Transcript Based on Predictive Hebbian Learning

Model of Bee Foraging
Based on Predictive
Hebbian Learning
Iain Black
Simply put, learning in our brains takes
place by changing the strength of
connections between individual neurons.
How the connection strengths are set
depends on the learning rules. Artificial
neural networks work in this way too, and
many different learning rules have been
developed for learning different tasks. The
aim of this project is to investigate the
operation of a learning rule that is
biologically plausible - predictive Hebbian
learning.
The idea is to implement a simple neural
network that models the behaviour of a bee
foraging for nectar in a field of flowers,
based on the work of Montague et al.
(Nature 377:725-728, 1995). This simulator
will allow an investigation of how the bee's
behaviour is influenced by the learning
parameters of the model.
The bee will fly around a field of blue and
yellow flowers and hope to get a reward
(quantity of nectar) for each flower it lands
on. However there is a catch!
The bee will be rewarded with 2uL nectar
for yellow flowers and 6uL for blue
flowers, however, there is only a 1/3 chance
the blue flowers will have this reward.
With the rewards the same overall, what
will the bee choose?
Sensory
Input
Nectar
START
S
Y
N
RANDOMLY POSITION
BEE AT TOP OF ARENA
WITH RANDOM
INITIAL HEADING
B
Action
R
WY
r(t)
WN
P
WB
δ(t)
The above diagram shows the structure of the artificial
neuron that is used to model the bee’s decision
making. The bee will receive sensory inputs of yellow,
neutral (out of field) and blue which represents the
bee’s vision. Depending on the bee’s previous
experience (i.e., constant reward failure from blue
flowers), it will make a decision to continue on it’s
path or re-orientate to try and find a more favourable
part of the field.
NO
HAS BEE
LANDED?
YES
REPEAT
RECEIVE
OUTPUT OF P
NO
REORIENTATION
CONTINUE
ON PATH BY
1 STEP
REORIENTATON
MAKE RANDOM
CHANGE IN
HEADING AND
TAKE A STEP
GET (OR NOT)
NECTAR FROM
FLOWER
UPDATE
SENSORY
WEIGHTS
RESTART
The flowchart shows the basic decisions the bee must make
during it’s foraging. This sequence is repeated for a desired
number of times, a pattern should emerge.
The bee will move around the field attempting to
forage nectar in an efficient manner. The slider
represents the percentage of blue/yellow the bee is
viewing at the current time. The bee should tend to
favour one colour over the other. Different learning
rules should provide different results.