TRANSFORMATION

1
TRANSFORMATION
Biomedical engineering Group School of Electrical
Engineering Sharif University of Technology

• Feed-forward NONLINEAR TRANSFORMATIONS

2
Linear Transformation Extentions
3
Transfer Function
4
NONLINEAR TRANSFORMATIONS

• The impressively complex behavior exhibited by
  neural systems is unlikely to be fully explained
  by linear transformations alone.
• Nonlinear operations are central to neural
  transformations
• Nonlinearities vastly increase computational
  power and storage capacity

5
Multiplying two variables
6
(No Transcript)
7
(No Transcript)
8
First Set Coefficients
9
2nd Set Coefficients
10
Simulation Results
200 LIF neurons in each of x,y,z
400 in the middle layer
The network is driven by ramp input to x and y
11
NEGATIVE WEIGHTS AND NEURAL INHIBITION

• The approach does not distinguish between
  positive and negative elements of the connection
  weight matrix
• Significant differences between neural excitation
  and neural inhibition
• Subclass of neurons using GABA to inhibit the
  firing of postsynaptic neurons
• Projections to these interneurons are only
  excitatory
• not a mixture of positive and negative connection
  weights

12
Implementation
13
Simulation for yx
14
AN EXAMPLE THE VESTIBULAR SYSTEM
otolith signal
gravitational
inertial
B
A
15
Otolith Signal
16
modularization of a vestibular circuit
biased to the left
3 sub-populations
17
(No Transcript)
18
(No Transcript)