Long Term Stable States
- Usually, semantic memories remain quite stable.
- My concept of a horse hasn't changed much in the last
several years.
- The process of forming a CA should lead to a mostly stable set
of neurons in the CA.
- The process looks like this.
- Initially weights have a base level and there is no CA. (Of course
neural genesis in the neonate is much more complex than this.)
- An instance is presented causing neurons to fire.
- This raises the strength of synapses between neurons in the eventual
CA.
- This is repeated until the strength gets to a point where
reverberation is enabled.
- Once reverberation starts there is a lot of activation within
a CA, and this leads to a corresponding growth of inter-CA
synaptic strength.
- However, this strength stabilises because the neurons are
not always on when a CA is active (due to neural fatigue).
- Problem 1: When do you forget a CA?
- If a CA is not ignited for a long time, synaptic strengths may be
slightly reduced, but reverberation in sub-CAs should keep the
weights high.
- So, you do not forget semantic memories.
- (In other CAs, memories may be erased because there isn't reverberation
in sub-CAs, or there is a lot of use of the neurons without igniting
the CA).
