| What
is a neuron?
A
neuron is a memory cell which can autonomously evaluate the distance
between an incoming pattern and the model stored in its memory. If
this distance falls within its influence field, the neuron claims to
recognize the pattern. Put many neurons in parallel and a pattern can
be recognized by multiple neurons with different levels of confidence.
A key patented feature of the CogniMem neurons is then triggered: they
respond one at a time starting by the most confident and this ordering
is automatic!
A robust non-linear
classifer
- CogniMem implements two re-konw
classifiers:
- K-Nearest Neighbor for best
template matching
- Radial Basis Function to generate a
decision space with cases unknown and uncertain classification
An adaptive model
generator
- CogniMem lets you build a decision
space with no efforts, by simply teaching correct and consistent examples
- A neuron is automatically committed if
the vector to learn and its associated category represent novelty
- All committed neurons with a
contradicting response automatically shrink their influence fields
|
