Vidna kognicija II

1
Vidna kognicija II

• Danko Nikolic

2
New lectureSmall delays in synchronization
3
(No Transcript)
4
Kuffler (1953) Discharge patterns and functional
organisation of vertebrate retina, Journal of
Neurophysiology
Increase in stimulus intensity
Stimulus onset
5
A simulation test
Van Rulen and Thorpe (2001) Rate Coding Versus
Temporal Order CodingWhat the Retinal Ganglion
Cells Tell the Visual Cortex, Neural Computation,
13
6
Somato-sensory information
Johansson and Birznieks (2004) First spikes in
ensembles of human tactile afferents code complex
spatial fingertip events. Nature Neuroscience.
7
Hippocampus
O'Keefe J., Recce M.L. (1993). Phase relationship
between hippocampal place units and the EEG theta
rhythm. Hippocampus 3 317 330.
Harris K.D., Henze D.A., Hirase H., Leinekugel
X., Dragoi G., Czurko A. Buzsaki G. (2002). Spike
train dynamics predicts theta-related phase
precession in hippocampal pyramidal cells. Nature
13417(6890) 738-41.
8
Measuring small delays

• Fitting a function and taking its maximum value
  for the estimate.

Cosine fit
9
Phase offsets can be measured with
sub-millisecond precision

• Schneider and Nikolic, Journal of Neuroscience
  Methods (2006).

10
(No Transcript)
11
Large networks
12
Relative firing time ms
13
Extraction of the firing sequence
Nikolic, Journal of Comp. Neuroscience (2007)
Schneider, Havenith and Nikolic, Neural
Computation (2006)
Schneider and Nikolic, Journal of Neuroscience
Methods (2006)
14
Non-parametric detection of temporal order
Nikolic, Journal of Comp. Neuroscience (in press).
15
Example Stimulus dependence
1)
2)
16
(No Transcript)
17
(No Transcript)
18
Firing sequences change dynamically
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
Havenith et al., (in preparation)
23
Conclusion Firing sequences

• Short time delays can serves as a code for
  carrying stimulus-related information that is as
  reliable as is the neuronal firing rate.
• Stronger synchronization increases the
  reliability of the code.

24
Binding problem
25
Perceptual integration and organization
26
(No Transcript)
27
(No Transcript)
28
Hierarchical coding by extraction of feature
combinations
Grandmother cell
29
Problems combinatorial explosion and novel
combinations
Combinatorial explosion There are many different
grandmothers and each can be seen from many
different perspectives.
Novel combinations Some grandmothers are seen
for the first time. There is no chance to learn
all the possible combination of features that
make a grandmother.
30
Perceptual organization through synchronization
of action potentials
31
Split bar experiment
(Gray et. al, 1989)
32
Synchrony at different scales
33
Conflicting bar experiment
Engel, A.K., Koenig, P. Singer, W. (1991)
Kreiter, A.K. Singer, W. (1996).
34
Mechanisms Tangential connections
35
An important role of attention
Higher brain areas and awake states

• In early visual areas- In higher visual areas
  V4, MT.

Infero-temporal cortex and recognition of faces
Mechanisms of synchronization
Mechanisms of detection
36
Attention and rates in V4

• Modulation of rate firing rate responses in V4
• Moran Desimone, 1985

37
Attention and synchrony in V4

• - Fries et al., 2001
• Investigated strength of synchrony
• Spike-triggered averages

Delay period
Stimulus period
38
Attention in early visual areas

• Roelfsema et al., 2004
• Non-modulation of synchrony in V1

39
Infero-temporal cortex

• Hirabayashi Miyashita
• Perception of faces

Face
Non-face

• Synchronization is stronger when faces are
  perceived.

40
Mechanisms

• For large part unknown
• To a high degree theoretical answers
• Models, simulations
• Three types of mechanism are considered
• bottom-up
• lateral interactions
• top-down

41
Bottom-up
Common input
Input is not shared
42
Lateral interactions
Tangential connections
43
Top-down
Lower visual area
Higher visual area
Feedback connections
44
(No Transcript)
45
The brain as a liquid-state machine
46
(No Transcript)
47
primary visual cortex

• Parallel recordings by multiple Michigan probes.
• Up to 48 channels.
• Cat visual cortex, area 17
• Anesthesia

48
(No Transcript)
49
main result
50
persistence of information
51
temporal superposition
XOR
52
code invariance
53
correlations matter
54
precise spike timing
55
a subject to noise
56
conclusions

• - memory information about previously shown
  images is available for a prolonged period of
  time.
• - superposition information about previously and
  currently shown stimuli is available
  simultaneously.
• - non-linearity information about non-linear
  transformations of input properties can be
  extracted by linear classifiers.
• rates and timing information is coded partially
  in neuronal firing rates and partially in the
  precise timing of neuronal spiking activity.
• 2nd order correlation the advantage of using
  additional non-linear classification methods was
  limited to the use of pair-wise correlations
  between neurons.