SFN2010-v6.pdf

Transcript

1. A top-down component of decision- related activity in monkey MT Ali Moeeny MD PhD Bruce Cumming MD PhD Laboratory of Sensory-motor Research National Eye Institute, NIH, Bethesda, MD 1
2. Sources of Decision Variation Reported Direction 2
3. Sources of Decision Variation MT Reported Direction 2
4. Sources of Decision Variation MT Reported Direction No incoming signal 2
5. Sources of Decision Variation MT Reported Direction No incoming signal 2
6. Sources of Decision Variation MT Reported Direction No incoming signal 2
7. Sources of Decision Variation MT Feedforward models Reported Direction No incoming signal 2
8. Sources of Decision Variation MT Feedforward models Reported Direction No incoming signal High rate Low rate Random noise 2
9. Sources of Decision Variation MT Feedforward models Reported Direction No incoming signal High rate Low rate Random noise 2
10. Sources of Decision Variation MT Reported Direction No incoming signal 3
11. Sources of Decision Variation MT Top-down models Reported Direction No incoming signal 3
12. Sources of Decision Variation Feature-based Attention MT Top-down models Reported Direction No incoming signal 3
13. Sources of Decision Variation Feature-based Attention MT Top-down models Reported Direction No incoming signal High rate Low rate 3
14. Hypothesis • If we can manipulate the way a subject perceives an ambiguous stimulus, that manipulation should change the firing rate of sensory neurons 4
15. Demo 5
16. Demo • Ambiguous • Stable • Binocular disparity can make it unambiguous -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- • Trick Start a trial with only a brief pulse of disparity 5
17. Design Fixation Disparity Rotation Time 500 ms 0 2000 ms 1000 ms %15 of trials Initial Disparity Rewarded by chance %85 of trials Rewarded if correct 6
18. Design Fixation Disparity Rotation Time 500 ms 0 2000 ms 1000 ms %15 of trials Initial Disparity Rewarded by chance %85 of trials Rewarded if correct 6
19. Design Fixation Disparity Rotation Time 500 ms 0 2000 ms 1000 ms %15 of trials Initial Disparity Rewarded by chance %85 of trials Rewarded if correct 6
20. Design Fixation Disparity Rotation Time 500 ms 0 2000 ms 1000 ms %15 of trials Initial Disparity Rewarded by chance %85 of trials Rewarded if correct 6
21. Design Fixation Disparity Rotation Time 500 ms 0 2000 ms 1000 ms %15 of trials Initial Disparity Rewarded by chance %85 of trials Rewarded if correct 6
22. Average Performance During Recording Disparity (degrees) - 0.035 0 + 0.035 Rightward in front Leftward in front %Leftward is in front %100 %50 %0 7
23. Average Performance During Recording Disparity (degrees) - 0.035 0 + 0.035 Rightward in front Leftward in front %Leftward is in front %100 %50 %0 -0.04 Initial disparity 7
24. Average Performance During Recording Disparity (degrees) - 0.035 0 + 0.035 Rightward in front Leftward in front %Leftward is in front %100 %50 %0 -0.04 Initial disparity Initial disparity +0.04 7
25. Example Neuron Time (ms) Firing Rate (spike/s) dae119 0 500 1000 1500 2000 120 60 0 8
26. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. 0 500 1000 1500 2000 120 60 0 8
27. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. 0 500 1000 1500 2000 120 60 0 8
28. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Null rotation direction 0 500 1000 1500 2000 120 60 0 8
29. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Null rotation direction 0 500 1000 1500 2000 120 60 0 8
30. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
31. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Feedforward model 8
32. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Feedforward model 8
33. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Feedforward model 8
34. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Feedforward model 8
35. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Feedforward model Delay 8
36. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
37. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Top-down model 8
38. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 Predictions from Top-down model 8
39. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
40. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
41. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
42. Example Neuron Time (ms) Firing Rate (spike/s) dae119 Preferred rotation dir. Preferred Initial disparity Null Initial disparity Null rotation direction 0 500 1000 1500 2000 120 60 0 8
43. Population Responses Time (ms) Firing Rate (spikes/s) 44 cells / 2 monkeys Preferred rotation dir. Preferred initial disparity Null initial disparity Null rotation direction 0 500 1000 1500 2000 50 25 0 9
44. Population Responses Time (ms) Firing Rate (spikes/s) 44 cells / 2 monkeys Preferred initial disparity Null initial disparity 0 500 1000 1500 2000 50 25 0 9
45. Population Responses Time (ms) Firing Rate (spikes/s) 44 cells / 2 monkeys Preferred initial disparity Null initial disparity 0 500 1000 1500 2000 50 25 0 Mean ROC = 0.67 larger than 0.5 (p<0.001) 9
46. Population Responses Time (ms) Firing Rate (spikes/s) 44 cells / 2 monkeys Preferred initial disparity Null initial disparity 0 500 1000 1500 2000 50 25 0 Mean ROC = 0.67 larger than 0.5 (p<0.001) Delay 9
47. Control Task Time 500 ms 0 2000 ms 1000 ms 10
48. Control Task Time 500 ms 0 2000 ms 1000 ms Fixation 10
49. Control Task Time 500 ms 0 2000 ms 1000 ms Fixation Rotation Disparity Inside RF 10
50. Control Task Time 500 ms 0 2000 ms 1000 ms Fixation Rotation Disparity Inside RF Rotation Disparity Outside RF Rewarded for correct choice 10
51. Demo 11
52. Control Task • To ensure that the monkey was attending to the stimulus outside the RF we only used data from the sessions that his threshold was close to those measured in training • In practice this required that the animal only do this task for several weeks consequently data was collected on separate neurons 12
53. Population Responses Time (ms) Firing Rate 20 cells / 1 monkey Preferred Initial disparity Null Initial disparity 0 500 1000 1500 2000 13
54. Population Responses Time (ms) Firing Rate 20 cells / 1 monkey Preferred Initial disparity Null Initial disparity 0 500 1000 1500 2000 Mean ROC = 0.55 smaller than main effect (p<0.001) No purely feedforward model is sufficient to explain the main effect 13
55. Conclusions • This directly demonstrates a top-down signal that contributes to the correlation between choices and firing rates in MT • This top-down signal is large enough to generate the reported choice probabilities of MT neurons 14
56. Thanks! Any Questions? 15