-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathindex.html
More file actions
338 lines (304 loc) · 17.8 KB
/
index.html
File metadata and controls
338 lines (304 loc) · 17.8 KB
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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
<html>
<head>
<title>SfN 2014: Cipollini & Cottrell</title>
<link href='styles/libs/jquery-ui.css' rel='stylesheet' type='text/css'>
<link href='styles/stuff.css' rel='stylesheet' type='text/css'>
<link href='styles/poster5.css' rel='stylesheet' type='text/css'>
<script type="text/javascript" src="scripts/libs/jquery-1.11.1.min.js"></script>
<script type="text/javascript" src="scripts/libs/jquery-ui.js"></script>
<script type="text/javascript" src="scripts/poster5.js"></script>
</head>
<body>
<div id="poster">
<!-- title bar -->
<div class="poster-title"
data-logo="images/logo.png"
data-qrcode="images/qrcode.ringo.png"
data-title="A developmental approach to long-distance connectivity and interhemispheric collaboration"
data-authors="Ben Cipollini, Garrison Cottrell"
data-affiliations="UC San Diego, Department of Computer Science"
data-author-affiliations="1,1">
</div>
<div id="section-summary-left" class="section-buttons section-buttons-expanded">
<!-- box 1: the concept -->
<div class="section section-summary"
data-number="1"
data-title="Are large brains lateralized because long delays cause hemispheric independence?"
data-short-blurb="Large brains have longer delays between the hemispheres.
Does this bias circuits to use shorter, intra-hemispheric axons?"
data-blurb="
<div style='width: 50%; float: left'>
Large brains have longer delays between the hemispheres.
Does this bias circuits to use shorter, intra-hemispheric axons?
</div>
<div style='float: right;'>
<img width=50 src='images/figs/brain.jpg' style='float: left; margin-right: 15px;'></img>
<div style='float: left'>
Small brain,<br/>short delay.
<br/>
<br/>
<br/>
Large brain, <br/>long delay.
</div>
<img width=125 src='images/figs/brain.jpg' style='margin-left: 10px;'></img>
</div>">
<div class="content-pane" style="">
<!-- model info -->
<div id="ringo-model" style="float: left">
<img src="images/figs/Ringo_model.png" width=550></img>
<br/><div style="width:100%; font-size: 14px; text-align: center">Adapted from Ringo et. al (1994)</div>
</div>
<div style="margin-left: 25px; float: left; width:500px;">
<p>
<b>Model:</b>
<ul>
<li>Model runs in time
<li>Each connection has a delay
<ul>
<li>Interhemispheric: 1 or 10
<li>All others: 1
</ul>
<li>Model runs for a fixed number of time steps.
<li>Model is trained on a paired associates task: given some input, produce the output.
</ul>
</p>
<p>
<br/>
<b>Training & Analysis:</b>
<ul>
<li>Train network to 100% performance.
<li>Remove interhemispheric connections.
<ul><li>Network with more error uses those connections more!</ul>
<li>Train models with many different total time steps
</ul>
</p>
</div>
</div>
</div>
<!-- box 2: ringo results -->
<div class="section section-summary"
data-number="2"
data-title="No; long delays do not reduce communication."
data-short-blurb="We re-analyzed a rate-coded model (Ringo et al., 1994) and found only a short delay to the onset of communication."
data-blurb="We re-analyzed a well-cited rate-coded model (Ringo et al., 1994)
and found <b>only a short delay to the onset of communication</b>--otherwise no difference.
There are a few potential mechanisms to overcome even this.
<br/><br/>One of these mechanisms to overcome this onset difference is temporal predictability.
">
<div class="content-pane" style="">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 50px;">
Model Results
</div>
<!-- model behavioral result -->
<div style="float: left; width: 520px; height: 420px; background-color: white; ">
<div style="position: relative; padding: 25px; width: 550px; height: 330px; left: -40px;">
<img class="abs" src="images/figs/Ringo-results-1.png" style="position: absolute; top: 0px; left: 0px;"></img>
<img class="abs" id="ringo-results-moving" src="images/figs/Ringo-results-2.png" style="position: absolute; top: 110px; left: 116px;"></img>
</div>
<div style="float: none; text-align:center">
Network run time
</div>
</div>
<div style="float: left; width: 50%; margin-left: 25px;">
<ul>
<li>Network with delay=10 (circles) has little error at early time steps,
then drops off.
<li>Network with delay=1 (triangles) has error early,
then drops off.
</ul>
<p>
Ringo et al. (1994) and others take this as evidence of less
interhemispheric communication, requiring "multiple passes"
to overcome.
</p>
<p>
<b>We know there will be an onset difference of 9 time-steps.
What remains after we account for that?</b>
</p>
<a id="animate-ringo" href="#">shifted</a>
<a id="unanimate-ringo" href="#">original</a>
</div>
<div id="no-diff" style="display: none; width:100%;">
<div style="width: 100%; font-size: 36px; color: red; font-weight: bold; text-align: center; padding-top: 25px">
There is no difference besides the onset delay.
</div>
<div style=" padding: 50px 0px 0px 200px; width:100%;">
Even this onset delay is overstated by the model:
<ul>
<li>There are many long intrahemispheric delays
<li>Difference in delays are extremely short (5-20ms)
<li>Temporal correlations allow hemispheres to predict each others activity.
<li>There is a distribution of delays!
</ul>
</div>
<div style="width: 50%">
<img width=500 src="images/figs/berbel_innocenti_histograms.png" />
<div style="width:100%; font-size: 14px; text-align: center">Axon diameter distribution in the cat (post-natal day 150)<br/>Adapted from Berbel & Innocenti (1988)</div>
</div>
</div>
<!-- model movie for comparison -->
<div id="ringo-compare-movie" style="width: 100%; padding: 350px 0px 270px 0px;">
<div id="ringo-cc-movie" style="width: 50%">
<img src="images/figs/cc/1.png" width=500 style=""></img>
<div style="width:100%; text-align: center">Original network</div>
</div>
<div id="ringo-cc-no-movie" style="width: 45%; margin-left: 25px">
<img src="images/figs/cc-no/1.png" width=500 style=""></img>
<div style="width:100%; text-align: center">Network without interhemispheric connections</div>
</div>
</div>
</div>
</div>
</div>
<div id="main-content" class="main-content main-content-collapsed">
</div>
<div id="section-summary-right" class="section-buttons section-buttons-expanded">
<!-- -->
<div class="section section-summary"
data-number="3"
data-title="<b><i>Unreliable</i></b> delays likely exist during development.</i></b>"
data-blurb="Most axons in early development are thin and unmyelinated fibers.
These have been shown to have unreliable delays (Faisal et al., 2008).">
<div class="content-pane" style="width: 100%; float: left;">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 50px;">
Prediction seems to matter.
</div>
<div style="width: 50%">
<img width=500 src="images/figs/berbel_innocenti_histograms_marked_box.png" />
<div style="width:100%; font-size: 14px; text-align: center">Axon diameter distribution in the cat (post-natal day 150)<br/>Adapted from Berbel & Innocenti (1988)</div>
</div>
<div style="float: left; width: 45%; margin-left: 25px; height:500px;">
At birth, most axons are thin and unmyelinated (bottom plot).
<ul>
<li>Axons with diameter < 0.5um have unreliable conduction delays,
due to a number of noise sources (Faisal et al., 2008).</li>
<li>Longer axons have more variability.</li>
</ul>
Over development, axons become thicker and myelinate.
<ul>
<li>Early in development, longer axons should have more variability.</li>
<li>Over development, variability should decrease and disappear.</li>
</ul>
</div>
<div style="width: 100%; font-size: 36px; font-weight: bold; text-align: center; padding-top:20px;">
What is the effect of length-dependent delay variability<br/>over development?
</div>
</div>
</div>
<!-- -->
<div class="section section-summary"
data-number="4"
data-title="In our model, unreliable delays reduce connectivity."
data-blurb="Using a similar model, we show that changes in delay reliability
can explain changes in local to long distance functional circuits.">
<div class="content-pane" style="position: relative;">
<div>
<div id="model" style="">
<img width=500 src="images/figs/le_model.png"></img>
</div>
<div style="margin-left: 25px; float: left; width:500px;">
<p>
<b>Same as Ringo model, but</b>:
<ul>
<li>"Split" inputs and outputs (for more nuanced testing)
<li>Interhemispheric delay is fixed at 10.
<li>Total number of time steps is fixed at 35.
<li>Noise is added such that interhemispheric connections have 1% noise.
</ul>
</p>
<p>
<b>Run three experiments</b>:
<ul>
<li><a href="#expt1">Expt 1</a>: Show that, over model training ("development") with constant noise, that noise reduces long-distance communication.
<li><a href="#expt2">Expt 2</a>: Show that, over model training ("development") with decreasing noise, that long-distance communication develops.
<li><a href="#expt3">Expt 3</a>: Show that the reduction of communication is more than what is due to the onset delay.
</ul>
</div>
</div>
<a name="expt1">
<div id="expt1" style="width: 100%; padding: 200px 0px 0px 0px;">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 25px;">
Expt 1: Constant Noise
</div>
<img width=800 src="images/figs/expt1_inter_matters.png" style="display: block; margin:auto"></img>
<div style="width: 100%; text-align: center; padding-top: 25px;">
Over learning, the amount of lesion-induced error is less for the noise vs. no-noise
condition, suggesting a bias for local (intrahemispheric) circuits
over long-distance (interhemispheric) circuits.
</div>
</div>
<a name="expt2">
<div style="width: 100%; padding: 200px 0px 0px 0px;">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 25px;">
Expt 2: Tapered Noise (simulating maturation)
</div>
<div style="width: 100%;">
<table style="margin-left: 100px;">
<tr>
<td><img width=200 src="images/figs/expt2_noise_trajectory.png" style="display: block; margin:auto"></img></td>
<td style="padding-left: 50px;"><img width=600 src="images/figs/expt2_developmental_results.png"></img></td>
</tr>
<tr>
<td align=center>Noise reduces over training, roughly simulating axonal maturation.</td>
<td style="padding-left: 50px;"align=center>
<div style="width: 100%; text-align: center; padding-top: 25px; font-size:24px;">
Over learning, the difference in lesion-induced error between noise
and no-noise conditions decreases, showing that an initial bias for
local (intrahemispheric) circuits disappears.
</div>
</td>
</tr>
</table>
</div>
</div>
<a name="expt3">
<div id="expt3" style="padding: 200px 0px 200px 0px;">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 25px;">
Expt 3: Analyzing communication
</div>
<div style="width:100%; display: block; padding: 0px 150px 0px 150px; box-sizing: border-box;">
<img width=500 src="images/figs/inter_vs_intra_patterns.png" style="display: block; margin:auto"></img>
<ul>
<li><b>intra</b>hemispheric patterns: a hemisphere's output can be made from its own input.
<li><b>inter</b>hemispheric patterns: a hemisphere's output can be made only after knowing both hemisphere's inputs.
</ul>
</div>
<div style="width:100%; padding-top: 25px;">
<div style="width: 100%; text-align: center; font-size: 36px; font-weight: bold; padding-bottom: 25px;">
Results
</div>
<img width=500 src="images/figs/expt3_all_results.png" style="display: block; margin:auto"></img>
<div style="width: 100%; text-align: center; padding: 25px 150px 0px 150px; box-sizing: border-box">
There is an effect of delay onset for all conditions and patterns.
In the noise condition, the inter- patterns show an additional effect
of noise.
</div>
</div>
</div>
</div>
</div>
</div>
<!-- footer -->
<div class="poster-footer">
<div style="width:50%; height: 100%; float: left;">
<div style="float: left; margin-right: 25px; height: 100%; left: -50px">References:</div>
<div style="font-size: 18px">
<li>Berbel, & Innocenti. (1988). The devel. of the corpus callosum in cats. J of comparative neurol., 276(1).</li>
<li>Faisal et al. (2008). Noise in the nervous system. Nature Rev. Neuro., 9(4).</li>
<li>Lewis & Elman. (2008). Growth-related neural reorg. & the autism phenotype. Devel. Science, 11(1).</li>
<li>Ringo et al. (1994). Time is of the essence. Cereb. Cortex, 4(4).</li>
</div>
</div>
<div style="width:50%; float: left;">
<div>
<div style="float: left; position: relative; left: -10px; height: 100%;">Conclusions:</div>
<div style="font-size: 24px;">
<li>Delay magnitude has little effect on rate-coded networks; <i>unreliable</i> delays matter.</li>
<li>In development, axons are thin & unmyelinated, and likely have unreliable delays.</li>
<li>Our model suggests an early bias for using shorter axons with more reliable delays.</li>
</div>
</div>
</div>
</div>
</body>
</html>