Artificial Neural Network Models of Real Neural Computation

Artificial Neural Network Models of Real Neural Computation

Minus times Minus equals Plus:The reason for this we need not discuss.

W.H. Auden

f ( net ) =

11+e−net

: 'logistic'

w1

w2

w0

wn

.

.

.

1.0

p1

pn

p2 a net =w0 + p1w1 + p2w2 +K + pnwn

a = f (net)

pi

pi

pi

Inputs Outputs

a1

a2

a3

Hidden Units

Training set:

p1 → a*1

M

pi → a* i

M

Error = a

j* i −aj( )

j∑

2

Minimize error

f net( ) =net : 'linear'

Retinal Receptive Field of spatially tuned neuron in area 7a

Experiment

Model

Experiment Model Hidden Units

ExternalInputs

outputs

ExternalInput

Weights

RecurrentWeights

1.0 p1 p2

Non-Zero Weight

<-<-<- Training data

Performance