Cue → Competition → Spark → Recall • Nakano-style Associatron (associative memory dynamics)

Cue
N = 64×64 = 4096
Episode Age (Write Target)
Overwrite Mode (Strong Write) P=0
Flow:
1) Choose target → 2) Draw cue → 3) Store trace (repeat to strengthen)
Clear cue → Recall runs across rooms (write target is for learning, not recall)
Cue Channel: Smell
Strength (cue weight)
0.75 gate 0.0% / mem 0.0%
Door threshold θ
0.28
Hesitation Δ
0.10
Display β
12 Top: -
Hallway (room opening likelihood)
Draw Cue
Draw a partial cue. Add noise to test basin robustness.
Camera Exposure 1.00 OFF finger → cue line → stop → auto recall
- Learning stores traces (reinforce similar, forget over time).
- Recall is not lookup: a cue activates candidates, they compete, and one basin wins.
- Spark reveals overlap inside the winning basin: shared structure becomes visible.
State
Episode traces (stored memories)
Door is still closed.
Tatami (8×8): place lights first
- When the door opens, “place” comes first.
- If the cue is vague, the light spreads.
Spark: Shared Features Ignite
- Winner-room memories are stacked, weighted by cue similarity.
- Hotter means “more shared here.”
Recalled Pattern (64×64)
- Recall runs using the final winner memory (across rooms).
- Low-rank mixing (fast).