We need to have a context that indicates when ramps should pay attention to the interval that's occurring.
First Trial
Light ON:
A occurs...
All ramps associated with A fire.
If light OFF before another event happens, no ramps are updated
If B occurs before light OFF, a coin is flipped for each ramp, chosen ones are updated to time A->B
Next light ON
A occurs,
Light stays ON but B hasn't happened
For the ramps that hit threshold when B should've happened, flip coin again and update selected ramps to ETR until another event happens or the light turns off (they are late for the next trial, probably)
We'd like to see if 1) there's some convergence by randomly updating these ramps in the context of learning 2) how the model performs by letting the ramps decay when they don't see the terminating event
We need to have a context that indicates when ramps should pay attention to the interval that's occurring.
First Trial Light ON: A occurs... All ramps associated with A fire. If light OFF before another event happens, no ramps are updated If B occurs before light OFF, a coin is flipped for each ramp, chosen ones are updated to time A->B
Next light ON A occurs, Light stays ON but B hasn't happened For the ramps that hit threshold when B should've happened, flip coin again and update selected ramps to ETR until another event happens or the light turns off (they are late for the next trial, probably)
We'd like to see if 1) there's some convergence by randomly updating these ramps in the context of learning 2) how the model performs by letting the ramps decay when they don't see the terminating event