32
38 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors 1,2,3,4 1,2 3,4 5,6,7,8 5,6 7,8 urora 2007, Vienna 32
39 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors 1,2,3,4 5,6,7,8 1,2 3,4 5,6 7, urora 2007, Vienna 32
40 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors Divide between processors 1,2,3,4 5,6,7,8 1,2 3,4 5, ,8 7 8 urora 2007, Vienna 32
41 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors Divide between processors Load data on each processor (core + generate scenarios) 1,2,3,4 5,6,7,8 1,2 3,4 5, ,8 7 8 urora 2007, Vienna 32
42 ootstrapping Set up (shape of) Scenario Tree Convert to lgebra Tree (without data) ssign processors Divide between processors Load data on each processor (core + generate scenarios) 1,2,3,4 5,6,7,8 1,2 3,4 5, Solve 7,8 7 8 urora 2007, Vienna 32
43 Results Stages lk ssets Scenarios Constraints Variables iter time procs machine ,831,873 64,159, ,982, G/L ,415,937 96,239, ,469, G/L ,831, ,646, ,443, G/L ,039, ,875,799 1,010,507, HPCx urora 2007, Vienna 33
44 Results Stages lk ssets Scenarios Constraints Variables iter time procs machine ,831,873 64,159, ,982, G/L ,415,937 96,239, ,469, G/L ,831, ,646, ,443, G/L ,039, ,875,799 1,010,507, HPCx Parallel Efficiency/Scaling nodes peak Mem time Comm Cholesky Solves MatVectProd M 2587 (1.00) (1.00) 956 (1.00) 28.8 (1.00) M 1303 (0.99) (1.00) 485 (0.98) 18.0 (0.80) M 688 (0.94) (0.98) 270 (0.88) 13.0 (0.55) M 348 (0.93) (0.99) 139 (0.86) 9.0 (0.40) M 179 (0.90) 3 93 (0.99) 73 (0.82) 5.8 (0.31) M 94 (0.86) 2 47 (0.98) 39 (0.76) 3.9 (0.23) urora 2007, Vienna 33
45 Further Developments: Warmstarting Possibilities for warmstarting are rarely used in Stochastic Programmming Solution approaches for SP (enders Decomposition/IPM) don t warmstart well Interior Point Methods cannot be warmstarted urora 2007, Vienna 34
46 Results (Efficient Frontier) Exp(X) Mean Variance formulation: max IE(X) ρvar(x) StdVar(X) urora 2007, Vienna 35
47 Results (Efficient Frontier) Exp(X) Mean Variance formulation: max IE(X) ρvar(x) StdVar(X) constraints variables procs ρ = ,321 76, , , ,982,604 16,316, ,575, ,478, %-60% of iterations saved on average urora 2007, Vienna 36
48 Idea: Use Warmstart to speed up solution of Stochastic Program Solve problem on a reduced scenario tree first Copy solution vectors from nearby scenarios to construct starting point for full problem. Solve full problem (using this starting point) Saves up to 60% of iterations on problem up to 4096 scenarios. urora 2007, Vienna 37
49 Construction of the warm-start iterate urora 2007, Vienna 38
50 Construction of the warm-start iterate Nodes in the reduced tree: the solution is already available urora 2007, Vienna 38
51 Construction of the warm-start iterate Nodes in the reduced tree: the solution is already available Remaining nodes: copy the solution from the corresponding reduced-tree node urora 2007, Vienna 38
52 Conclusions: OOPS is a general purpose IPM solver with object-oriented linear algebra Will scale up to massively parallel architecture Can solve problems with 10 9 variables using direct factorization methods llows the solution of realistic financial planning problems Future Work on OOPS: Extend warmstarting procedure: IPM difficult to warmstart Multi-Step Scheme Complexity of such a scheme Carry over to other structures (PDE constrained optimization) Incorporation of iterative solvers (structured pre-conditioners) Integrate into structured modelling language urora 2007, Vienna 39
53 Object-Oriented Parallel Solver (OOPS): References: J. Gondzio and. Grothey, Parallel interior point solver for structured quadratic programs: application to financial planning problems, nnals of OR 152 (2007), Vol 1, pp J. Gondzio and. Grothey, Solving nonlinear portfolio optimization problems with the primal-dual interior point method, European Journal of Operational Research, 181 (2007), Vol 3, p J. Gondzio and. Grothey, Exploiting Structure in Parallel Implementation of Interior Point Methods for Optimization, Tech. Rep. MS , School of Maths, University of Edinburgh, Dec J. Gondzio and. Grothey, Direct Solution of Linear Systems of Size 10 9 rising in Optimization with Interior Point Methods,in: Parallel Processing and pplied Mathematics, Lecture Notes in Computer Sciences urora 2007, Vienna 40
Parallel Interior Point Solver for Structured Quadratic Programs: Application to Financial Planning Problems
Parallel Interior Point Solver for Structured uadratic Programs: Application to Financial Planning Problems Jacek Gondzio Andreas Grothey April 15th, 2003 MS-03-001 For other papers in this series see
More informationBig Data Optimization at SAS
Big Data Optimization at SAS Imre Pólik et al. SAS Institute Cary, NC, USA Edinburgh, 2013 Outline 1 Optimization at SAS 2 Big Data Optimization at SAS The SAS HPA architecture Support vector machines
More informationAn Overview Of Software For Convex Optimization. Brian Borchers Department of Mathematics New Mexico Tech Socorro, NM 87801 borchers@nmt.
An Overview Of Software For Convex Optimization Brian Borchers Department of Mathematics New Mexico Tech Socorro, NM 87801 borchers@nmt.edu In fact, the great watershed in optimization isn t between linearity
More informationCASH FLOW MATCHING PROBLEM WITH CVaR CONSTRAINTS: A CASE STUDY WITH PORTFOLIO SAFEGUARD. Danjue Shang and Stan Uryasev
CASH FLOW MATCHING PROBLEM WITH CVaR CONSTRAINTS: A CASE STUDY WITH PORTFOLIO SAFEGUARD Danjue Shang and Stan Uryasev PROJECT REPORT #2011-1 Risk Management and Financial Engineering Lab Department of
More informationSolving polynomial least squares problems via semidefinite programming relaxations
Solving polynomial least squares problems via semidefinite programming relaxations Sunyoung Kim and Masakazu Kojima August 2007, revised in November, 2007 Abstract. A polynomial optimization problem whose
More informationYousef Saad University of Minnesota Computer Science and Engineering. CRM Montreal - April 30, 2008
A tutorial on: Iterative methods for Sparse Matrix Problems Yousef Saad University of Minnesota Computer Science and Engineering CRM Montreal - April 30, 2008 Outline Part 1 Sparse matrices and sparsity
More informationParallel Computing for Option Pricing Based on the Backward Stochastic Differential Equation
Parallel Computing for Option Pricing Based on the Backward Stochastic Differential Equation Ying Peng, Bin Gong, Hui Liu, and Yanxin Zhang School of Computer Science and Technology, Shandong University,
More informationComputing a Nearest Correlation Matrix with Factor Structure
Computing a Nearest Correlation Matrix with Factor Structure Nick Higham School of Mathematics The University of Manchester higham@ma.man.ac.uk http://www.ma.man.ac.uk/~higham/ Joint work with Rüdiger
More informationTwo-Stage Stochastic Linear Programs
Two-Stage Stochastic Linear Programs Operations Research Anthony Papavasiliou 1 / 27 Two-Stage Stochastic Linear Programs 1 Short Reviews Probability Spaces and Random Variables Convex Analysis 2 Deterministic
More informationOptimization Methods in Finance
Optimization Methods in Finance Gerard Cornuejols Reha Tütüncü Carnegie Mellon University, Pittsburgh, PA 15213 USA January 2006 2 Foreword Optimization models play an increasingly important role in financial
More informationMixed Precision Iterative Refinement Methods Energy Efficiency on Hybrid Hardware Platforms
Mixed Precision Iterative Refinement Methods Energy Efficiency on Hybrid Hardware Platforms Björn Rocker Hamburg, June 17th 2010 Engineering Mathematics and Computing Lab (EMCL) KIT University of the State
More informationA matrix-free preconditioner for sparse symmetric positive definite systems and least square problems
A matrix-free preconditioner for sparse symmetric positive definite systems and least square problems Stefania Bellavia Dipartimento di Ingegneria Industriale Università degli Studi di Firenze Joint work
More informationIntroduction to Stochastic Optimization in Supply Chain and Logistic Optimization
Introduction to Stochastic Optimization in Supply Chain and Logistic Optimization John R. Birge Northwestern University IMA Tutorial, Stochastic Optimization, September 00 1 Outline Overview Part I - Models
More informationStochastic programming approach to ALM in Finnish pension insurance companies p.1/36
Stochastic programming approach to ALM in Finnish pension insurance companies Aktuaaritoiminnan kehittämissäätiön syysseminaari 17.11.2004 Teemu Pennanen Helsinki School of Economics Stochastic programming
More informationUsing Microsoft Excel to build Efficient Frontiers via the Mean Variance Optimization Method
Using Microsoft Excel to build Efficient Frontiers via the Mean Variance Optimization Method Submitted by John Alexander McNair ID #: 0061216 Date: April 14, 2003 The Optimal Portfolio Problem Consider
More informationFRIEDRICH-ALEXANDER-UNIVERSITÄT ERLANGEN-NÜRNBERG
FRIEDRICH-ALEXANDER-UNIVERSITÄT ERLANGEN-NÜRNBERG INSTITUT FÜR INFORMATIK (MATHEMATISCHE MASCHINEN UND DATENVERARBEITUNG) Lehrstuhl für Informatik 10 (Systemsimulation) Massively Parallel Multilevel Finite
More informationChapter 1 INTRODUCTION. 1.1 Background
Chapter 1 INTRODUCTION 1.1 Background This thesis attempts to enhance the body of knowledge regarding quantitative equity (stocks) portfolio selection. A major step in quantitative management of investment
More informationAdvanced Lecture on Mathematical Science and Information Science I. Optimization in Finance
Advanced Lecture on Mathematical Science and Information Science I Optimization in Finance Reha H. Tütüncü Visiting Associate Professor Dept. of Mathematical and Computing Sciences Tokyo Institute of Technology
More informationA Direct Numerical Method for Observability Analysis
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL 15, NO 2, MAY 2000 625 A Direct Numerical Method for Observability Analysis Bei Gou and Ali Abur, Senior Member, IEEE Abstract This paper presents an algebraic method
More informationDATA ANALYSIS II. Matrix Algorithms
DATA ANALYSIS II Matrix Algorithms Similarity Matrix Given a dataset D = {x i }, i=1,..,n consisting of n points in R d, let A denote the n n symmetric similarity matrix between the points, given as where
More informationLecture 3. Linear Programming. 3B1B Optimization Michaelmas 2015 A. Zisserman. Extreme solutions. Simplex method. Interior point method
Lecture 3 3B1B Optimization Michaelmas 2015 A. Zisserman Linear Programming Extreme solutions Simplex method Interior point method Integer programming and relaxation The Optimization Tree Linear Programming
More informationSupport Vector Machines with Clustering for Training with Very Large Datasets
Support Vector Machines with Clustering for Training with Very Large Datasets Theodoros Evgeniou Technology Management INSEAD Bd de Constance, Fontainebleau 77300, France theodoros.evgeniou@insead.fr Massimiliano
More informationAN INTRODUCTION TO NUMERICAL METHODS AND ANALYSIS
AN INTRODUCTION TO NUMERICAL METHODS AND ANALYSIS Revised Edition James Epperson Mathematical Reviews BICENTENNIAL 0, 1 8 0 7 z ewiley wu 2007 r71 BICENTENNIAL WILEY-INTERSCIENCE A John Wiley & Sons, Inc.,
More informationSupport Vector Machines
Support Vector Machines Charlie Frogner 1 MIT 2011 1 Slides mostly stolen from Ryan Rifkin (Google). Plan Regularization derivation of SVMs. Analyzing the SVM problem: optimization, duality. Geometric
More informationDistributed communication-aware load balancing with TreeMatch in Charm++
Distributed communication-aware load balancing with TreeMatch in Charm++ The 9th Scheduling for Large Scale Systems Workshop, Lyon, France Emmanuel Jeannot Guillaume Mercier Francois Tessier In collaboration
More informationFast Multipole Method for particle interactions: an open source parallel library component
Fast Multipole Method for particle interactions: an open source parallel library component F. A. Cruz 1,M.G.Knepley 2,andL.A.Barba 1 1 Department of Mathematics, University of Bristol, University Walk,
More informationACCELERATING COMMERCIAL LINEAR DYNAMIC AND NONLINEAR IMPLICIT FEA SOFTWARE THROUGH HIGH- PERFORMANCE COMPUTING
ACCELERATING COMMERCIAL LINEAR DYNAMIC AND Vladimir Belsky Director of Solver Development* Luis Crivelli Director of Solver Development* Matt Dunbar Chief Architect* Mikhail Belyi Development Group Manager*
More informationOptimization applications in finance, securities, banking and insurance
IBM Software IBM ILOG Optimization and Analytical Decision Support Solutions White Paper Optimization applications in finance, securities, banking and insurance 2 Optimization applications in finance,
More informationParallel Data Selection Based on Neurodynamic Optimization in the Era of Big Data
Parallel Data Selection Based on Neurodynamic Optimization in the Era of Big Data Jun Wang Department of Mechanical and Automation Engineering The Chinese University of Hong Kong Shatin, New Territories,
More informationSECOND DERIVATIVE TEST FOR CONSTRAINED EXTREMA
SECOND DERIVATIVE TEST FOR CONSTRAINED EXTREMA This handout presents the second derivative test for a local extrema of a Lagrange multiplier problem. The Section 1 presents a geometric motivation for the
More informationMesh Generation and Load Balancing
Mesh Generation and Load Balancing Stan Tomov Innovative Computing Laboratory Computer Science Department The University of Tennessee April 04, 2012 CS 594 04/04/2012 Slide 1 / 19 Outline Motivation Reliable
More informationParallel Programming at the Exascale Era: A Case Study on Parallelizing Matrix Assembly For Unstructured Meshes
Parallel Programming at the Exascale Era: A Case Study on Parallelizing Matrix Assembly For Unstructured Meshes Eric Petit, Loïc Thebault, Quang V. Dinh May 2014 EXA2CT Consortium 2 WPs Organization Proto-Applications
More informationBest practices for efficient HPC performance with large models
Best practices for efficient HPC performance with large models Dr. Hößl Bernhard, CADFEM (Austria) GmbH PRACE Autumn School 2013 - Industry Oriented HPC Simulations, September 21-27, University of Ljubljana,
More informationNumerical Methods I Solving Linear Systems: Sparse Matrices, Iterative Methods and Non-Square Systems
Numerical Methods I Solving Linear Systems: Sparse Matrices, Iterative Methods and Non-Square Systems Aleksandar Donev Courant Institute, NYU 1 donev@courant.nyu.edu 1 Course G63.2010.001 / G22.2420-001,
More informationWhy is SAS/OR important? For whom is SAS/OR designed?
Fact Sheet What does SAS/OR software do? SAS/OR software provides a powerful array of optimization, simulation and project scheduling techniques to identify the actions that will produce the best results,
More informationPower-Aware High-Performance Scientific Computing
Power-Aware High-Performance Scientific Computing Padma Raghavan Scalable Computing Laboratory Department of Computer Science Engineering The Pennsylvania State University http://www.cse.psu.edu/~raghavan
More informationSummer course on Convex Optimization. Fifth Lecture Interior-Point Methods (1) Michel Baes, K.U.Leuven Bharath Rangarajan, U.
Summer course on Convex Optimization Fifth Lecture Interior-Point Methods (1) Michel Baes, K.U.Leuven Bharath Rangarajan, U.Minnesota Interior-Point Methods: the rebirth of an old idea Suppose that f is
More informationOptimization Modeling for Mining Engineers
Optimization Modeling for Mining Engineers Alexandra M. Newman Division of Economics and Business Slide 1 Colorado School of Mines Seminar Outline Linear Programming Integer Linear Programming Slide 2
More informationA Weighted-Sum Mixed Integer Program for Bi-Objective Dynamic Portfolio Optimization
AUTOMATYKA 2009 Tom 3 Zeszyt 2 Bartosz Sawik* A Weighted-Sum Mixed Integer Program for Bi-Objective Dynamic Portfolio Optimization. Introduction The optimal security selection is a classical portfolio
More informationAPPM4720/5720: Fast algorithms for big data. Gunnar Martinsson The University of Colorado at Boulder
APPM4720/5720: Fast algorithms for big data Gunnar Martinsson The University of Colorado at Boulder Course objectives: The purpose of this course is to teach efficient algorithms for processing very large
More informationDynamic Asset Allocation Using Stochastic Programming and Stochastic Dynamic Programming Techniques
Dynamic Asset Allocation Using Stochastic Programming and Stochastic Dynamic Programming Techniques Gerd Infanger Stanford University Winter 2011/2012 MS&E348/Infanger 1 Outline Motivation Background and
More information2004 Networks UK Publishers. Reprinted with permission.
Riikka Susitaival and Samuli Aalto. Adaptive load balancing with OSPF. In Proceedings of the Second International Working Conference on Performance Modelling and Evaluation of Heterogeneous Networks (HET
More informationSeveral Views of Support Vector Machines
Several Views of Support Vector Machines Ryan M. Rifkin Honda Research Institute USA, Inc. Human Intention Understanding Group 2007 Tikhonov Regularization We are considering algorithms of the form min
More informationSupply Chain Design and Inventory Management Optimization in the Motors Industry
A publication of 1171 CHEMICAL ENGINEERING TRANSACTIONS VOL. 32, 2013 Chief Editors: Sauro Pierucci, Jiří J. Klemeš Copyright 2013, AIDIC Servizi S.r.l., ISBN 978-88-95608-23-5; ISSN 1974-9791 The Italian
More informationOPTIMIZATION AND FORECASTING WITH FINANCIAL TIME SERIES
OPTIMIZATION AND FORECASTING WITH FINANCIAL TIME SERIES Allan Din Geneva Research Collaboration Notes from seminar at CERN, June 25, 2002 General scope of GRC research activities Econophysics paradigm
More information2014-2015 The Master s Degree with Thesis Course Descriptions in Industrial Engineering
2014-2015 The Master s Degree with Thesis Course Descriptions in Industrial Engineering Compulsory Courses IENG540 Optimization Models and Algorithms In the course important deterministic optimization
More informationMathematical finance and linear programming (optimization)
Mathematical finance and linear programming (optimization) Geir Dahl September 15, 2009 1 Introduction The purpose of this short note is to explain how linear programming (LP) (=linear optimization) may
More informationNonlinear Optimization: Algorithms 3: Interior-point methods
Nonlinear Optimization: Algorithms 3: Interior-point methods INSEAD, Spring 2006 Jean-Philippe Vert Ecole des Mines de Paris Jean-Philippe.Vert@mines.org Nonlinear optimization c 2006 Jean-Philippe Vert,
More informationTOMLAB - For fast and robust largescale optimization in MATLAB
The TOMLAB Optimization Environment is a powerful optimization and modeling package for solving applied optimization problems in MATLAB. TOMLAB provides a wide range of features, tools and services for
More informationParallel Algorithm for Dense Matrix Multiplication
Parallel Algorithm for Dense Matrix Multiplication CSE633 Parallel Algorithms Fall 2012 Ortega, Patricia Outline Problem definition Assumptions Implementation Test Results Future work Conclusions Problem
More informationNon-negative Matrix Factorization (NMF) in Semi-supervised Learning Reducing Dimension and Maintaining Meaning
Non-negative Matrix Factorization (NMF) in Semi-supervised Learning Reducing Dimension and Maintaining Meaning SAMSI 10 May 2013 Outline Introduction to NMF Applications Motivations NMF as a middle step
More informationIntroduction to Linear Programming (LP) Mathematical Programming (MP) Concept
Introduction to Linear Programming (LP) Mathematical Programming Concept LP Concept Standard Form Assumptions Consequences of Assumptions Solution Approach Solution Methods Typical Formulations Massachusetts
More informationScalable Distributed Schur Complement Solvers for Internal and External Flow Computations on Many-Core Architectures
Scalable Distributed Schur Complement Solvers for Internal and External Flow Computations on Many-Core Architectures Dr.-Ing. Achim Basermann, Dr. Hans-Peter Kersken, Melven Zöllner** German Aerospace
More informationAN INTERFACE STRIP PRECONDITIONER FOR DOMAIN DECOMPOSITION METHODS
AN INTERFACE STRIP PRECONDITIONER FOR DOMAIN DECOMPOSITION METHODS by M. Storti, L. Dalcín, R. Paz Centro Internacional de Métodos Numéricos en Ingeniería - CIMEC INTEC, (CONICET-UNL), Santa Fe, Argentina
More informationMinkowski Sum of Polytopes Defined by Their Vertices
Minkowski Sum of Polytopes Defined by Their Vertices Vincent Delos, Denis Teissandier To cite this version: Vincent Delos, Denis Teissandier. Minkowski Sum of Polytopes Defined by Their Vertices. Journal
More informationFast Iterative Solvers for Integral Equation Based Techniques in Electromagnetics
Fast Iterative Solvers for Integral Equation Based Techniques in Electromagnetics Mario Echeverri, PhD. Student (2 nd year, presently doing a research period abroad) ID:30360 Tutor: Prof. Francesca Vipiana,
More informationYALES2 porting on the Xeon- Phi Early results
YALES2 porting on the Xeon- Phi Early results Othman Bouizi Ghislain Lartigue Innovation and Pathfinding Architecture Group in Europe, Exascale Lab. Paris CRIHAN - Demi-journée calcul intensif, 16 juin
More informationPortfolio selection based on upper and lower exponential possibility distributions
European Journal of Operational Research 114 (1999) 115±126 Theory and Methodology Portfolio selection based on upper and lower exponential possibility distributions Hideo Tanaka *, Peijun Guo Department
More informationFPGA Acceleration of Mean Variance Framework for Optimal Asset Allocation
FPGA Acceleration of Mean Variance Framework for Optimal Asset Allocation Ali Irturk, Bridget Benson, Nikolay Laptev, Ryan Kastner Department of Computer Science and Engineering University of California,
More informationFUZZY CLUSTERING ANALYSIS OF DATA MINING: APPLICATION TO AN ACCIDENT MINING SYSTEM
International Journal of Innovative Computing, Information and Control ICIC International c 0 ISSN 34-48 Volume 8, Number 8, August 0 pp. 4 FUZZY CLUSTERING ANALYSIS OF DATA MINING: APPLICATION TO AN ACCIDENT
More informationClass #6: Non-linear classification. ML4Bio 2012 February 17 th, 2012 Quaid Morris
Class #6: Non-linear classification ML4Bio 2012 February 17 th, 2012 Quaid Morris 1 Module #: Title of Module 2 Review Overview Linear separability Non-linear classification Linear Support Vector Machines
More informationLinear Programming for Optimization. Mark A. Schulze, Ph.D. Perceptive Scientific Instruments, Inc.
1. Introduction Linear Programming for Optimization Mark A. Schulze, Ph.D. Perceptive Scientific Instruments, Inc. 1.1 Definition Linear programming is the name of a branch of applied mathematics that
More informationModern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh
Modern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh Peter Richtárik Week 3 Randomized Coordinate Descent With Arbitrary Sampling January 27, 2016 1 / 30 The Problem
More informationOptimal Scheduling for Dependent Details Processing Using MS Excel Solver
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 8, No 2 Sofia 2008 Optimal Scheduling for Dependent Details Processing Using MS Excel Solver Daniela Borissova Institute of
More informationAsset Liability Management / Liability Driven Investment Optimization (LDIOpt)
Asset Liability Management / Liability Driven Investment Optimization (LDIOpt) Introduction ALM CASH FLOWS OptiRisk Liability Driven Investment Optimization LDIOpt is an asset and liability management
More informationApplying Stochastic Programming Models in Financial Risk Management. Xi Yang
Applying Stochastic Programming Models in Financial Risk Management Xi Yang Doctor of Philosophy University of Edinburgh 2009 Declaration I declare that this thesis was composed by myself and that the
More informationOperations Research and Financial Engineering. Courses
Operations Research and Financial Engineering Courses ORF 504/FIN 504 Financial Econometrics Professor Jianqing Fan This course covers econometric and statistical methods as applied to finance. Topics
More informationProximal mapping via network optimization
L. Vandenberghe EE236C (Spring 23-4) Proximal mapping via network optimization minimum cut and maximum flow problems parametric minimum cut problem application to proximal mapping Introduction this lecture:
More informationA Genetic Algorithm Approach for Solving a Flexible Job Shop Scheduling Problem
A Genetic Algorithm Approach for Solving a Flexible Job Shop Scheduling Problem Sayedmohammadreza Vaghefinezhad 1, Kuan Yew Wong 2 1 Department of Manufacturing & Industrial Engineering, Faculty of Mechanical
More informationAbstract. 1. Introduction. Caparica, Portugal b CEG, IST-UTL, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
Ian David Lockhart Bogle and Michael Fairweather (Editors), Proceedings of the 22nd European Symposium on Computer Aided Process Engineering, 17-20 June 2012, London. 2012 Elsevier B.V. All rights reserved.
More informationThis paper is not to be removed from the Examination Halls
~~FN3023 ZB d0 This paper is not to be removed from the Examination Halls UNIVERSITY OF LONDON FN3023 ZB BSc degrees and Diplomas for Graduates in Economics, Management, Finance and the Social Sciences,
More informationA Simultaneous Solution for General Linear Equations on a Ring or Hierarchical Cluster
Acta Technica Jaurinensis Vol. 3. No. 1. 010 A Simultaneous Solution for General Linear Equations on a Ring or Hierarchical Cluster G. Molnárka, N. Varjasi Széchenyi István University Győr, Hungary, H-906
More informationEfficient and Robust Allocation Algorithms in Clouds under Memory Constraints
Efficient and Robust Allocation Algorithms in Clouds under Memory Constraints Olivier Beaumont,, Paul Renaud-Goud Inria & University of Bordeaux Bordeaux, France 9th Scheduling for Large Scale Systems
More informationInterior Point Methods and Linear Programming
Interior Point Methods and Linear Programming Robert Robere University of Toronto December 13, 2012 Abstract The linear programming problem is usually solved through the use of one of two algorithms: either
More informationCourse Syllabus For Operations Management. Management Information Systems
For Operations Management and Management Information Systems Department School Year First Year First Year First Year Second year Second year Second year Third year Third year Third year Third year Third
More informationHPC enabling of OpenFOAM R for CFD applications
HPC enabling of OpenFOAM R for CFD applications Towards the exascale: OpenFOAM perspective Ivan Spisso 25-27 March 2015, Casalecchio di Reno, BOLOGNA. SuperComputing Applications and Innovation Department,
More informationBuilding and Solving Large-scale Stochastic Programs on an Aæordable Distributed Computing System æ Emmanuel Fragniçere y, Jacek Gondzio z, Jean-Philippe Vial Logilab, HEC, Section of Management Studies,
More informationSimplified Benders cuts for Facility Location
Simplified Benders cuts for Facility Location Matteo Fischetti, University of Padova based on joint work with Ivana Ljubic (ESSEC, Paris) and Markus Sinnl (ISOR, Vienna) Barcelona, November 2015 1 Apology
More informationSolving Large Scale Optimization Problems via Grid and Cluster Computing
Solving Large Scale Optimization Problems via Grid and Cluster Computing Katsuki Fujisawa, Masakazu Kojima Akiko Takeda and Makoto Yamashita Abstract. Solving large scale optimization problems requires
More informationSanjeev Kumar. contribute
RESEARCH ISSUES IN DATAA MINING Sanjeev Kumar I.A.S.R.I., Library Avenue, Pusa, New Delhi-110012 sanjeevk@iasri.res.in 1. Introduction The field of data mining and knowledgee discovery is emerging as a
More informationModelling Incomplete Markets for Long Term Asset Liability Management
INI DQF Workshop on Modelling Philosophy 22 April 2005 Modelling Incomplete Markets for Long Term Asset Liability Management M A H Dempster Centre for Financial Research Judge Institute of Management University
More informationOptimization 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 informationControlling the Internet in the era of Software Defined and Virtualized Networks. Fernando Paganini Universidad ORT Uruguay
Controlling the Internet in the era of Software Defined and Virtualized Networks Fernando Paganini Universidad ORT Uruguay CDS@20, Caltech 2014 Motivation 1. The Internet grew in its first 30 years with
More informationNumerical Methods for Pricing Exotic Options
Numerical Methods for Pricing Exotic Options Dimitra Bampou Supervisor: Dr. Daniel Kuhn Second Marker: Professor Berç Rustem 18 June 2008 2 Numerical Methods for Pricing Exotic Options 0BAbstract 3 Abstract
More informationDepartment of Chemical Engineering ChE-101: Approaches to Chemical Engineering Problem Solving MATLAB Tutorial VI
Department of Chemical Engineering ChE-101: Approaches to Chemical Engineering Problem Solving MATLAB Tutorial VI Solving a System of Linear Algebraic Equations (last updated 5/19/05 by GGB) Objectives:
More informationIt s Not A Disease: The Parallel Solver Packages MUMPS, PaStiX & SuperLU
It s Not A Disease: The Parallel Solver Packages MUMPS, PaStiX & SuperLU A. Windisch PhD Seminar: High Performance Computing II G. Haase March 29 th, 2012, Graz Outline 1 MUMPS 2 PaStiX 3 SuperLU 4 Summary
More informationLECTURE 5: DUALITY AND SENSITIVITY ANALYSIS. 1. Dual linear program 2. Duality theory 3. Sensitivity analysis 4. Dual simplex method
LECTURE 5: DUALITY AND SENSITIVITY ANALYSIS 1. Dual linear program 2. Duality theory 3. Sensitivity analysis 4. Dual simplex method Introduction to dual linear program Given a constraint matrix A, right
More informationStochastic programming approaches to pricing in non-life insurance
Stochastic programming approaches to pricing in non-life insurance Martin Branda Charles University in Prague Department of Probability and Mathematical Statistics 11th International Conference on COMPUTATIONAL
More informationBIG DATA PROBLEMS AND LARGE-SCALE OPTIMIZATION: A DISTRIBUTED ALGORITHM FOR MATRIX FACTORIZATION
BIG DATA PROBLEMS AND LARGE-SCALE OPTIMIZATION: A DISTRIBUTED ALGORITHM FOR MATRIX FACTORIZATION Ş. İlker Birbil Sabancı University Ali Taylan Cemgil 1, Hazal Koptagel 1, Figen Öztoprak 2, Umut Şimşekli
More informationChapter 2: Binomial Methods and the Black-Scholes Formula
Chapter 2: Binomial Methods and the Black-Scholes Formula 2.1 Binomial Trees We consider a financial market consisting of a bond B t = B(t), a stock S t = S(t), and a call-option C t = C(t), where the
More informationConic optimization: examples and software
Conic optimization: examples and software Etienne de Klerk Tilburg University, The Netherlands Etienne de Klerk (Tilburg University) Conic optimization: examples and software 1 / 16 Outline Conic optimization
More informationAn interval linear programming contractor
An interval linear programming contractor Introduction Milan Hladík Abstract. We consider linear programming with interval data. One of the most challenging problems in this topic is to determine or tight
More informationIntroduction: Models, Model Building and Mathematical Optimization The Importance of Modeling Langauges for Solving Real World Problems
Introduction: Models, Model Building and Mathematical Optimization The Importance of Modeling Langauges for Solving Real World Problems Josef Kallrath Structure of the Lecture: the Modeling Process survey
More informationLimited Memory Solution of Complementarity Problems arising in Video Games
Laboratoire d Arithmétique, Calcul formel et d Optimisation UMR CNRS 69 Limited Memory Solution of Complementarity Problems arising in Video Games Michael C. Ferris Andrew J. Wathen Paul Armand Rapport
More informationSOLVING LINEAR SYSTEMS
SOLVING LINEAR SYSTEMS Linear systems Ax = b occur widely in applied mathematics They occur as direct formulations of real world problems; but more often, they occur as a part of the numerical analysis
More informationMaximize Performance and Scalability of RADIOSS* Structural Analysis Software on Intel Xeon Processor E7 v2 Family-Based Platforms
Maximize Performance and Scalability of RADIOSS* Structural Analysis Software on Family-Based Platforms Executive Summary Complex simulations of structural and systems performance, such as car crash simulations,
More informationLeast-Squares Intersection of Lines
Least-Squares Intersection of Lines Johannes Traa - UIUC 2013 This write-up derives the least-squares solution for the intersection of lines. In the general case, a set of lines will not intersect at a
More informationEquilibrium computation: Part 1
Equilibrium computation: Part 1 Nicola Gatti 1 Troels Bjerre Sorensen 2 1 Politecnico di Milano, Italy 2 Duke University, USA Nicola Gatti and Troels Bjerre Sørensen ( Politecnico di Milano, Italy, Equilibrium
More informationThe Methodology for the Mid-Term Forecasting of the Financial Results of Firms Belikov V.V., Gataullina R.I., Tregub I.V.
The Methodology for the Mid-Term Forecasting of the Financial Results of Firms Belikov V.V., Gataullina R.I., Tregub I.V. Abstract. The article is dedicated to the mid-term forecasting of the Company financial
More informationLarge-Scale Similarity and Distance Metric Learning
Large-Scale Similarity and Distance Metric Learning Aurélien Bellet Télécom ParisTech Joint work with K. Liu, Y. Shi and F. Sha (USC), S. Clémençon and I. Colin (Télécom ParisTech) Séminaire Criteo March
More information