Transcript No Slide Title

optic nerve
Striate
Cortex
(V1)
1 deg
Hubel & Wiesel
Striate
Cortex
(V1)
1 deg
Hubel & Wiesel
Spikes from an LGN Neuron: 62 Repeats of each stimulus
S1
trial #
1
....
Firing Rate (Hz)
62
time
Butts et al. 2010
S2
S3
80
60
response
(spikes/s)
40
20
0
-25º
Sclar & Freeman 1982
0º
orientation
+25º
Odorant Receptors
amines
lactones
acids
sulfur
terpenes
aldehydes
ketones
aromatics
alcohols
esters
Hallem & Carlson 2006
IT face cell
Striate
Cortex
(V1)
1 deg
Tsao et al. 2006
x xxx
LGN
x xx x
x xxx
+
+
X = excitation
= inhibition
Striate Cortex
Hubel & Wiesel 1962
+
80
80% contrast
40% contrast
response
(spikes/s)
R1
0
-25º
Sclar & Freeman 1982
0º
orientation
+25º
1.0
attend in
attend out
V4
response
0.5
0.0
-90º
-60º
-30º
0º
orientation
McAdams & Maunsell 1999
30º
60º
90º
adapting
direction
100
spikes/s
50
0
-180
Kohn & Movshon 2004
0
180
direction of motion
waterfall illusion
Early: 65 to 85 ms
Late: >150 ms
(2 or 3 spikes)
140 spikes/s
 = 45°
 = 90°
 = 135°
Pack & Born 2001
140 spikes/s
Lorençeau et al. 1993
trial #
Spikes from an MT Neuron: Identical Stimulus, 210 Repeats
sp/sec
Shadlen & Newsome 1994
time (ms)
Outline: neural coding lecture, pt 2
Population coding: a case study
Problems in understanding decoding
A cheat sheet for your homework assignment
Population coding: a case study
the cricket wind direction sensing system (first-order neurons)
Bacon & Murphey J. Physiol. 1984 352:601-623
Population coding: a case study
the cricket wind direction sensing system (second-order neurons)
First-order neuron projections to the terminal ganglion are organized according to preferred wind direction.
There are four second-order neurons, and their dendrites are organized along the same divisions.
see http://www.biol.sc.edu/~vogt/courses/neuro/neurolabs.html
Bacon & Murphey J. Physiol. 1984 352:601-623
Population coding: a case study
cell 1
cell 2
cell 3
cell 4
v
r / rmax
wind direction (degrees)
P. Dayan & L.F. Abbott Theoretical Neuroscience MIT Press
Population coding: a case study
v
P. Dayan & L.F. Abbott Theoretical Neuroscience MIT Press
Outline: neural coding lecture, pt 2
Population coding: a case study
Problems in understanding decoding
A cheat sheet for your homework assignment
Problems in understanding decoding
Which spike trains are being decoded to produce a percept?
Some criteria:
Stimuli that produce different percepts should produce discernable
changes in the spiking of the candidate neurons.
Differences in the spiking of candidate neurons should be
sufficiently reliable to account for the acuity of the percept.
Noise in the activity of the candidate neurons should predict noise
in the percept.
Artificially stimulating the candidate neurons should affect the
percept.
Silencing or removing the candidate neurons should affect the
percept.
adapted from Parker & Newsome, Annu. Rev. Neurosci. 1998. 21:227–77.
Problems in understanding decoding
Is information encoded in spike timing or spike rate?
In principle, either spike
timing or spike rate can
carry information about a
stimulus.
adapted from Gollisch & Meister Science 2008 319:1108-11
Problems in understanding decoding
How much of a spike train should we consider?
Behavioral performance can help
tell us what portion of a spike train
we should consider.
Cury & Uchida Neuron 2010 68:570-585
Problems in understanding decoding
Is the optimal decoding algorithm always used by the organism?
rapidly adapting
slowly adapting
psychophysical
rapidly adapting
type 2
rapidly adapting
type 1
The “lower envelope model”: Sensory thresholds are specified by
the neuron that has the lowest threshold for stimulus in question.
Johansson & Vallbo, J. Physiol. 1979 297:405-422
Problems in understanding decoding
Is the optimal decoding algorithm always used by the organism?
rapidly adapting
slowly adapting
… but single neurons can exhibit better acuity than the organism as a whole!
Johansson & Vallbo, J. Physiol. 1979 297:405-422
Problems in understanding decoding
Does each neuron provide independent information to the decoder?
The “pooling model”:
Sensory thresholds can be improved by pooling independent information from many neurons.
Problems in understanding decoding
Does each neuron provide independent information to the decoder?
Problems in understanding decoding
Does each neuron provide independent information to the decoder?
There is lots of evidence that activity in nearby neurons is often not independent.
Outline: neural coding lecture, pt 2
Population coding: a case study
Problems in understanding decoding
A cheat sheet for your homework assignment
Principal component analysis: a method for reducing the dimensionality of a data set by
defining a reduced set of axes which account for much of the variance in the data.
principal component 1
accounts for a large
part of the variance
(“body size”)
principal component 2
accounts for a smaller
part of the variance
Linear discriminant analysis: a method for classifying samples within a data set based
on drawing a linear boundary (a line or plane) which best separates different categories of
samples.
discriminant