Download presentation
Presentation is loading. Please wait.
Published byJarred Jenney Modified over 10 years ago
1
Remembering to decide: discrimination of temporally separated stimuli (selecting the best apple) Paul Miller Brandeis University
2
Parametric Working Memory and Sequential Discrimination Experiments by group of R. Romo et al., UNAM Nature 399:470 (1999), Cereb. Cort. 13:1196 (2003)
3
Choose f1 > f2 f2f1
4
or f2 > f1 f1f2
5
Rastergram: f1(Hz) 10 14 18 22 26 30 34 basedelay Trial-averaged firing rate Firing rate (Hz) 0 30 Time (sec)0.5 3.5 (from Miller et al. Cerebral Cortex 2003) Tuning curve of memory activity Firing rate (Hz) Stimulus, f1 (Hz) 5 18 1034 Romo et al. Nature 1999
6
A continuous attractor acts as an integrator Time Input Memory activity
7
... but integration yields magnitude x time Time Input Memory activity
8
Problem: How can a network compare an incoming stimulus with an earlier one in memory? Especially as discrimination ≡ subtraction whereas integration ≡ addition Sequential Discrimination Integral feedback control: memory neurons (M) inhibit their inputs (D). Solution: - + ∫ r D dt Input
9
rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 1 low cue1delaycue2
10
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 1 low cue1delaycue2
11
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 2 higher cue1delaycue2
12
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 2 lower Threshold not reached cue1delaycue2
13
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 1 high cue1delaycue2
14
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 2 lower Threshold not reached cue1delaycue2
15
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue1delaycue2 cue 2 higher
16
A continuous attractor for memory
18
Feedback too high Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
19
Feedback too high Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
20
Feedback too high Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
21
Feedback too low Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
22
Feedback too low Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
23
Feedback too low Gradient is proportional to difference between current needed to produce a firing rate and the feedback current generated by that firing rate.
24
Continuous or discrete memory? Note psychophysics: for most continuous quantities, we can only remember (even recognize?) them in discrete categories Except when quantity is encoded across different neurons (eg vision, pitch)
25
Simulation results Look at Discriminating neuron Memory = Discrete Integrator
26
Activity of model discriminating neuron. basedelaycomparison
27
basedelaycomparison Activity of model discriminating neuron.
28
Trial-averaged firing rate through time of model discriminating neuron for different pairs of stimuli f1 = 34Hz f1 = 10Hz f2>f1 f2<f1 Base, f1 Delay Comparison, f2 Time (sec) 0 0.5 3.5 4 0 100 Firing rate (Hz) f1 = 22Hz Miller and Wang, PNAS 2006
29
Base tuning Comparison tuning Delay tuning f2>f1 f2<f1
30
Trial-averaged firing rate through time from experimental data of Romo (prefrontal cortex) Base, f1 Delay Comparison, f2 Time (sec) 0 0.5 3.5 4 0 35 Firing rate (Hz) f2>f1 f2<f1 f1=12Hz f1=20Hz f1=28Hz
31
PFC cell from Romo's data: Initial tuning +ve to f1 : final tuning to +f2-f1 Base, f1 Delay Comparison, f2 Time (sec) 0 0.5 3.5 4 0 60 Firing rate (Hz) f2>f1 f2<f1 f1=10Hz f1=22Hz f1=34Hz
32
PFC cell from Romo's data Initial tuning -ve to f1 : final tuning to +f1-f2 Base, f1 Delay Comparison, f2 Time (sec) 0 0.5 3.5 4 0 35 Firing rate (Hz) f2<f1 f2>f1 f1=10Hz f1=22Hz f1=28Hz
33
Decision-making as a competition between pools
34
f1=22Hz Probability of choosing f2>f1 from simulations
35
f1=14Hz f1=22Hz Probability of choosing f2>f1 from simulations
36
f1=14Hz f1=22Hz f1=30Hz Probability of choosing f2>f1 from simulations Miller, in preparation
37
Probability of choosing f2>f1 from experiment f1 = 20Hzf1 = 30Hz f2
38
Probability of choosing f2>f1 from experiment = fix f2 (20Hz), vary f1 = fix f1 (20Hz), vary f2
39
Probability of choosing f2>f1 from experiment Hernandez et al, 1997 = fix f2 (20Hz), vary f1 = fix f1 (20Hz), vary f2 = fix f2 (30Hz), vary f1 = fix f1 (30Hz), vary f2
40
fixed f1=22Hzfixed f1=30Hz Probability of choosing f2>f1 from simulations
41
fixed f1=22Hzfixed f1=30Hz Probability of choosing f2>f1 from simulations fixed f2=22Hzfixed f2=30Hz
42
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 1:low Is magnitude dissociated from duration of input?
43
Input rDrD rMrM I D =Input -W MD r M t t t t cue1delaycue2 cue1 delay cue2 cue 1:longer Is magnitude dissociated from duration of input?
44
Duration of initial stimulus:= 0.5s Is magnitude dissociated from duration of input? Simulation results
45
Duration of initial stimulus:= 0.5s = 0.25s Is magnitude dissociated from duration of input? Simulation results
46
Duration of initial stimulus:= 0.5s = 0.25s = 0.75s + Is magnitude dissociated from duration of input? Simulation results
47
From Luna et al., Nat Neurosci 2005 Is magnitude dissociated from duration of input? Experimental results
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.