/
sigLatDistOverDev.m
176 lines (162 loc) · 6.29 KB
/
sigLatDistOverDev.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
clear all; close all
figure(1); clf
set(gcf, 'color', ones(1,3));
p.wmax = 1; % single synapse hard bound
p.m = 10; % desired number of synapses per neuron (wmax = Wmax/m)
p.n = 100; % n neurons
p.trainint = 10; % Time interval between inputs
p.nsteps = 100; % time-steps to simulate -- each time-step is 1 burst duration.
p.pn = .01; % probability of external stimulation of at least one neuron at any time
p.trainingInd = 1:10; % index of training neurons
p.beta = .115; % strength of feedforward inhibition
p.alpha = 30; % strength of neural adaptation
p.eta = .025; % learning rate parameter
p.epsilon = .2; % relative strength of heterosynaptic LTD
p.tau = 4; % time constant of adaptation
p.gamma= .01; % strength of recurrent inhibition
wmaxSplit = 2; % single synapse hard bound to induce splitting (increased to encourage fewer stronger synapses)
gammaSplit =.18; % increased strength of recurrent inhibition to induce splitting
Niter = [1 500 500];
gammas = sigmf(1:Niter(end),[1/200 500])*gammaSplit; % gradually increase gamma to gammaSplit
Wmax = p.wmax*p.m;
wmax = p.wmax;
gamma = p.gamma;
m = p.m;
k = length(p.trainingInd);
trainint = p.trainint;
nsteps = p.nsteps;
n = p.n;
pn = p.pn;
HowClamped = 10;
HowOn = 10;
%Psyl inputs
% training inputs
trainingNeurons{1}.nIDs = 1:k;
trainingNeurons{2}.nIDs = 1:k;
trainingNeurons{1}.tind = find(mod(1:nsteps,trainint)==1); %
trainingNeurons{2}.tind = find(mod(1:nsteps,trainint)==1); %
trainingNeurons{1}.candLat = 1:trainint;
trainingNeurons{2}.candLat = 1:trainint;
trainingNeurons{1}.thres = 4;
trainingNeurons{2}.thres = 4;
Input = -HowClamped*ones(k, nsteps); %clamp training neurons (effectively giving them higher threshold)
Input(:,mod(1:nsteps,trainint)==1) = HowOn; % rhythmic activation of training neurons
PsylInput = Input;
trainingNeuronsPsyl = trainingNeurons; clear trainingNeurons;
%Alternating Inputs
% training inputs
trainingNeurons{1}.nIDs = 1:k/2;
trainingNeurons{2}.nIDs = (k/2+1):k;
trainingNeurons{1}.tind = find(mod(1:nsteps,2*trainint)==1); %repmat([true(1,trainint) false(1,trainint)],1,nsteps/trainint/2);
trainingNeurons{2}.tind = find(mod(1:nsteps,2*trainint)==trainint+1); %repmat([false(1,trainint) true(1,trainint)],1,nsteps/trainint/2);
trainingNeurons{1}.candLat = 1:trainint;
trainingNeurons{2}.candLat = 1:trainint;
trainingNeurons{1}.thres = 2;
trainingNeurons{2}.thres = 2;
Input =-HowClamped*ones(k, nsteps); % clamp training neurons (effectively giving them higher threshold)
Input(trainingNeurons{1}.nIDs,mod(1:nsteps,2*trainint)==1) = HowOn; % alternating rhythmic activation of training neurons
Input(trainingNeurons{2}.nIDs,mod(1:nsteps,2*trainint)==trainint+1) = HowOn; % alternating rhythmic activation of training neurons
AltInput = Input;
trainingNeuronsAlt = trainingNeurons; clear trainingNeurons
%%
% for each seed
for seedi = 1:50
rng(seedi);
p.wmax = wmax;
p.gamma = gamma;
p.m = m;
w0 = 2*rand(p.n)*Wmax/p.n;
% subsong stage
eta = p.eta;
p.eta = 0;
nsteps = p.nsteps;
nstepsSubsong = 1000;
p.nsteps = nstepsSubsong;
w = w0;
p.w = w;
trainingNeurons{1}.nIDs = 1:k;
trainingNeurons{2}.nIDs = 1:k;
isOnset = rand(1,nstepsSubsong)>.9;
trainingNeurons{1}.tind = find(isOnset);
trainingNeurons{2}.tind = find(isOnset);
trainingNeurons{1}.candLat = 1:trainint;
trainingNeurons{2}.candLat = 1:trainint; %1:trainint;
trainingNeurons{1}.thres = 12;
trainingNeurons{2}.thres = 12;
Input =-HowClamped*ones(k, nstepsSubsong); % clamp training neurons (effectively giving them higher threshold)
Input(trainingNeurons{1}.nIDs,isOnset) = HowOn; % alternating rhythmic activation of training neurons
Input(trainingNeurons{2}.nIDs,isOnset) = HowOn; % alternating rhythmic activation of training neurons
bdyn = double(rand(n,nstepsSubsong)>=(1-pn)); % Random activation
bdyn(1:k,:) = Input;
p.input = bdyn;
[w xdyn] = HVCIter(p);
LatencySubsong{seedi} = findLatency(xdyn, trainingNeurons);
% recovering original params
p.eta = eta;
p.nsteps = nsteps;
% set(gca, 'color', 'none')
trainingNeurons = trainingNeuronsPsyl;
for i = 1:Niter(1)
% Construct input
bdyn = double(rand(n,nsteps)>=(1-pn)); % Random activation
bdyn(1:k,:) = PsylInput;
p.w = w;
p.input = bdyn;
% One 'bout' of learning
[w xdyn] = HVCIter(p);
end
LatencyEarlyPsyl{seedi} = findLatency(xdyn, trainingNeurons);
for i = (Niter(1)+1):Niter(2)
% Construct input
bdyn = double(rand(n,nsteps)>=(1-pn)); % Random activation
bdyn(1:k,:) = PsylInput;
p.w = w;
p.input = bdyn;
% One 'bout' of learning
[w xdyn] = HVCIter(p);
end
p.wmax = wmaxSplit;
p.m = Wmax/p.wmax;
trainingNeurons = trainingNeuronsAlt;
niter = Niter(3);
for i = 1:niter
% Construct input
bdyn = double(rand(n,nsteps)>=(1-pn)); % Random activation
bdyn(1:k,:) = AltInput;
p.w = w;
p.input = bdyn;
p.gamma = gammas(i);
[w xdyn] = HVCIter(p);
end
LatencyLatePsyl{seedi} = findLatency(xdyn, trainingNeurons);
seedi
end
%%
%save('C:\Users\emackev\Documents\MATLAB\code\misc_elm\HVCmodel\sigLatDistOverDev1');
%% compiling and plotting
%load 'C:\Users\emackev\Documents\MATLAB\code\misc_elm\HVCmodel\sigLatDistOverDev1';
Cases = {LatencySubsong LatencyEarlyPsyl LatencyLatePsyl};
figure(2);
Colors = .8*ones(3,3);
FS = 7; % labels
FS_axes = 5; % axis labels
for i = 1:length(Cases)
thisCase = Cases{i};
CompiledLats{i} = [];
for seedi = 1:length(thisCase);
CompiledLats{i} = [CompiledLats{i} thisCase{seedi}{1}.mode(thisCase{seedi}{1}.FireDur)];
end
subplot(3,1,i)
[n,x] = hist(CompiledLats{i}, 1:10);
bar(x*10, n/sum(n), 'facecolor', Colors(i,:))
ylabel('Fraction','fontsize',FS)
xlim([0 110])
ylim([0 1.1*max(n(:))/sum(n)])
box off
set(gca,'color','none','tickdir','out','ticklength',[0.025 0.025])
set(gca,'fontsize',FS_axes)
end
xlabel('Latency (ms)','fontsize',FS)
figw = 3;
figh = 4;
set(gcf, 'color', [1 1 1],'papersize', [figw figh], 'paperposition', [0 0 figw figh])