30
Creative Modeling for Technology Visionaries Qualitative & Simplified Quantitative Modeling for Product Creation Module 13: Squeezing Little Models David E. Goldberg University of Illinois at Urbana- Champaign Urbana, Illinois 61801 [email protected]

MTV 13: Squeezing Little Models

Embed Size (px)

DESCRIPTION

Module 13 from Creative Modeling for Tech Vision. Picks up on Little Models theme of module 11 and shows how to get more out of first efforts by "squeezing" the little model to learn more.

Citation preview

Page 1: MTV 13: Squeezing Little Models

Creative Modeling forTechnology VisionariesQualitative & Simplified Quantitative Modeling for Product Creation

Module 13: Squeezing Little Models

David E. GoldbergUniversity of Illinois at Urbana-ChampaignUrbana, Illinois 61801

[email protected]

Page 2: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.2

Squeezing Little ModelsWould like to move from qual to quant in difficult domains.

Once model obtained, how to we improve it, squeeze it, and extend it?

In this way, little modeling begets more little modeling.

Want to squeeze the most out of little models possible.

Page 3: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.3

This ModuleReview deciding-doing model:

Beyond deciding-doing:Stretching: auxiliary models.

Modifying: modification to functional form.

Reusing: Same math, different app.

Generalizing little models: EOPs and ETPs

Some solvable classes.

Page 4: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.4

Deciding and Doing Model

Team size: n

Discussing what is to be done: T1

Total time to do the task alone: T2

Total time required for task completion:

Model integration via summation.

Page 5: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.5

Do the Math

Take derivative of T(n) with respect to n.

Set to zero.

Do it.

Efficient team size

Optimal time:

Page 6: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.6

Consider Turning Point DerivationTdecide = T1n

Tdo = T2/nSet equal to each other.

T1n* = T2/n*Same as before:

Not generally the case, but not bad approximation.

Page 7: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.7

A Specific Case

Page 8: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.8

Stretching a Model

Ways to stretch a model.

Dimensional analysis can help reveal essential form.

Can recast in useful terms.

Can add auxiliary models to basic model.

Page 9: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.9

Ratios Reveal Structure

Can non-dimensionalize with respect to optimal solutions.

Can non-dimensionalize with respect to meaningful benchmarks.

Deciding-doing examples:Non-dimensional representation: & ν.

Speed-up representation.

Page 10: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.10

Non-Dimensionalizing

Optimal team size:

Optimal time:

Plug into deciding-doing equation:

Page 11: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.11

Dimensionless Form

Page 12: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.12

Aside: Power Laws & log-logPower law: f(x) = axb

Consider log-log transformation.

ln f(x) = ln axb = ln a + blnx

log-log transforms power law to linear curve with slope b (and intercept lna).

Page 13: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.13

Speedup as Dimensionless Recasting

Page 14: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.14

Auxiliary Models: Decision Quality

Beyond efficiency: Quality

Solutions successfully proposed by individual team members with a probability p.

Solution quality:

Q increases monotonically with increased n.

Q is high when

Page 15: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.15

An Illustrative Example

Page 16: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.16

Time and Quality RelationshipFor n<n*, longer completion and lower solution quality.

For n>n*, a better solution quality is achieved in exchange for longer completion time.

Page 17: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.17

Modifying a Little Model

What if deciding and doing are not linear & hyperbolic respectively?

Can modify the form.

For example, imagine that pairwise interactions are important in decision.

Deciding might be quadratic function of n.

Page 18: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.18

Nonlinearity in Deciding & Doing

Pairwise communications

More generally,

Likewise,

Page 19: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.19

A Power-Law Model

For , monotonically decreases with increased

Complexity of decision making increases, the efficient team size decreases.

Likewise, monotonically increases withless shirking or more synergy reduces the efficient team size.

Page 20: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.20

Power-Law Solution

n:

T:

Dimensionless form:

Page 21: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.21

Reusing Little Models

Consider problem of sizing breakouts in a conference (IlliGAL 2005021).

Have a big meeting of size m.

Have k = m/n breakout groups of size n.

Incur time of discussion Td per member in breakout groups and Tr reporting per team.

T = Td n + Tr k = Td n + Tr m/n

Page 22: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.22

Same Form as Deciding/DoingSubstitute T2 = Tr m and T2 = Td we are back to

Similar reasoning can be used to look at flat organizations.

d

r

T

mTn *

Page 23: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.23

Generalizing Little Models

Elementary optimization problems (EOPs).

Elementary turning points (ETPs).

Some solvable classes.

Helps to know what to look for.

Page 24: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.24

Elementary Optimization Problems

Elementary optimization problems (EOPs). A function of one variable.

The sum of a monotonically increasing function and a monotonically decreasing function.

Twice differentiable.

Page 25: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.25

Extensions of EOPMultiplicative form

Explicit EOP and implicit EOP

Single optimum simple EOP (sEOP)

Clearly nothing special about deciding, doing, time, quality, etc.

Have done transaction costs & span of control.

Page 26: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.26

Some examples of explicit sEOPs

Power law:

Exponential:

Logarithmic:

Mixed forms occasionally have explicit solution.

Page 27: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.27

Exponential-Exponential

Page 28: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.28

Power Law of Invertible Function

Page 29: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.29

Similar Argument for Turning Points

Have increasing and decreasing function.

Interested when they are equal.

f(n) = g(n).

Need explicit solution for n for greatest utility in inspection.

Page 30: MTV 13: Squeezing Little Models

Creative Modeling for Tech Visionaries © 2007 David E. Goldberg. All rights reserved.30

Bottom LineCan squeeze a lot out of little models.

Stretch them (visualize effectively and use auxiliary models).

Modify them (add complexity, accuracy).

Reuse them (in other domains).

Generalize them in EOPs and ETPs.

Seeking explicit models that yield qualitative & quantitative insight cheaply.

Can we extend them further & integrate multiple models for complex domains?