What is thinking? The dynamics of mental exploration

1
What is thinking?The dynamics of mental
exploration
2

• Thinking is a process by which a
  computational system can generate an effective
  action in a novel situation, based on exploring
  the possibilities (often combinatorial) implicit
  in previously acquired knowledge.

3
Sound waves in a simple gas molecular viewpoint
molecules j of mass m, molecular forces m
dvj/dt Fj (Newtons laws for 1028
molecules) Fj - Sk grad V(rj rk) no sound
waves without molecular interactions Navier-Stok
es viewpoint pressure fluctuations in a fluid
mean density r compressibility k k-1 (?2/?x2
?2/?y2 ?2/?z2) p r ?2p/?t2 The existence of
molecules and their microscopic interactions
has disappeared
4
(No Transcript)
5
Drift and diffusion model of decision-making Slid
ing motion along a coordinate Drift caused by
available evidence (slope) Diffusion (random walk
addition) caused by noise Decision made when
green dot reaches an end Reaching lower end
means decision correct Reaching upper end means
erroneous decision
6
(No Transcript)
7
(No Transcript)
8
Mental exploration A
protracted evolution of the pattern of neural
activity, while an animal is not yet taking
actions and while sensory input may be constant
(or irrelevant), followed by an apt behavioral
action that directly relates to the activity
states during the exploration.
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
Where action potentials spikes are generated
in a typical rat hippocampal place cell during
exploration of a familiar space
Box environment with visual clues on walls
13
from Wills, Lever, Cacucci, Burgess and OKeefe
(2005)Science 308, 873
14
(No Transcript)
15
(No Transcript)
16
Mapping brain cell activity to a useful spatial
display
cell location in brain
spatial display for a rectangular
environment arrowheads at position of maximal
activity of corresponding neuron
17
(No Transcript)
18
(No Transcript)
19
near water neuron (connections learned)
Spatial representation of activity
pattern (reordering of pattern in
hippocampus) Strongly active neurons in
hippocampus when animal is at location w
o
20
from Wills, Lever, Cacucci, Burgess and OKeefe
(2005)Science 308, 873
21
Animal in R environment Animal in H
environment
Rectangle H environment
environment display display
22
(No Transcript)
23
Mental exploration While stationary, mentally
explore extensively to search for water in
present environment make a stable activity clump
in any particular environment make the activity
clump explore that environment If water is found,
find a (the?) mental pathway between present
physical location and location of water Remember
the pathway so that it can be mentally
repeated Use mental recapitulation of motion to
guide corresponding physical motion along the
mental path
24
(No Transcript)
25
Exploration in one
dimension Each neuron has a location of maximum
activity Arrange neurons in this natural order
for display purposes Connect each neuron k to M
others (denoted by j) that are most strongly
active when k is active. This defines the
connection matrix Tkj neurons that fire
together wire together
c
Tkj
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
Firing rate adapation
input
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
Mental exploration While stationary, mentally
explore extensively to search for water in
present environment If water is found, find a
(the?) mental pathway between present physical
location and location of water Remember the
pathway so that it can be mentally repeated Use
mental recapitulation of motion to guide
corresponding physical motion along the mental
path
38
Mentally learning a physical trajectorymove
along a pathsensory input dominates place
cells Skj (activity of neuron k ) (activity
of neuron j) accumulate Skj throughout this
motionFor all synapses that are non-zero (i.e.
Tkj 1)if Skj gt threshold valueincrease Tkj
by 50Strengthens synapses that would be
useful along the path
39
(No Transcript)
40
MLS
41
Motor controllerintegrate-and-fire neuron
slow excitatory pathway
each output spike approximately
reverses direction of motion with random spread ?
60o
input
balanced fast inhibition
long-lasting self-inhibition Ca-dependent Inhibi
tory currents
42
No control signal
control input from olfactory
cells Gaussian spatial odor profile
43
MLS
Area E activity is a moving bump representing
intended action Area A has two bumps of
input Sensory input represents where the animal
is Input from area E reflects the intended
position When these coincide, area A has maximal
activity When well separated, area a has little
activity Motor system will move animal to
intended location
44
(No Transcript)
45
recapitulate mental success to guide physical
motion
mentally explore for w
46
Activity-position movie 10 frames/sec Red
instantaneous location of animal Black
points . Center of receptive field of
strongly active place cells 100 randomly chosen
place cells in the interior of a T environment
47
Explanation of movie The
mouse goes one branch to another, not directly
repeating. (perhaps a learned behavior) Hippocamp
al place cells have indirect inputs from the
sensoryvestibular system. They have a selective
filter on this input resulting in spatial
receptive fields. When the animal is at X, the
cells with place field centers near X are
strongly active, driven by sensory input that
characterizes being at X Being at X causes
corresponding neural activity WRONG There is
NO sensory input to these place cells (in E
ca3?) during the movie. A moving cluster of place
cells with is active through mutual feedback.
When this cluster is at X, the animal wants to
be at X (i.e., this activity causes the motor
system to move the animal to X). Intrinsic
neurodynamics makes the active cluster moves in
mental space, causing the animal to move
correspondingly in real space. CAUSALITY IS
REVERSED
48
(No Transcript)