Coercion through Optimization: A Classification of Optimization Techniques

1
Coercion through Optimization A Classification
of Optimization Techniques
Modeling and Simulation Technology Research
Initiative Computer Science Department University
of Virginia

• Sarah Waziruddin
• David Brogan
• Paul F. Reynolds, Jr.

2
Modeling and Simulation Technology Research
Initiative

• Simulation Design and Transformation for Reuse
• COERCE
• Coercibility the practices and methods for
  capturing designer knowledge in software
  (Carnahan et al.)
• Coercion a user-guided, semi-automated software
  transformation process (Waziruddin et al.)
• Composability
• Reusing components possible with acceptable
  amounts of revision, to meet new requirements
  (Bartholet et al.)

3
Simulation Transformation

• Very desirable to meet new requirements
• Code modification
• Where to make changes?
• What to add, remove, or adjust?

Manual process difficult and time consuming

• Is there room for automation?

4
Coercion

• Increase efficiency of the code modification
  process
• Language tools
• Automatic visualization
• Sensitivity Analysis
• Optimization

Seeking insight
5
Optimization

• Traditionally used to find best parameter values
• Generates additional insight
• Identify brittle systems
• Explore novel circumstances
• Detect correlations
• Discover constraints
• Bound search space

6
P The Coercion Process
Btarget
S?
S?
.
.
7
P The Coercion Process
8
Paths to Coercion
Sn
S0
Sn
S0
Sn
S0
.
.
.
.
9
Dangerous Divergence
Sn
S0
10
Distance Function
S0
Btarget
Si Ii
Distance Function D(Si, Ii)
11
The Convergence of Coercion

• How can we guarantee coercion will
  terminate?D(Sn, In)0
• We benefit from our successes
• Better satisfy Btarget with good transformations
• We learn from our mistakes
• Gain insight from bad transformations
• Coercion invariant guarantees convergence
• D(Si, Ii) lt D(Si-1, Ii-1)
• Progressively reduce distance to Btarget

12
Focus on User Interaction with Optimization Tools

• Insight dependence and generation

Random Code Generation
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Code Modification
13
Running Example
14
Simulated Annealing
15
Simulated Annealing
16
Genetic Algorithms
17
Genetic Algorithms
18
Gradient-Based Search
19
Gradient-Based Search
ExploreUnknown
Simulated Annealing
Genetic Algorithms
Gradient-Based Search
Exploit Insight
20
Response Surface Methodology
21
Response Surface Methodology
ExploreUnknown
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Exploit Insight
22
Optimization Techniques
Explore Unknown
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Exploit Insight
23
Identify/classify best practices for optimization
in coercion

• Set-up time
• Computation time
• Technique preemption

24
Set-up Time
More Set-up Time
Explore Unknown
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Exploit Insight
25
Computation Time
More Computation Time
Explore Unknown
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Exploit Insight
26
Revisit Criterion Technique Preemption
More Preemption
Explore Unknown
Simulated Annealing
Genetic Algorithms
Response Surface Methodology
Gradient-Based Search
Exploit Insight
27
Conclusion

• Provided formalism for insight in coercion
• Evaluated optimization techniques for
  contributions to insight
• Provided criteria for optimization techniques in
  coercion

28
Acknowledgements

• Defense Modeling and Simulation Office
• National Science Foundation (ITR 0426971)

Modeling and Simulation Technology Research
Initiative