O slideshow foi denunciado.
Utilizamos seu perfil e dados de atividades no LinkedIn para personalizar e exibir anúncios mais relevantes. Altere suas preferências de anúncios quando desejar.
Making decisions for complex, dynamic
problems with imperfect knowledge
The application of control systems engineering
to ...
A group effort
2
@ehekler
Outline
• Epistemological target
• Control Systems Engineering
• Our work
– Encapsulate previous knowledge
– Define dynami...
From generally useful
to useful for me
Epistemological
target
4
@ehekler
Embracing (plausibly) meaningful variability
@ehekler
“In General”
~50%
“Personalization/
Matchmaking”
~35%
Idiosyncratic/...
Specific Solutions
for Specific Problems
Design &
Engineering
“On Average”
Science
“On Average” Evidence
for General Probl...
Specific Solutions
for Specific Problems
Design &
Engineering
“On Average”
Science
“On Average” Evidence
for General Probl...
Specific Solutions
for Specific Problems
Design &
Engineering
“On Average”
Science
“On Average” Evidence
for General Probl...
Specific Solutions
for Specific Problems
Design &
Engineering
“On Average”
Science
“On Average” Evidence
for General Probl...
Making decisions in
complex, dynamic
systems with imperfect
knowledge
Control Systems
Engineering
10
@ehekler
Control Systems Engineering
NSF IIS-1449751: EAGER: Defining a Dynamical Behavioral
Model to Support a Just in Time Adapti...
Describe & predict: System identification
-100
100
300
500
700
900
1100
1300
1500
0
2000
4000
6000
8000
10000
12000
14000
...
Martin, Rivera, & Hekler Am. Control Conference (2015)
Control: Model-predictive control
@ehekler
13
Continuous improvement: Adaptive control
@ehekler Flickr - Dave Gray
14
Systematically
managing and
mitigating imperfect
knowledge to support
dynamic evidence-
based decisions
Our work
15
@ehekl...
Encapsulate previous knowledge (theory)
@ehekler
16
Dynamical model of Social Cognitive Theory
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et a...
One inventory
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler
18
Differential equations (first order shown)
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et a...
Simulation: Low vs. high self-efficacy
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2...
Simulation: Habituation
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler
21
Secondary data analysis: Validation
Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014...
Define dynamic decision(s) of intervention
@ehekler
23
Daily “ambitious but doable” step goals
Hekler (PI), Rivera (Co-PI), NSF IIS-1449751
-15
-10
-5
0
5
10
15
20
0
2000
4000
6...
Intervention decisions
Martin, Rivera, & Hekler, 2015; 2016, American Control Conference
25
@ehekler
Simulation
Martin, Rivera, & Hekler, 2016, American Control Conference
26
@ehekler
Devise a system
identification
experiment
@ehekler
Need: replicable
(for estimation & validation),
random/pseudo-random,
e...
Experimental design: System identification
28
@ehekler
Multisine “pseudo-random” signals
Martin, Rivera, & Hekler, 2015, American Control Conference
29
@ehekler
Pilot study: “Just Walk”
Fitbit Zip
30
@ehekler
“Black box” modeling to develop descriptive
models & examine individual differences
@ehekler
31
Participants
• 22 inactive, overweight Android users
– BMI 33.7 ± 6.7
– 47 ± 6.2 years
– 87% women
Living anywhere in the...
Preliminary results: Average effects
6,827 (SE = 647) Average median steps in the last cycle
45% (SD = 36) Average increas...
Multiple-Input Single-Output
AutoRegressive with eXternal inputs model (ARX model)
34
@ehekler
Preliminary ARX parametric estimation
35
@ehekler
Preliminary dynamical modeling results
0
6000
External cues (Goals)
0
200
400
Outcome Expectancy for Reinforcement (Availa...
Model fit criterion
37
@ehekler
Three Example Individualized Computational Models via Black-Box
System ID: Goals-Expected Points-Granted Points model; B: ...
Semi-physical modeling: Examine mechanistic
dynamical models idiographically
@ehekler
39
Almost done… Stay tuned! 
@ehekler
40
Devise model predictive controller
@ehekler
41
Idiographic trajectory model predictions
Hekler, et al. 2013 Health Education and Behavior@ehekler
42
Martin, Rivera, & Hekler Am. Control Conference (2015; 2016)
Model-predictive controller
@ehekler
43
“Closing the loop”
Open Loop MPC 
44
@ehekler
Simulation: MPC robustness
Martin, Rivera, & Hekler (2016)
45
@ehekler
Summary
46
@ehekler
Specific Solutions
for Specific Problems
Design &
Engineering
“On Average”
Science
“On Average” Evidence
for General Probl...
From “in general” to “for me”
Decisions for complex, dynamic problems
Manage & mitigate imperfect knowledge
@ehekler
48
Feedback and questions welcome!
Dr. Eric Hekler, Arizona State University
ehekler@asu.edu, @ehekler 49
Próximos SlideShares
Carregando em…5
×

Making decisions for complex, dynamic problems with imperfect knowledge: The application of control systems engineering to a behavioral intervention

563 visualizações

Publicada em

The purpose of this talk is to introduce a behavioral science audience to the logic of control systems engineering and how it could be used to create far more personalized, precise, and perpetually adapting behavioral interventions.

Publicada em: Ciências
  • Seja o primeiro a comentar

Making decisions for complex, dynamic problems with imperfect knowledge: The application of control systems engineering to a behavioral intervention

  1. 1. Making decisions for complex, dynamic problems with imperfect knowledge The application of control systems engineering to a behavioral intervention @ehekler Eric Hekler, PhD Arizona State University August 18, 2016 Flickr -Pat Castaldo 1
  2. 2. A group effort 2 @ehekler
  3. 3. Outline • Epistemological target • Control Systems Engineering • Our work – Encapsulate previous knowledge – Define dynamic decisions of an intervention – Devise a system ID experiment – Examine individual differences (ARX) – Examine mechanistic model (semi-physical modeling) – Devise model-predictive controller @ehekler 3
  4. 4. From generally useful to useful for me Epistemological target 4 @ehekler
  5. 5. Embracing (plausibly) meaningful variability @ehekler “In General” ~50% “Personalization/ Matchmaking” ~35% Idiosyncratic/ Subjective ~15% Hekler, et al. 2016, Agile Science, Translational Behavioral Medicine 5
  6. 6. Specific Solutions for Specific Problems Design & Engineering “On Average” Science “On Average” Evidence for General Problems Key Traditional pathway Emerging pathway Product Process Professional-led 6 @ehekler
  7. 7. Specific Solutions for Specific Problems Design & Engineering “On Average” Science “On Average” Evidence for General Problems Key Traditional pathway Emerging pathway Product Process Precise Evidence for Specific Problems Personalization Algorithm Science Professional-led Process Individualization Science 7 @ehekler
  8. 8. Specific Solutions for Specific Problems Design & Engineering “On Average” Science “On Average” Evidence for General Problems Key Traditional pathway Emerging pathway Product Process Precise Evidence for Specific Problems Personalization Algorithm Science Professional-led Process Individualization Science Citizen/Patient-led 8 @ehekler
  9. 9. Specific Solutions for Specific Problems Design & Engineering “On Average” Science “On Average” Evidence for General Problems Key Traditional pathway Emerging pathway Product Process Precise Evidence for Specific Problems Personalization Algorithm Science Professional-led Process Individualization Science 9 @ehekler
  10. 10. Making decisions in complex, dynamic systems with imperfect knowledge Control Systems Engineering 10 @ehekler
  11. 11. Control Systems Engineering NSF IIS-1449751: EAGER: Defining a Dynamical Behavioral Model to Support a Just in Time Adaptive Intervention, PIs, Hekler & Rivera @ehekler 11
  12. 12. Describe & predict: System identification -100 100 300 500 700 900 1100 1300 1500 0 2000 4000 6000 8000 10000 12000 14000 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 Points Stepsperday Days Points Provided (100, 300, 500) Fictionalized actual steps per day Daily step goal ((Baseline Median) to (Baseline Median+100% Baseline Median)) NSF IIS-1449751: Defining a Dynamical Behavioral Model to Support a Just in Time Adaptive Intervention, PIs, Hekler & Rivera@ehekler 12
  13. 13. Martin, Rivera, & Hekler Am. Control Conference (2015) Control: Model-predictive control @ehekler 13
  14. 14. Continuous improvement: Adaptive control @ehekler Flickr - Dave Gray 14
  15. 15. Systematically managing and mitigating imperfect knowledge to support dynamic evidence- based decisions Our work 15 @ehekler
  16. 16. Encapsulate previous knowledge (theory) @ehekler 16
  17. 17. Dynamical model of Social Cognitive Theory Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler 17
  18. 18. One inventory Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler 18
  19. 19. Differential equations (first order shown) Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler 19
  20. 20. Simulation: Low vs. high self-efficacy Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler Low Self-Efficacy High Self-Efficacy 20
  21. 21. Simulation: Habituation Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler 21
  22. 22. Secondary data analysis: Validation Riley, Martin, Rivera, Hekler, et al. 2016; Martin, Riley, Rivera, Hekler, et al. 2014@ehekler 22
  23. 23. Define dynamic decision(s) of intervention @ehekler 23
  24. 24. Daily “ambitious but doable” step goals Hekler (PI), Rivera (Co-PI), NSF IIS-1449751 -15 -10 -5 0 5 10 15 20 0 2000 4000 6000 8000 10000 12000 14000 AveChangeSelfEffficacy ActualDailySteps Recommended Goal Actual Steps Δ Self-Efficacy @ehekler 24
  25. 25. Intervention decisions Martin, Rivera, & Hekler, 2015; 2016, American Control Conference 25 @ehekler
  26. 26. Simulation Martin, Rivera, & Hekler, 2016, American Control Conference 26 @ehekler
  27. 27. Devise a system identification experiment @ehekler Need: replicable (for estimation & validation), random/pseudo-random, excitation over time 27
  28. 28. Experimental design: System identification 28 @ehekler
  29. 29. Multisine “pseudo-random” signals Martin, Rivera, & Hekler, 2015, American Control Conference 29 @ehekler
  30. 30. Pilot study: “Just Walk” Fitbit Zip 30 @ehekler
  31. 31. “Black box” modeling to develop descriptive models & examine individual differences @ehekler 31
  32. 32. Participants • 22 inactive, overweight Android users – BMI 33.7 ± 6.7 – 47 ± 6.2 years – 87% women Living anywhere in the US Average Baseline Median Steps: 4972 steps/day (SE = 482) 32 @ehekler
  33. 33. Preliminary results: Average effects 6,827 (SE = 647) Average median steps in the last cycle 45% (SD = 36) Average increase in median steps/day from baseline to final cycle 69% (SD = 24) Average goals met >90% Adherence to daily self-report 33 @ehekler
  34. 34. Multiple-Input Single-Output AutoRegressive with eXternal inputs model (ARX model) 34 @ehekler
  35. 35. Preliminary ARX parametric estimation 35 @ehekler
  36. 36. Preliminary dynamical modeling results 0 6000 External cues (Goals) 0 200 400 Outcome Expectancy for Reinforcement (Available points) 0 200 400 Reinforcement (Granted Points) 0 5 Predicted Busyness 0 5 Predicted Stress 0 5 Predicted Typical 0 1 Weekday (0) - Weekend (1) Time (days) 0 10 20 30 40 50 60 70 80 90 100 Steps 2000 4000 6000 8000 10000 12000 Baseline Estimation Validation Goals (98 =u8 ) Predicted Steps (28.66% fit) Actual Steps Figure 4. System ID “open-loop” experiment w/ predicted & actual steps from one participant from Just Walk 36 @ehekler
  37. 37. Model fit criterion 37 @ehekler
  38. 38. Three Example Individualized Computational Models via Black-Box System ID: Goals-Expected Points-Granted Points model; B: Predicted Busyness; S: Predicted Stress; T: Predicted Typical; W: Weekday-Weekend Modeling differences 38 @ehekler
  39. 39. Semi-physical modeling: Examine mechanistic dynamical models idiographically @ehekler 39
  40. 40. Almost done… Stay tuned!  @ehekler 40
  41. 41. Devise model predictive controller @ehekler 41
  42. 42. Idiographic trajectory model predictions Hekler, et al. 2013 Health Education and Behavior@ehekler 42
  43. 43. Martin, Rivera, & Hekler Am. Control Conference (2015; 2016) Model-predictive controller @ehekler 43
  44. 44. “Closing the loop” Open Loop MPC  44 @ehekler
  45. 45. Simulation: MPC robustness Martin, Rivera, & Hekler (2016) 45 @ehekler
  46. 46. Summary 46 @ehekler
  47. 47. Specific Solutions for Specific Problems Design & Engineering “On Average” Science “On Average” Evidence for General Problems Key Traditional pathway Emerging pathway Product Process Precise Evidence for Specific Problems Personalization Algorithm Science Professional-led Process Individualization Science 47 @ehekler
  48. 48. From “in general” to “for me” Decisions for complex, dynamic problems Manage & mitigate imperfect knowledge @ehekler 48
  49. 49. Feedback and questions welcome! Dr. Eric Hekler, Arizona State University ehekler@asu.edu, @ehekler 49

×