can nn to detect anomalcy using test data
detect new attack?


there are so many success reports 
database for simulating attack in 
usning nn but 
would they still work if we found other than 
quite suspicious

inteligent behaviour 

foreing language


"Jumping to conclusions is premature," but...

they say "biologically plousible synaptic plasticity, but...




we sometimes meet a description like spiking neurons is more efficient than traditional
but efficient behavior is not neccesarily intelligent




Synaptic strength could not grow indefinitely, but was kept in a range by means of a self-limiting mechanism which depended on synaptic strength.

a synapse could not change sign, which was genetically specified, but only strength.

ach synaptic weight w ij is randomly initialized at the beginning of the individual's life


Plain Hebb rule :can only strengthen the synapse proportionally to the correlated activity
of the pre-and post-synaptic neurons.

Postsynaptic rule :behaves as the plain Hebb rule, but in addition it weakens the synapse
when the postsynaptic node is active but the presynaptic is not.

Presynaptic rule :weakening occurs when the presynaptic unit is active but the postsy-
naptic is not.


Covariance rule :strengthens the synapse whenever the di erence between the activations
of the two neurons is less than half their maximum activity,otherwise the synapse is
weakened.In other words,this rule makes the synapse stronger when the two neurons
have similar activity and makes it weaker otherwise.




reference

S. Bengio, Y. Bengio, J. Cloutier, and J. Gecsei, "On the optimization of a synaptic learning rule," in Preprints Conf. Optimality in Artificial and Biological Neural Networks, Univ. of Texas, Dallas, Feb. 6--8, 1992.

An-experiment-in-genetic-connectionism