Neural Activity Decay
- Biological neurons retain activity if they do not fire
- Some of the activity leaks away but can contribute to firing
in future cycles
- The models close to Hopfield have decay turned off
- If we turn decay on (so that half of sub-firing activity is retained
after each step),
- The system functions largely the same as the earlier system
- In CANT the decay parameter divides the activity divisor, so
2 halves it
- If the number is dropped to 1.1 the system still functions largely the
same
- This is because any neurons that receive activity are also heavily
supressed
- Consequently no activity persists, and decay is largely irrelevant
in this net
- Why is decay relevant to the brain?
- Perhaps it is just a biological phenomena that we are stuck with, a
evolutionary mistake like the appendix.
- However, decay may be useful to deal with supressing noisy input
- Decay is also probably useful in allowing evidence to accumulate
and thus to encourage proto-CA formation.
