Attractor dynamics explain post-error adjustments

Talk at the NeuroComp 2018 conference in Paris.

Transcript

1. Attractor dynamics explains post-error adjustments NeuroComp @ Paris 2018 Kevin Berlemont Laboratoire de Physique Statistique, Ecole Normale Sup´ erieure, CNRS UMR 8550, PSL University, Paris, 75005, France.
2. PhD Advisor : Jean-Pierre Nadal (LPS - ENS, CAMS - EHESS) Fellowship : CDSN Ecole Normale Sup´ erieure Paris-Saclay 2/14
3. Introduction Perceptual Decision-Making Visual Discrimination 3/14
4. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories 3/14
5. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Cat or Dog ? 3/14
6. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Can be ambiguous Cat or Dog ? 3/14
7. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Can be ambiguous Cat or Dog ? 3/14
8. Random Dot Motion 4/14
9. Random Dot Motion Moving along ← or → ? J. D. Roitman and M. N. Shadlen (2002). “Response of neurons in the lateral intraparietal area during a combined visual discrimination reaction time task.”. In: Journal of Neuroscience 22.21, pp. 9475–9489. arXiv: NIHMS150003. 5/14
10. Random Dot Motion Moving along ← or → ? 5/14
11. Schematic process of decision-making Visual Stimulus Coding Decision-Making 6/14
12. Schematic process of decision-making Visual Stimulus Coding Decision-Making 6/14
13. Attractor Model R L ICD(t) = −ICD,max exp(−t/τCD) dSi dt = − Si τs + (1 − Si ) γHi Hi = axi − b 1 − exp [−d (axi − b)] C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
14. Attractor Model R L IL IR ICD(t) = −ICD,max exp(−t/τCD) xL = I0 + IL + Inoise,L C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
15. Attractor Model R L IL IR ICD(t) = −ICD,max exp(−t/τCD) xL = I0 + IL + Inoise,L + JLLSL C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
16. Attractor Model R L IL IR ICD(t) = −ICD,max exp(−t/τCD) xL = I0 + IL + Inoise,L + JLLSL − JLRSR C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
17. Attractor Model R L IL IR Decision ICD(t) = −ICD,max exp(−t/τCD) 9000 10000 11000 12000 13000 14000 0.1 0.2 0.3 0.4 Time (ms) Threshold C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
18. Attractor Model R L IL IR Decision Corollary discharge ICD(t) = −ICD,max exp(−t/τCD) C. C. Lo and X. J. Wang (2006). “Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks”. In: Nature Neuroscience 9.7, pp. 956–963. K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 7/14
19. Energy Landscape 8/14
20. Energy Landscape 8/14
21. Energy Landscape 8/14
22. Energy Landscape 8/14
23. Energy Landscape 8/14
24. Energy Landscape 8/14
25. Post-error adjustments Two choices tasks C. Danielmeier and M. Ullsperger (2011). “Post-error adjustments”. In: Frontiers in Psychology 2.SEP, pp. 1–10. I. Jentzsch and C. Dudschig (2009). “Why do we slow down after an error? Mechanisms underlying the effects of posterror slowing”. In: Quarterly Journal of Experimental Psychology 62.2, pp. 209–218. 9/14
26. Post-error adjustments Two choices tasks No feedback on the response C. Danielmeier and M. Ullsperger (2011). “Post-error adjustments”. In: Frontiers in Psychology 2.SEP, pp. 1–10. I. Jentzsch and C. Dudschig (2009). “Why do we slow down after an error? Mechanisms underlying the effects of posterror slowing”. In: Quarterly Journal of Experimental Psychology 62.2, pp. 209–218. 9/14
27. Post-error adjustments Two choices tasks No feedback on the response Higher response time after an error C. Danielmeier and M. Ullsperger (2011). “Post-error adjustments”. In: Frontiers in Psychology 2.SEP, pp. 1–10. I. Jentzsch and C. Dudschig (2009). “Why do we slow down after an error? Mechanisms underlying the effects of posterror slowing”. In: Quarterly Journal of Experimental Psychology 62.2, pp. 209–218. 9/14
28. Post-error adjustments Two choices tasks No feedback on the response Higher response time after an error Accuracy improved after an error C. Danielmeier and M. Ullsperger (2011). “Post-error adjustments”. In: Frontiers in Psychology 2.SEP, pp. 1–10. I. Jentzsch and C. Dudschig (2009). “Why do we slow down after an error? Mechanisms underlying the effects of posterror slowing”. In: Quarterly Journal of Experimental Psychology 62.2, pp. 209–218. 9/14
29. Post-error adjustments C. Danielmeier and M. Ullsperger (2011). “Post-error adjustments”. In: Frontiers in Psychology 2.SEP, pp. 1–10. 9/14
30. Post-error slowing in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 Coherence level (%) PES (ms) Corollary discharge (nA) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 10/14
31. Post-error slowing in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 Coherence level (%) PES (ms) Corollary discharge (nA) Coherence level (%) PES (ms) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 10/14
32. Post-error slowing in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 Coherence level (%) Coherence level (%) PES (ms) PES (ms) Corollary discharge (nA) Coherence level (%) PES (ms) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 10/14
33. Post-error improvement in accuracy in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 PIA Corollary discharge (nA) Coherence level (%) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 11/14
34. Post-error improvement in accuracy in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 PIA PIA Corollary discharge (nA) Coherence level (%) Coherence level (%) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 11/14
35. Post-error improvement in accuracy in the model Response-Stimulus interval: 500ms. 0.02 0.03 0.035 0.042 0.047 0.054 0.061 0.071 0.08 0.09 0.1 PIA PIA PIA Corollary discharge (nA) Coherence level (%) Coherence level (%) Coherence level (%) K. Berlemont and J.-P. Nadal (2018). “Perceptual decision making: Biases in post-error reaction times explained by attractor network dynamics”. In: arXiv: 1803.00795. 11/14
36. Explanation by the dynamics -2.5 -2.0 -1.5 -1.0 -0.5 -1.5 -1.0 -2.0 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -0.5 Coherence level : 10 Coherence level : -10 Coherence level (%) PES (ms) PIA Coherence level (%) 12/14
37. Explanation by the dynamics Neutral attractor -2.5 -2.0 -1.5 -1.0 -0.5 -1.5 Decision L Decision R -1.0 -2.0 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 10 Approximate decision line Coherence level : -10 12/14
38. Explanation by the dynamics Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor -2.5 -2.0 -1.5 -1.0 -0.5 -1.5 Decision L Decision R -1.0 -2.0 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 10 Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation Coherence level : -10 12/14
39. Explanation by the dynamics Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor -2.5 -2.0 -1.5 -1.0 -0.5 -1.5 Decision L Decision R -1.0 -2.0 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Invariant manifold for stimulus L Invariant manifold for stimulus R Attraction pool for the current trial Coherence level : 10 Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation Coherence level : -10 12/14
40. Explanation by the dynamics Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor -2.5 -2.0 -1.5 -1.0 -0.5 -1.5 Decision L Decision R -1.0 -2.0 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Typical post-correct decision Typical post-error decision Invariant manifold for stimulus L Invariant manifold for stimulus R Attraction pool for the current trial Coherence level : 10 Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation Coherence level : -10 12/14
41. Explanation by the dynamics -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 -2.5 -2.0 -1.5 -1.0 -2.5 -2.0 -1.5 -1.0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 Coherence level : 20 Coherence level : -20 PIA Coherence level (%) Coherence level (%) PES (ms) 13/14
42. Explanation by the dynamics -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 -2.5 -2.0 -1.5 -1.0 Decision L Decision R -2.5 -2.0 -1.5 -1.0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 20 Coherence level : -20 Neutral attractor Approximate decision line 13/14
43. Explanation by the dynamics -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 -2.5 -2.0 -1.5 -1.0 Decision L Decision R -2.5 -2.0 -1.5 -1.0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 20 Coherence level : -20 Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation 13/14
44. Explanation by the dynamics -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 -2.5 -2.0 -1.5 -1.0 Decision L Decision R -2.5 -2.0 -1.5 -1.0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 20 Coherence level : -20 Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor Invariant manifold for stimulus L Invariant manifold for stimulus R Attraction pool for the current trial Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation 13/14
45. Explanation by the dynamics -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 -2.5 -2.0 -1.5 -1.0 Decision L Decision R -2.5 -2.0 -1.5 -1.0 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 Decision L Decision R Coherence level : 20 Coherence level : -20 Mean Post-correct relaxation Mean Post-error relaxation Neutral attractor Typical post-correct decision Typical post-error decision Invariant manifold for stimulus L Invariant manifold for stimulus R Attraction pool for the current trial Approximate decision line Ending point of post-error's relaxation Ending point of post-correct's relaxation 13/14
46. Conclusion Biophysically inspired attractor network model 14/14
47. Conclusion Biophysically inspired attractor network model Decision tasks can be traited as continous sessions 14/14
48. Conclusion Biophysically inspired attractor network model Decision tasks can be traited as continous sessions Post-error adjustments are observed and explained 14/14
49. Conclusion Biophysically inspired attractor network model Decision tasks can be traited as continous sessions Post-error adjustments are observed and explained Suggest to test experimentally some effects 14/14
50. Temporal variations PES 15/14
51. Temporal variations PIA 16/14
52. Quantiles Variations 17/14
53. Phase Diagram Tau 500 ms 18/14
54. Other regime 19/14
55. Same noise 20/14
56. Synaptics distributions 21/14