Neuromorphic Smart Controller for Seismically Excited Structures - Design Optimization of Active and Passive Structural Control Systems - page 221

A G SI
i

= ⋅ (1)

where G OFC i are the nodal gains for the or-
bitofrontal cortex nodes. The gain is updated
according to the learning rule

A i

i

where the subscript i represents the i th sens-
ing stream and G A i are the learning gains for
the amygdala nodes. The sensory input is
the motivation that makes the control system
operates. The gain is updated according to
the learning rule

( ) ,

∆ G = ⋅ b SI MO ES ⋅ − (4)

OFC i

i

where b is the learning rate for the orbito-
frontal cortex nodes, ES is the emotional
signal. The ES is to correlate the sensory
input with the control output. In other
words, the emotional signal is generated
considering proper linkage between sensory
input and control action. These concepts are
implemented by Equations (2) and (4). The
overall model output, MO is determined by
the following equation

∑

∆ G = ⋅ a SI ⋅ max 0 ES − A (2)

A i i

i

i

where a is a learning rate of the amygdala
node selectable between 0 (no learning) and
1 (instant adaptation) and ES is the emotional
signal. Note, ES (also called the reward
signal) is an internally generated signal un-
like the sensory input which is an external
signal that is measured by sensors. Indeed,
the main learning process in the system hap-
pens in the adaptive gains of the amygdala
and orbitofrontal cortex that is explained in
following section.

∑ ∑ MO = A OFC . (5)

i i

i i

From the above model output equation, it is
realized that the overall model output is the subtrac-
tion of excitatory signals from the amygdala and
the inhibitory signals from orbitofrontal cortex.
Note, the learning trend in the amygdala is never
unlearn a connection by taking maximum value
between 0 and the difference between ES and
amygdala gains; once learned, it is permanent,
giving the system the ability to retain emotional
connections for as long as necessary. However,
the orbitofrontal cortex can both learn and unlearn
(forget). Basically, this is inspired by biology:
based on both good and bad experiences, the
amygdala constantly learns the associations be-
tween the sensory input signals and the reward
signal and tends to behave based on the learned
associations. On the other hand, the orbitofrontal
cortex inhibitory signals act to prevent any inap-
propriate actions to be issued by the amygdala.

• Orbitofrontal Cortex (OFC): Tracks the

mismatch between the system's predictions
and the actual received reinforcement, and
learns to inhibit the system output in pro-
portion to the mismatch. The OFC reduces
the strength of amygdala which is no lon-
ger appropriate as the goal or the context
has been changed. The OFC regulates the
mapping of the stimuli to the emotional
reaction occurring through the amygdala.
For any sensory input signal SI i , the output
signal of a corresponding orbitofrontal cor-
tex node OFC i is

OFC = G ⋅ , SI (3)

i OFC i

i