Scribd Logo
Upload

Welcome to Scribd!

  • 25 views

    Premise - Design and Criterion Space - Pareto Optimality

    Uploaded by

    Rsn786
    This document discusses Pareto optimality, which is a concept in multi-objective optimization involving problems with two or more objectives. It provides examples to illustrate: 1) The feasible region or "design space" is graphed based on the objectives and constraints. Corner points are identified. 2) The design space is converted to "criterion space" where the axes are the objective functions rather than the variables. 3) Pareto optimal solutions exist along the "Pareto front" where improving one objective requires worsening another. No single solution is best, as people must balance priorities. 4) Solutions not on the Pareto front are "Pareto inefficient" as both objectives could be

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Premise - Design and Criterion Space - Pareto Optimality

    Uploaded by

    Rsn786
    25 views22 pages
    This document discusses Pareto optimality, which is a concept in multi-objective optimization involving problems with two or more objectives. It provides examples to illustrate: 1) The feasible region or "design space" is graphed based on the objectives and constraints. Corner points are identified. 2) The design space is converted to "criterion space" where the axes are the objective functions rather than the variables. 3) Pareto optimal solutions exist along the "Pareto front" where improving one objective requires worsening another. No single solution is best, as people must balance priorities. 4) Solutions not on the Pareto front are "Pareto inefficient" as both objectives could be

    Original Title

    ParetoPtm

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses Pareto optimality, which is a concept in multi-objective optimization involving problems with two or more objectives. It provides examples to illustrate: 1) The feasible region or "design space" is graphed based on the objectives and constraints. Corner points are identified. 2) The design space is converted to "criterion space" where the axes are the objective functions rather than the variables. 3) Pareto optimal solutions exist along the "Pareto front" where improving one objective requires worsening another. No single solution is best, as people must balance priorities. 4) Solutions not on the Pareto front are "Pareto inefficient" as both objectives could be

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    25 views22 pages

    Premise - Design and Criterion Space - Pareto Optimality

    Uploaded by

    Rsn786
    This document discusses Pareto optimality, which is a concept in multi-objective optimization involving problems with two or more objectives. It provides examples to illustrate: 1) The feasible region or "design space" is graphed based on the objectives and constraints. Corner points are identified. 2) The design space is converted to "criterion space" where the axes are the objective functions rather than the variables. 3) Pareto optimal solutions exist along the "Pareto front" where improving one objective requires worsening another. No single solution is best, as people must balance priorities. 4) Solutions not on the Pareto front are "Pareto inefficient" as both objectives could be

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 22

    Pareto Optimality

    • Premise
    • Design and Criterion Space
    • Pareto Optimality
    The Premise
    In Unit 3 (Linear programming), we learned how to
    solve problems like this one:
    A small business advertises through traditional
    media and personal appearances. Each ad
    campaign in traditional media costs about $2000,
    generating 2 new customers and 1 positive rating
    per month. Each personal appearance costs $500,
    generating 2 new customers and 5 positive ratings.
    The company wants at least 16 new customers and
    28 positive ratings per month. Minimize the
    advertising costs for this company.
    The Premise
    Now suppose the company in the previous example
    used linear programming to minimize costs, and
    found that their optimal solution was to skip
    traditional advertising altogether and just make
    personal appearances.
    Soon, the executives would realize that they were
    spending so much time at personal appearances
    that they might be tempted to hire another
    executive – hardly a way to minimize cost!
    The Premise
    This sort of scenario plays out quite often and is
    called multiobjective optimization. In it, there are
    two or more objectives (in this case, minimize cost
    AND time) along with constraints.
    Frequently, objectives are confused with constraints
    when initially framing a problem.
    For example, do you really only have 4 hours a week
    to spend doing chores? Or is that just a way of
    saying you want to minimize time?
    Constraints should only be used when there is a
    hard, non-negotiable limit on the resource.
    Practice Problem 1
    Decide if each restriction is best written as a
    constraint or an objective:
    a) A musician wants to spend at least 4 hours a day
    practicing.
    b) There are only 8 cups of flour in the pantry.
    c) The doorway is only 3 feet wide.
    d) A CEO wants at least 5 new contracts this year.
    e) An office manager wants to limit the number of
    lost staplers to 10 per month.
    f) A family must bring in at least $2000 to cover
    rent.
    Pareto Optimality
    One way to find good solutions to multiobjective
    problems is with Pareto optimality, named after
    economist Vilfredo Pareto.

    Pareto noticed that many economic solutions


    helped some people while hurting others. He was
    interested in finding solutions that helped some
    people without hurting anyone else. Solutions like
    this are now called “Pareto improvements.”
    Design Space
    The first step when solving a multiobjective
    problem is to get a handle on the feasible
    region. In our example, there were two
    constraints: minimum numbers of customers
    and positive ratings.

    (These should probably be objectives too, but


    let’s keep it simple!)
    Design Space
    Customers were gained at a rate of 2 per ad
    campaign and 2 per appearance, and the
    company wanted a minimum of 16:
    2x1 + 2x2 ≥ 16
    Positive reviews came in at 1 per ad and 5 per
    appearance, with a minimum of 28:
    x1 + 5x2 ≥ 28
    These create the feasible region:
    Design Space
    The graph of the feasible region for the constraints
    is called the design space for the problem. The two
    axes are x1 and x2, and the corner points are where
    the constraints cross.
    We already have a money objective:
    x2 minimize cost = 2000x1 + 500x2
    Now, suppose each personal
    (0, 8) (3, 5) appearance takes 2 hours and
    x1
    each ad campaign takes 1 hour.
    (28, 0) Our second objective is then:
    minimize time = x1 + 2x2.
    Design Space
    Evaluating the corner points for both objectives, we
    get:
    x2 cost $8500
    cost $4000 time 13h
    time 16h (optimal time)
    (optimal cost)(0, 8) (3, 5)
    x1
    (28, 0) cost $56000
    time 28h
    cost = 2000x1 + 500x2
    (horrible!)
    time = x1 + 2x2
    Design Space
    If there were only one point that minimized
    both cost and time, obviously that would be the
    solution; but often, there are different optimal
    points for each objective.
    x2 The entire line segment
    cost $4000 between the two
    time 16h cost $8500
    (0, 8) (3, 5) time 13h
    optimal points
    x1 represents possible
    (28, 0) solutions.
    Practice Problem 2
    Suppose a university outreach program provides
    two services: campus tour days and publicity
    mailing days. The outcomes and requirements of
    both services are shown here:
    Service Cost Hours # applicants # donors
    Tour 100 32 10 4
    mailing 500 16 2 6
    requirement ≤ 2600 ≤ 256 Maximize Maximize

    Graph the feasible region, find the two optimal


    points, and highlight the segment between them.
    Criterion Space
    After graphing the feasible region and finding its
    boundaries in design space, the next step is to
    convert it to criterion space. In criterion space,
    the axes are no longer x1 and x2, but f1 and f2,
    which are the two objective functions.
    Every point in design space has a corresponding
    point in criterion space, but it’s sufficient to just
    locate the corner points.
    Criterion Space
    In our business example, we had three corner
    points with the objective function values shown.
    If we call “cost” f1 and “time” f2, we can create
    the criterion space:
    design space criterion space
    x2 (56000, 28)
    f2
    cost $4000 cost $8500
    time 16h time 13h
    (0, 8) (4000, 16)
    (3, 5)
    x1 (8500, 13)
    (28, 0) f1
    cost $56000
    time 28h
    Practice Problem 3
    Convert your design space from problem 2 into
    criterion space.

    Highlight the segment between the two optimal


    points from problem 2.
    Pareto Optimality
    In the business example, we were trying to minimize
    time and cost.
    Note that the orange point in criterion space is the
    lowest value of f2 (time) and the red point is the
    lowest value of f1 (cost). The edge between them is
    called the Pareto Front.
    Any point on this front is considered
    f2
    “Pareto optimal”. By moving along
    the curve, you could minimize cost
    at the expense of time, or minimize
    time at the expense of cost, but you
    f1 can not improve both at once.
    Pareto Optimality
    What that means is, there is no mathematical “best”
    point along the Pareto front. Actual people would
    have to get together and figure out how they
    wanted to balance their priorities.
    The slope of our Pareto curve is
    f2 -3/4500, which reduces to -1/1500.
    (4000, 16) Using the slope, Cost time
    we can find 4000 16
    (8500, 13) some Pareto 5500 15
    f1 optimal 7000 14
    solutions: 8500 13
    Practice Problem 4
    Using the slope of the Pareto curve from your
    criterion space in problem 3, find all five whole-
    number and two fractional Pareto-optimal
    solutions.
    Pareto Improvements
    Another implication of the Pareto front is that any
    point in the feasible region that is not on the
    Pareto front is a bad solution. Either objective, or
    both, can be improved at no penalty to the other.
    Pareto improvement!
    Recall that an improvement f2
    that helps one objective
    without harming the other is
    called a Pareto improvement. f1
    not Pareto optimal
    (“Pareto inefficient”)
    Practice Problem 5
    Determine whether these practices by the
    university outreach program are Pareto optimal;
    if so, report the number of donors and
    applications produced:
    a) 7 tour days, 2 mailing days
    b) 1 tour day, 5 mailing days
    c) 5 tour days, 4.2 mailing days
    d) 6.5 tour days, 2.5 mailing days
    e) 7.5 tour days, 1 mailing day
    Pareto Curves and Solutions
    When there is an obvious solution, Pareto
    curves will find it. Here are two examples:
    (goal: minimization) (goal: maximization)
    f2 f2

    f1 f1
    The blue point minimizes Although orange is on the
    both f1 and f2. There is Pareto front, moving to
    only one Pareto-optimal purple costs very little f2
    solution. for huge gains in f1.
    Practice Problem 6
    If there is an obvious solution, identify it by
    color. If not, write “no obvious solution”.
    (goal: minimization) (goal: maximization)
    a. d.

    b. e.

    c. f.

    You might also like