6/7/18
1
DFT Summer School:Lecture 3John P. SpencerProfessorSchool of PsychologyUniversity of East AngliaNorwich, UK
How neural is DFT?ªDoes DFT simply create an analogy to neural systems?ªOr is DFT ‘neural’ in a deeper sense?
DFT is deeply neural…ªWe’ll show this using an analysis method called the
distribution of population activation (DPA)ªDPA creates a direct mapping from electrophysiological
recording to the neural population dynamics captured in DF models
ªBut more than that, we’ll see direct neurophysiological evidence for the types of neural interactions instantiated in DF models à neural evidence for DFT
6/7/18
2
Receptive fields of single neuronsSingle cells are ‘tuned’ to metric dimensionsªThey fire maximally to certain
stimuliªBut tuning curves are generally
quite broadªAs a consequence, it’s hard to
glean much from the firing rates of individual cells
ªBy contrast, response across multiple cells can be quite informative
Spi
ke R
ate
Feature DimensionValue 3Value 2
A B CNeurons
Value 1
(a)
Value 1 Value 3
Spi
ke R
ate
(c)
(b) Neuron A Neuron C
Spi
ke R
ate
1 3
Neuron B
2 1 32 1 32
Value 2
A CBA CB A CB
Does the whole population really matter?Yes!ªDeactivate A and visually
evoke an eye movement to A à still get an eye movement to A
ªDeactivate A and visually evoke an eye movement to B à get an eye movement biased toward D, even though B is still engaged (A has an influence, even though at the periphery)
Amplitude
+40°
+20°
0°
-20°
-40°2°5° 10° 20° 30° 50°
A
B
C
B
A
C
Amplitude
+40°
+20°
0°
-20°
-40°2°5° 10° 20° 30° 50°
A
C
BB
Result (A)
C
Amplitude
+40°
+20°
0°
-20°
-40°2°5° 10° 20° 30° 50°
A
C
BB
ResultD
D
(a)
(b)
(c)
A
Angula
rD
irec
tion
Angula
rD
irec
tion
Angula
rD
irec
tion
6/7/18
3
What about non-topographic areas?ªMotor cortex is not topographic…but it is functionally
topographic
Population vectors
0°
30°
60°90°
120°
150°
180°
210°
240°270°
300°
330°
0°
30°
60°90°
120°
150°
180°
210°
240°270°
300°
330°0° 90° 180° 270° 360°
Spik
e R
ate
Movement Direction
(a) (b) (c)
6/7/18
4
Population vector predicts movement direction prior to actual movement
DPA generalizes this approachªCells contribute their entire tuning curves, not just their
maximal direction
acti
vat
ion
feature space
6/7/18
5
Evidence for surround inhibition
From excitation close to inhibition far
6/7/18
6
Neural repulsion effect!
DPA in motor cortex
a
b
c
4 →
23
5 6
Start trial 500 ms PS
1
RS MVT
RTPP
1000 ms
← →← →
6/7/18
7
complete information(a)
PS500
RS1500
56
12
340
2
4
6
8
10
time (ms)targets
popula
tion a
ctiv
atio
n
two-target information(b)
PS500
RS1500
56
12
340
2
4
6
8
10
time (ms)targets
popula
tion a
ctiv
atio
n
three-target information(c)
PS500
RS1500
56
12
340
2
4
6
8
10
time (ms)targets
popula
tion a
ctiv
atio
n
12
34
56
6 PS
RS
acti
vat
ion
12
34
56
6 PS
RS
acti
vat
ion
timetargets
timetargets
complete information (DNF model)(d)
two-target information (DNF model)(e)
DFT is not a neural analogyªEvidence suggests that the brain actually work this wayªNeural population dynamics captured by DFT are
observable in cortex