RationalBoundsOfControl_ControlProcesses2019.pdf

Symposium on the Geometry of Control Representations
Control Processes 2019
Brown University

Transcript

1. Sebastian Musslick Princeton Neuroscience Institute Symposium on the Geometry of Control Representations Control Processes 2019 Brown University Slides available at: https://speakerdeck.com/musslick
2. Cognitive control – reconfigure information processing away from default (automatic) settings (Cohen et al., 1990; Botvinick & Cohen, 2015) read email follow talk
3. Cognitive control is associated with constraints in processing (Posner & Snyder, 1975; Shiffrin & Schneider, 1977)
4. read email follow talk Cognitive control is associated with constraints in processing (Posner & Snyder, 1975; Shiffrin & Schneider, 1977)
5. Bounds of cognitive control are… § A defining feature of cognitive control (Posner & Snyder, 1997; Shiffrin & Schneider, 1977) § Operationalized in the form of dual-task interference § A premise of general theories of cognition § ACT-R (Anderson, 1986; 2013) § EPIC (Meyer & Kieras, 1997) § SOAR (Laird, 2012) § Multi-Threaded Cognition (Salvucci & Taatgen, 2008, 2010) § Bounded Rationality (Simon, 1957) § An explanatory variable in recent models of control allocation § Opportunity Cost Model (Kurzban, Duckworth, Kable & Myers, 2013) § Expected Value of Control Theory (Shenhav, Botvinick & Cohen, 2013; Musslick, Shenhav, Botvinick & Cohen, 2015) § Value of Computation (Lieder & Griffiths, 2015; Lieder, Shenhav, Musslick & Griffiths, 2018)
6. Bounds of cognitive control are… § A defining feature of cognitive control (Posner & Snyder, 1997; Shiffrin & Schneider, 1977) § A premise of general theories of cognition § ACT-R (Anderson, 1986; 2013) § EPIC (Meyer & Kieras, 1997) § SOAR (Laird, 2012) § Multi-Threaded Cognition (Salvucci & Taatgen, 2008, 2010) § An explanatory variable in recent models of control allocation § Opportunity Cost Model (Kurzban, Duckworth, Kable & Myers, 2013) § Expected Value of Control Theory (Shenhav, Botvinick & Cohen, 2013; Musslick, Shenhav, Botvinick & Cohen, 2015) § Value of Computation (Lieder & Griffiths, 2015; Lieder, Shenhav, Musslick & Griffiths, 2018) Structural limitations? Metabolic constraints?
7. shared representation for learning efficiency separate representation For multitasking capacity versus control
10. Under which conditions can we multitask?
11. Name the color of the following stimulus and, at the same time, point to where it is… BROWN
12. BLUE
13. YELLOW
14. point left if the written word is RED point right if the written word is GREEN RED
15. GREEN
16. RED
17. RED
18. Name the color of the following stimulus and, at the same time: point left if the written word is RED point right if the written word is GREEN RED
19. GREEN
20. RED
21. Accuracy Results Color Naming + Location Pointing Word Mapping Color Naming + Word Mapping Accuracy by Task Task Percent Correct (%) 0 20 40 60 80 100 74 87 5.1 (first part) (second part) (third part) Anne Mennen Abigail Novick
22. (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) verbal manual response color word location stimulus internal (hidden) representation
23. verbal manual response color word location stimulus internal (hidden) representation (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016)
24. color word location verbal manual stimulus internal (hidden) representation response (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
25. stimulus internal (hidden) representation response color word location verbal manual (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
26. stimulus internal (hidden) representation response color word location verbal manual multitasking is possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
27. stimulus internal (hidden) representation response color word location verbal manual multitasking is not possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
28. stimulus internal (hidden) representation response color word location verbal manual multitasking is not possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016)
29. stimulus internal (hidden) representation response color word location verbal manual (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control purpose of cognitive control is to limit interference
30. stimulus internal (hidden) representation response color word location verbal manual multitasking is not possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
31. stimulus internal (hidden) representation response color word location verbal manual multitasking is possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016) control
32. stimulus internal (hidden) representation response color word location verbal manual multitasking is possible (Cohen et al., 1990 ; Feng et al., 2014; Musslick et al., 2016)
33. stimulus internal (hidden) representation response color word location verbal manual multiple-resource hypothesis (Allport, 1972; Allport, 1980; Meyer & Kieras, 1997; Navon & Gopher, 1979; Wickens, 1984; Salvucci & Taatgen, 2008)
34. Capacity for control-dependent processing… … decreases with amount of shared representation and is virtually invariant to network size, as is evident from § Computational simulations (Feng et al., 2014; Musslick et al., 2016, 2017) § Graph-theoretic analyses (Musslick et al., 2016; Alon et al., 2017) Multitasking Capacity Amount of Shared Representation Network Size
35. § Computational simulations (Feng et al., 2014; Musslick et al., 2016, 2017) § Graph-theoretic analyses (Musslick et al., 2016; Alon et al., 2017) § Estimation techniques derived from statistical mechanics (Petri et al., under review) Capacity for control-dependent processing… … decreases with amount of shared representation and is virtually invariant to network size, as is evident from
36. Capacity for control-dependent processing… … decreases with network depth (Alon et al., 2017) Network Depth 60 40 20 0 Amount of Shared Representation Multitasking Capacity Network Size = (# Units Within a Layer) Daniel Reichman
37. color word verbal manual
38. color word verbal manual (Musslick & Cohen, in press)
39. color word verbal manual (Musslick & Cohen, in press)
40. (Musslick & Cohen, in press)
42. Why Shared Representations? § Multi-Task Learning: Improved Learning Efficiency & Generalization Performance (e.g. Baxter, 1995; Caruana, 1997; Collobert & Weston, 2008; Bengio et al., 2013) shared intermediate representation auxilliary task 1 output auxilliary task 2 output primary task output …
43. color word verbal manual
44. word verbal manual 2 x training signal color
45. color word verbal manual
46. 1 task executable at a time color word verbal manual
47. color word verbal manual
48. color word verbal manual
49. word verbal manual 1 x training signal color
50. color word verbal manual
51. 2 tasks executable at a time color word verbal manual
52. ∝ multitasking capacity # of tasks sharing an input dimension Andrew Saxe color word verbal manual (time to learn)2 (Musslick et al., 2017)
53. § .. shallow linear networks (Saxe et al., 2017) § …in shallow non-linear networks with simple task environments (Musslick et al., 2017) § …in deep convolutional neural networks with complex task environments (Ravi et al., in prep) Tradeoff between learning efficiency and multitasking capability in 70 75 80 85 Iterations Required To Train 42 44 46 48 50 52 Multitasking Accuracy (%) 0 0.2 0.4 0.6 0.8 1 Initial Task Correlation 70 75 80 85 Iterations Required To Train 42 44 46 48 50 52 Multitasking Accuracy (%) 0 0.2 0.4 0.6 0.8 1 Initial Task Correlation Initial Task Correlation Sharing representations is optimal if § …the benefit of shared representations for learning is high, § …the cost associated with executing tasks sequentially is low, § …time horizon is finite (Ravi et al., in prep; Sagiv, Musslick, Niv & Cohen, 2018) Amount of Sharing
55. Similarity Between Task Representations as a Function ofTraining
56. Similarity Between Task Representations as a Function ofTraining
57. Similarity Between Task Representations as a Function ofTraining
58. Similarity Between Task Representations as a Function ofTraining
59. Similarity Between Task Representations as a Function ofTraining
60. Multitasking Training Study color word location verbal manual Abigail Novick
61. Multitasking Training Study color word location verbal manual Abigail Novick
62. Multitasking Training Study color word location verbal manual Abigail Novick word reading word pointing
63. Multitasking Training Study color word location verbal manual Abigail Novick word reading word pointing 1 2 3 4 1 2 3 4 session session
64. learning efficiency multitasking capacity shared representation separate representation controlled automatic
65. Jonathan Cohen Ted Willke Biswadip Dey Kayhan Ozcimder Andrew Saxe Abigail Novick Anne Mennen Penina Krieger Yotam Sagiv Sachin Ravi Daniel Reichman Giovanni Petri Thank you!