Talk: Rencontre des Jeunes Physiciens 2021

Transcript

1. Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions Rencontres des Jeunes Physicien·ne·s 2021 Kevin Berlemont Laboratoire de Physique de l’ENS Center for Neural Science - NYU
2. PhD Advisor: Jean-Pierre Nadal (LPENS, CAMS - EHESS) In Collaboration with: Jerˆ ome Sackur (LSCP - EHESS) and Jean-R´ emy Martin (CRCN - ULB) Thanks to: Laurent Bonasse-Gahot (CAMS - EHESS) Fellowship: CDSN Ecole Normale Sup´ erieure Paris-Saclay 2/13
3. Introduction Perceptual Decision-Making Visual Discrimination 3/13
4. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories 3/13
5. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Cat or Dog ? 3/13
6. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Can be ambiguous Cat or Dog ? 3/13
7. Introduction Perceptual Decision-Making Visual Discrimination Choice between categories Can be ambiguous Cat or Dog ? 3/13
8. Experimental setup 200 ms 100 ms 200 ms Decision Time K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 4/13
9. Confidence and behavior 5/13
10. Confidence and behavior 5/13
11. Attractor Model R L dSi dt = − Si τs + (1 − Si ) γf (Ii,tot) f (Ii,tot) = af (Ii,tot) − b 1 − exp [−d (af (Ii,tot) − b)] K.-F. Wong and X.-J. Wang (2006). “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions”. Journal of Neuroscience 26.4, pp. 1314–1328. 6/13
12. Attractor Model R L IL IR IL,tot = I0 + IL + Inoise,L K.-F. Wong and X.-J. Wang (2006). “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions”. Journal of Neuroscience 26.4, pp. 1314–1328. 6/13
13. Attractor Model R L IL IR IL,tot = I0 + IL + Inoise,L + JLLSL K.-F. Wong and X.-J. Wang (2006). “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions”. Journal of Neuroscience 26.4, pp. 1314–1328. 6/13
14. Attractor Model R L IL IR IL,tot = I0 + IL + Inoise,L + JLLSL − JLRSR K.-F. Wong and X.-J. Wang (2006). “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions”. Journal of Neuroscience 26.4, pp. 1314–1328. 6/13
15. Attractor Model R L IL IR Decision θ Time (ms) ri (Hz) K.-F. Wong and X.-J. Wang (2006). “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions”. Journal of Neuroscience 26.4, pp. 1314–1328. 6/13
16. Typical timecourse Stimulus 1 Stimulus 2 Decision 1 Decision 2 RSI 7/13
17. Typical timecourse Stimulus 1 Stimulus 2 Decision 1 Decision 2 RSI 7/13
18. Behavioral results K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 8/13
19. Confidence modelisation θ Time (ms) ri (Hz) 9/13
20. Confidence modelisation θ Time (ms) ri (Hz) ∆r at time of the decision 9/13
21. Confidence modelisation ∆r (Hz) - Simulations 10/13
22. Confidence modelisation ∆r (Hz) - Simulations f (∆r) 0 2 4 6 8 10 Confidence (Data) 10/13
23. Histogram matching Confidence Participant 1 Confidence Confidence Confidence Confidence Confidence Participant 2 Participant 3 Participant 6 K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 11/13
24. Histogram matching Participant 3 Participant 2 Participant 1 Participant 6 Participant 5 Participant 4 K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 11/13
25. Matching with data K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 12/13
26. Matching with data K. Berlemont et al. (2020). “Nonlinear neural network dynamics accounts for human confidence in a sequence of perceptual decisions”. Scientific reports 10.1, pp. 1–16. 12/13
27. Conclusion Attractor neural networks can be fitted to individual participants. 13/13
28. Conclusion Attractor neural networks can be fitted to individual participants. It relates accurately confidence, response times and accuracy. 13/13
29. Conclusion Attractor neural networks can be fitted to individual participants. It relates accurately confidence, response times and accuracy. Non-linearity explains various effects observed in decision-making experiments. 13/13